WO1995031393A1

WO1995031393A1 - Elevator group control system

Info

Publication number: WO1995031393A1
Application number: PCT/JP1994/000795
Authority: WO
Inventors: Shintaro Tsuji
Original assignee: Mitsubishi Denki Kabushiki Kaisha
Priority date: 1994-05-17
Filing date: 1994-05-17
Publication date: 1995-11-23
Also published as: KR960704792A; EP0709332A1; DE69426420D1; EP0709332B1; DE69426420T2; KR0178322B1; JP3215426B2; EP0709332A4

Abstract

A system for group controlling a plurality of elevators in accordance with group control algorithm including a plurality of parameters. This system includes a means for searching an optimum set. An optimum set means an optimum combination among combinations of parameters to be given to the group control algorithm. A new set is generated by crossing-over or mutation. Superior sets are accumulated in a storage by utilizing the additional registration of excellent sets in a storage and the deletion poor sets therefrom. An optimum set is selected from these superior sets. Therefore, the searching of an optimum set can be done efficiently.

Description

Akitoda Title of Invention

Elebe overnight group management system

Technical field

The present invention relates to an elevator group management system, and more particularly, to an apparatus for efficiently searching for an optimal combination of control parameter overnight values. The background technology

The elevator group management system is a system for operating multiple elevators efficiently according to various traffic conditions in the building. The group management device in this system performs operation control such as elevator assignment according to the group management algorithm. Here, the group management algorithm executes and controls all functions and operations related to elevator operation, including allocation control of elevators.

The group management algorithm includes various control parameters, and in order to achieve efficient operation, appropriate parameters are set according to various traffic conditions in the building. You need to substitute a number. When a new hall call is registered in the hall call (call from Yerebe Ichiyu Hall) allocation control, which is one of the basic functions of group management, if a new hall call is registered, the new hall call and the registered hall are already registered. In order to evaluate the service status of the call, such as waiting time and forecast, the evaluation value Em is obtained for each elevator (car) according to the allocation evaluation function described below. Then, the elevator with the smallest evaluation value Em is assigned and selected as an overnight elevator. Then, light the hall lanterns, etc. provided at the landing and wait before the assigned car arrives. The car is displayed as a guide to the customer (this is called a forecast). As a function for obtaining the above-mentioned assignment evaluation value Em, for example, there is the following equation [1]. Here, i is the hall call number, and m is the elevator エレ.

Em = ∑ (W (i) ² + CaxM (i) + CbxY (i)} + Pm-Bm… [1]

However, Em _: Assignment evaluation value when a new hall call is assigned to elevator m W (i): Estimated waiting time for hall call i when a new hall call is assigned to elevator m

M (i): Probability of fullness of hall call i when a new hall call is assigned to elevator m (0≤M (i) ≤ l)

Y (i): Probability of forecast call deviating from hall call i when a new hall call is assigned to elevator m (0≤Y (i) ≤ l, where forecast deviates is a phenomenon in which elevators other than the overnight forecast first arrive first. Means)

Pm: Penalty for assigning a new hall call to the elevator m m Bm: Bonus for assigning a new hall call to the elevator m Ca: Packing evaluation coefficient

Cb: Out-of-forecast evaluation coefficient Here, the packed evaluation coefficient Ca is a weighting coefficient of the packed evaluation value M (i) with respect to the waiting time evaluation value W (i) ² . It is possible to operate with emphasis on passing.

The forecast deviation evaluation coefficient Cb is a weighting coefficient of the forecast deviation evaluation value Y (i) with respect to the waiting time evaluation value W (i) ² . Also, allocation can be performed with emphasis on prevention of deviation from forecast. Examples of the priority assignment function using the penalty Pm in the above equation [1] include the following [2] boarding time priority assignment function and [3] power saving priority assignment function as described below.

[2] The ride time priority assignment function is a function that makes it difficult to assign a call from a landing on the way to an elevator that has many car calls. For example, a value calculated by (riding time priority Pa) X (call number Nm) is set in the penalty Pm.

[3] The power saving priority assignment function is a function that makes it difficult to assign a new call to a suspended elevator. For example, the value indicated by the power saving priority Pb is set as the penalty Pm for the idle elevator, and 0 is set for the other elevators.

In addition, the priority assignment function using the bonus Bm in the above equation [1] includes, for example, [4] a proximity elevator priority assignment function, [5] a light load elevator priority assignment function, and [6] a specific elevator priority assignment function. There is.

[4] The proximity elevator priority assignment function facilitates assignment of an elevator (proximity elevator) near the operated button. For example, for a nearby elevator, the value indicated as the nearby elevator priority Ba is set to Bonus Bm, and for other elevators, 0 is not added to the bonus Bm.

C5] The light-load elevator priority assignment function is a function that makes it easier to assign an empty elevator or an elevator with a small load (light-load elevator). For example, for a lightly loaded elevator, the value indicated by the lightly loaded elevator overnight priority Bb is set as the bonus Bm, and for other elevators, 0 is set as the bonus Bm.

[6] The specific elevator priority assignment function makes it easy to assign a specific elevator Function. For example, the value indicated by the specific elevator priority Be is set to Bonus Bm for underground elevators, rooftop elevators, observation elevators, etc., and bonus Bm for all other elevators. 0 is set.

As described above, Ca, Cb, Pa, Pb, Ba, Bb, and Be are parameters for group management related to the assignment evaluation function [1]. Even if an assignment is performed using the above assignment evaluation function [1], an unexpected call may occur after that, causing a long wait. Therefore, the group management system has [7] additional assignment function and [8] assignment change function.

[7] The additional allocation function for long waiting calls is a function to perform additional allocation to rescue another elevator that can service earlier than the currently allocated elevator.

[8] The function to change the assignment for long waiting calls is to shift the assignment (and forecast) of calls in long waiting state to the rescue elevator overnight. Judgment criterion value DL is set to detect whether or not the user has been waiting for a long time. In addition, each elevator in the group management system has a [9] automatic full-passage function, and when the car load exceeds the reference value DB, it automatically passes through the landing even if it has already been assigned. I do. In addition, for calls that are automatically passed when full, rescue operation will be performed using the [10] Assignment change function.

[10] The function of changing the allocation for calls that are full is to shift the allocation and forecast of hall calls that have been automatically passed when they are full to another rescue elevator. The change of the forecast to the new assigned car is called a forecast change.

As mentioned above, the DL and DB mentioned above are also a parameter for group management. Various control parameters are also used for operations other than those related to the hall call. One night is used. For example, various control parameters are used as conditions for selecting or canceling the following operation patterns.

[11] Choice of driving at work · Elimination

Driving during commuting is selected when the start time of the commuting time period has elapsed, and the number of cars registered for the starting elevator on the main floor (main floor) has exceeded the judgment reference value D1 UPC. On the other hand, it is canceled when the end time of the work hours has passed.

[12] Selection of up peak operation · Elimination

Up-peak operation is selected when the number of passengers on the main floor exceeds the first criterion value DUP1 and an elevator departs, while the number of passengers on the main floor exceeds the second criterion value DUP2. Eliminated when no elevators depart during the DUPT decision time.

[13] Select and eliminate down-peak operation

Down-pick operation is selected when an elevator that descends with passengers of the first criterion value DDP1 or more occurs, while the elevator that descends with passengers of the second criterion value DDR2 or more falls in the judgment time. Dissolved during the DDPT when no units are generated. Each operation pattern includes the following control, but also includes control parameters.

[14] Driving at work

• Distribute the number of elevators designated as DIUPN on the main floor o

• In the departure adjustment operation that delays departure with the door open even if there is a call to respond to the starting elevator on the main floor, the departure time (referred to as the time to keep the door open) Set to the value D1UPT. • On the main floor, the number of elevators designated as the DI UPW waiting for doors to be opened is set to wait for doors to open, and other elevators are set to wait for doors to close.

[15] Up-peak operation

• Distribute the number of elevators designated as DUPN to the main floor.

[16] Down peak operation

• When calculating the waiting time, for the hall call in the direction to the main floor, the estimated waiting time is calculated to be longer by the amount corresponding to the priority DDPE.

[17] Distributed standby operation

• When an empty elevator occurs, this operation is performed when the elevator is waiting on the floor where a call is likely to occur next to reduce the waiting time, and is selected when the congestion operation is not selected. .

• When the number of empty elevators equal to or greater than the specified number D0HN continues for the specified time D0HT or more, the distributed standby operation is executed.

• The floor on which the elevator waits (standby floor) and the number of vehicles on standby may be used as control parameters. In addition, the following additional control also includes control parameters to control the number of operating units.

[18] Power saving operation

This is an operation to reduce power consumption by automatically reducing the number of operating units according to the service status.If the average waiting time for the last 5 minutes is less than the first service reference value DESW1, the current operating number is 1 The number of operating units will be increased by one when the service level is equal to or greater than the second service reference value DESW2. As mentioned above, the group management algorithm includes many parameters ing. These parameters are provided to satisfy various control objectives such as shortening waiting time, improving forecast accuracy, improving passenger comfort, and saving power. However, since some parameters have conflicting control objectives, the performance of group management is greatly affected by the combination of numerical values assigned to each parameter.

In other words, it is necessary to quickly search for the optimal combination of parameter values in order to carry out efficient group management operation according to the various traffic conditions of the building, which change from moment to moment, and the demands of customers. is there. In the following description, the combination (parameter sequence) of the parameter

Called "parameter overnight set" or simply "set." As a conventional optimal set search method, there is a brute force method in which all combinations of parameter values (that is, all sets) are examined (for example, Japanese Patent Publication No. Hei 4-5-151475). And JP-A-57-57168.

With the brute force method, there is almost no problem when the types of parameters are small.

However, as the number of parameter sets increases, the number of combinations of parameter set values to be considered increases dramatically. For this reason, it is extremely difficult to consider all combinations and select the optimal set. It will be difficult. The details are described below.

In the brute force method, the number of parameter types is M and the number of possible values of the parameter is L. If M = 3 and L = 6, the total number of simulations is L ^M = 2 16 times. Therefore, when the types of parameters and the number of possible values are large, whether the simulation is performed or the test operation of the elevator with an actual group It takes a very long time to determine a combination of parameter values, which is not practical.

Thus, in the system described in Japanese Patent Publication No. 5-244067, a proposal has been made to reduce the number of simulations. In other words, in this system, for example, when there are two parameters, first find the optimal value for the first parameter, then fix the value of the first parameter to the optimal value, It is a method of finding a value. According to _c This sequential method is referred to as a sequential method of this, if, all simulation times in the case of M = 3 and L = 6 becomes the L x M = 1 8 times, shea Muret compared to the round robin The number of sessions is greatly reduced.

However, the above-mentioned sequential method is effective only when there is no correlation in the parameter overnight, and when there are many highly correlated parameters such as the parameter group for group management control , Not applicable.

Disclosure of the invention

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described conventional problems, and it has been found that, for a group management parameter group having a special property having a strong correlation, even if the number of parameter value sets is enormous, the optimal set can be efficiently performed. specifically _c also aims to explore, the present invention is, on the basis of general technical called "Yaden algorithm" (Genet ic a lgorithm), realizing a unique search method for optimal set search The purpose is to do.

(1) In order to achieve this object, the search device according to the present invention comprises:

Storage means for storing a plurality of sets;

Generating means for selecting one or more sets from the storage means as a parent and generating one or more new sets partially inheriting the properties of the parent; Evaluation means for obtaining an execution result when the group management algorithm is executed using the new set as a group management performance value;

Selecting means for improving the plurality of sets stored in the storage means by using the addition of the new set to the storage means and the deletion of the bad set from the storage means; and

Extracting means for extracting the optimum set from the plurality of sets stored and improved in the storage means based on the group management performance value.

According to the above configuration, the generation of a superior set can be increased by the generation of a genetic set and the selection of a superior set, and at the same time, a child set that inherits only the good properties of the parent set. (New set) can be stored in the storage means. That is, by repeating a series of cycles, a plurality of sets stored in the storage means can be sequentially updated to improve the quality. Then, finally, the optimal set is extracted from the storage means based on the group management performance value. Each numerical value constituting the optimal set is substituted into each parameter in the group management algorithm, and group management such as elevator assignment is performed.

As described above, according to the present invention, it is possible to efficiently search for the most excellent set or the set having a content very similar to the best set. That is, the amount of calculation and the number of simulations can be reduced, so that the search can be sped up.

(2) In the present invention, the generating means includes:

Numerical exchange means for generating two new sets by exchanging some numerical values with each other between two sets selected from the storage means; and A new value replacement means for generating one new set by replacing some parameter values in one set with new randomly generated values; A generation method selecting means for stochastically selecting the numerical value exchange and the new value replacement,

including.

According to the above configuration, the generation of the new set is performed by the "crossover"

(crossover) and "mutation" by new value replacement means and-are selected at random.

In general, crossover has the property of converging solutions, and mutation has the property of producing diversity in solutions. Therefore, according to the crossover, the contents of the set group stored in the storage means can be converged, but on the other hand, the diversity of the set group is lost at an early stage, and the local solution (local solution) is lost. ion) and lose the true solution (optimal set). In that case, the mutation can break away from the local solution. In that sense, crossovers and mutations are mutually exclusive.

—On the other hand, mutations tend to destroy good solutions searched by crossover. In that sense, crossover and mutation are competitive. Therefore, it is necessary to appropriately set the ratio between the crossover rate (crossover rate) and the mutation rate (mutation rate). In any case, by appropriately using both crossover and mutation, the advantages of both can be fully utilized and the probability of generating an excellent new set can be improved.o

However, in some cases, it is possible to search for the optimal set only by crossover or mutation.

(3) In the present invention, the generating means includes:

A parent selecting means for selecting one or two sets from the storage means; and a parameter selecting means for selecting, for the one or two sets, parameters for which the numerical value exchange or the new value replacement is performed. including.

According to the above configuration, the parent selecting means selects two parent sets (set pairs) from the memory means when the crossover is selected, and selects the parent set from the memory means when the mutation is selected. Select one parent set. The parameter overnight selection means selects a parameter overnight position (crossover position or mutation position) at which parameter values are exchanged in crossover and mutation.

(4) The above-mentioned parent selection means selects a parent based on parent selection criterion information for increasing the generation probability of an excellent new set. By using parent selection criteria information, the probability of generating an excellent new set can be increased.

(5) If the distance between sets is used as parent selection criterion information, the similarity between sets can be used as the selection criterion. This makes it possible to prioritize the diversity of the new set to be generated or give priority to the convergence of the new set to be generated.

(6) If a group management performance value is used as parent selection criterion information, a parent set can be selected based on the excellence of each set. This increases the probability that a good parent set will be selected, and consequently the probability that a better new set will be generated.

(7) If the number of similar sets is used as parent selection criterion information, the uniqueness of each set can be used as a selection criterion. As a result, the probability of selecting a set pair having different characteristics from each other is increased, and the diversity of the new set to be generated can be secured.

(8) It is desirable to modify the parent selection conditions in the parent selection means according to the search progress. For example, a plurality of parent selection conditions can be prepared and switched according to the progress, or the reference value of the parent selection conditions can be changed according to the progress. In this way, the probability of generating a good new set can be increased.

(9) The above parameter selection means selects a parameter based on parameter selection criterion information for increasing the generation probability of an excellent new set. _C Accordingly, capable of enhancing the probability of excellent new sets Bok is generated

(10) If a parameter value difference is used as parameter selection criterion information, similarity between parameters can be used as a selection criterion. Thereby, it is possible to give priority to the diversity of the new set to be generated, or to give priority to the convergence of the new set to be generated.

(11) If the degree of association with the elevator usage is used as parameter selection criterion information, the probability that a parameter having a high degree of relation to the elevator usage will be selected will be increased, and a better new set will be generated. The probability of breaking is increased.

(12) If the degree of relevance to the contents of the performance evaluation value is used as the parameter selection criterion information, the probability that a parameter having a high degree of relation to the contents of the performance evaluation value is selected increases, and a better new set can be created. The probability of generation is increased.

(13) It is desirable that the parameter selection conditions in the parameter selection means be modified according to the progress of the search. For example, a plurality of parameter selection conditions can be prepared and switched according to the progress, or the reference value of the parameter selection conditions can be changed according to the progress. In this way, the probability of generating an excellent new set can be increased.

(14) It is desirable that the selection probability of each generation method be corrected by the probability correction means according to the progress of the search. This makes it possible to take advantage of the features of both the global search by mutation and the local search by crossover, thereby improving search efficiency.

(15) In that case, it is desirable to correct the selection probability by probability correction means, for example, according to the success index. According to this, the degree of progress of the search can be determined from the degree of convergence of the search, and a selection probability suitable for the degree of progress can be set.

(16) Also, in order to achieve the above object, a search device according to the present invention comprises: a storage unit for storing a plurality of sets; By exchanging some parameter values between the two sets selected as parents from the storage means, two new sets that partially inherit the properties of the parents are generated. Numerical exchange means,

By replacing some parameter values in one set selected as a parent from the storage means with new randomly generated values, one new set partially inheriting the properties of the parent New price substitution means for generating

A generation method selecting means for stochastically selecting the numerical value exchange and the new value replacement,

Evaluation means for obtaining an execution result when the group management algorithm is executed using the new set as a group management performance value;

Additional means for adding and storing only good new sets satisfying predetermined additional conditions to the storage means;

Deleting means for deleting a bad set satisfying a predetermined deletion condition from the storage means;

Extracting means for extracting the optimum set from the plurality of sets stored and improved in the storage means, based on the group management performance value.

According to the above configuration, a set generation (crossover) by numerical exchange and a set generation (mutation) by replacement of a new value are randomly selected to generate a new set. Of the generated new sets, only the new sets that satisfy a predetermined additional condition are stored in the storage means. On the other hand, the bad set is deleted in the storage means.

When such a cycle is sequentially repeated, only the best set is stored in the storage means, and the optimum set is extracted therefrom. Then, each numerical value constituting the optimal set is substituted into each parameter of the group management algorithm.

Three Therefore, according to the present invention, it is possible to efficiently search for the most excellent set or the set having a content very close to it. In other words, the amount of calculation and the number of simulations can be reduced, so that the search can be speeded up.

(17) The search device according to the present invention preferably further includes an additional condition correcting means.

(18) The additional condition is determined based on, for example, the group management performance value of each set stored in the storage means, and the additional condition is gradually set stricter. In this way, only a good new set can always be stored in the storage means, and unnecessary processing can be reduced, and the probability of generating a good new set can be increased.

(19) The above-described deletion means performs deletion based on, for example, a group management performance value. In this way, it is possible to increase the excellence of the stored multiple parent sets as a whole, leaving only the excellent set.

(20) In addition, the above-described deletion means performs deletion based on, for example, the distance between sets. In this way, with respect to a plurality of sets in the storage means, it is possible to avoid a state in which similar sets are duplicated, and to ensure the diversity of sets.

(21) The search device according to the present invention preferably further has an initial setting means. The search time can be reduced by initializing using the initial set group that matches the search conditions as much as possible.

(22) The initial setting means desirably includes a first mode and a second mode. In the first mode, multiple sets prepared in advance are used as the initial set group, and in the second mode, multiple sets that were improved in the previous search are used as the initial set group. . If an appropriate mode is selected according to the state at the start of the search, the convergence of the search can be expedited.

(23) The search device according to the present invention preferably further includes a search end determination unit.

Four Including. This means determines the end of the search when the search progresses and a state where sufficient improvement can be expected. It is possible to avoid termination in an insufficient search state, and to avoid useless search.

(24) The number of sets evaluated is related to the number of times the optimization cycle is executed, and can be used as a termination criterion.

(25) The number of added sets indicates the degree of improvement in the storage unit, which can be used as a termination judgment criterion.

(26) The success index is the ratio of the number of evaluated sets to the number of added sets. Since it is an indirect expression of the convergence of the search, it can be used as a termination criterion.

(27) Since the distance between sets expresses the similarity of a plurality of sets in the storage unit as a whole, it can be used as a termination judgment criterion.

(28) The search device according to the present invention desirably further includes re-search determination means. This means determines a re-search based on a change in the precondition given at the start of the search. According to this, the optimum set can be automatically searched under new conditions. Here, the above-mentioned preconditions include, for example, elevator specifications, traffic flow specifications, performance reference values, control reference values, and the like.

(29) In the present invention, a group management performance value can be further stored in the storage means.

(30) In the present invention, a target value setting device can be connected to the search device in order to set a target value for the search. You can freely set the control target and search for the optimal set that matches the set target.

(31) In the present invention, when a new set is evaluated using a device dedicated to simulation, a simulation device is connected to the search device in addition to the group management device. Here, the simulation device includes the same group management algorithm as the group management algorithm included in the group management device. No. Then, the evaluation means sets the simulation execution result as a group management performance value. If a simulation device is used, a new set can be evaluated without interrupting group management.

(32) In the present invention, for example, the group management device is arranged in the same building together with the search device (and the simulation device). However, when it is necessary to arrange the search device (and the simulation device) separately from the group management device, the group management device and the search device are connected by a communication line. If one search device (and simulation device) is shared by multiple group management devices, system cost can be reduced.

(33) In the present invention, the simulation can be performed using a group management device as an actual device connected to the search device. Since a simulation device is not required, the system cost can be reduced.

(34) In the present invention, when the search device is arranged separately from the group management device, the group management device and the search device are connected by a communication line. If one search device is shared by a plurality of group management devices, the system cost can be reduced.

(35) In the present invention, the simulation device and the search device are connected, and the present invention can be extended and applied to evaluation of a group management algorithm and the like. (1) Relationship between the present invention and GA

Genetic algorithms (GAs) have been described in various documents (eg, “Measurement and Control” (Volume 32, No. 1, January 1993)). Current Status and Issues of Genetic Algorithms ”). The basic GA includes a series of cycles of initialization, parent selection, crossover, mutation, and alternation of generations.

The 34th Automatic Control Federation Lecture Meeting (January 21--22, 1991) "Preliminary Proceedings for Solving Elevator Group Management Car Assignment Problems for Definite Calls by Genetic Algorithm" Applied GA to elevator group management

6 one System has been described.

Here, this conventional system uses GA to perform an optimal assignment of "call permutations" for the entire night. Therefore, in the present invention and the conventional system, the common force search target differs for the system related to GA, and the basic configuration and the like are greatly different.

In short, the present invention does not merely use GA, but provides a new technology for searching for a new optimal parameter value set. It is made clear by the unique features of the present invention due to the unique nature of the parameter value set. BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 is a block diagram showing the overall configuration of Embodiment 1 of the system according to the present invention.

FIG. 2 is a diagram showing a configuration of a search device composed of a microcomputer.

FIG. 3 is a diagram showing a configuration inside the RAMIOC of FIG.

FIG. 4 is a diagram showing a configuration inside the ROM 10B of FIG.

Fig. 5 is a diagram showing the structure of the elevator specification data (ELS).

Figure 6 is a diagram showing the structure of traffic flow specification data (TRS).

FIG. 7 is a diagram showing a configuration of group management performance data (PRF).

FIG. 8 is a diagram showing a configuration of a parameter value set (EPS).

FIG. 9 is a flowchart showing the contents of the control program in the first embodiment.

FIG. 10 is a flowchart showing a search command program according to the first embodiment.

FIG. 11 is a flowchart illustrating a search main program according to the first embodiment. FIG. 12 is a flowchart illustrating a search start determination program according to the first embodiment.

FIG. 13 is a flowchart showing an initial setting program according to the first embodiment.

FIG. 14 is a flowchart showing a new set generation program in the first embodiment.

Figure 15 is a flow chart showing the evaluation program in Example 1.

FIG. 16 is a flowchart showing an additional program in the first embodiment.

FIG. 17 is a flowchart showing the deletion program in the first embodiment.

FIG. 18 is a flowchart showing the additional reference value correction program in the first embodiment.

FIG. 19 is a flowchart illustrating a search end determination program according to the first embodiment.

FIG. 20 is a flowchart showing an optimal set extraction program in the first embodiment.

FIG. 21 is a block diagram illustrating the system configuration of the second embodiment.

FIG. 22 is a diagram illustrating a configuration of the RAM in the second embodiment.

FIG. 23 is a flowchart showing an additional reference value correction program in the second embodiment.

FIG. 24 is a flowchart illustrating a search start determination program according to the second embodiment.

FIG. 25 is a flowchart showing the initial setting program in the second embodiment. is there.

FIG. 26 is a flowchart showing a deletion program according to the third embodiment. O

FIG. 27 is a flowchart showing a search end determination program according to the fifth embodiment.

FIG. 28 is a flowchart illustrating a search end determination program according to the sixth embodiment.

FIG. 29 is a flowchart showing an optimum value extraction program according to the seventh embodiment.

FIG. 30 is a block diagram illustrating a system configuration according to the eighth embodiment. FIG. 31 is a flowchart showing a search main program according to the eighth embodiment.

FIG. 32 is a block diagram showing a system configuration of the ninth embodiment.

FIG. 33 is a flowchart showing a search main program according to the ninth embodiment.

FIG. 34 is a flowchart showing the appearance rate correction program in the ninth embodiment.

FIG. 35 is a flowchart showing the appearance rate correction program in the tenth embodiment.

FIG. 36 is a block diagram illustrating a system configuration of the eleventh embodiment.

FIG. 37 is a flowchart showing an operation main program in the eleventh embodiment.

FIG. 38 is a flowchart showing a part of the new set generation program in the eleventh embodiment.

FIG. 39 is a flowchart showing the selection condition correction program in the embodiment 11.

FIG. 40 is a flowchart showing a selection condition modification program in the embodiment 12.

FIG. 41 shows a part of a new set generation program in the embodiment 13

9 one It is a flow chart.

FIG. 42 is a flowchart showing a selection condition correction program in the embodiment 13.

FIG. 43 is a flowchart showing a selection condition correcting program in the embodiment 14.

FIG. 44 is a flowchart showing part of a new set generation program in the fifteenth embodiment.

FIG. 45 is a diagram illustrating an appearance rate for each parameter in the 15th embodiment. FIG. 46 is a flowchart showing a part of the new set generation program in the embodiment 16.

FIG. 47 is a diagram illustrating the system configuration of the seventeenth embodiment.

FIG. 48 is a diagram illustrating the system configuration of the eighteenth embodiment.

FIG. 49 is a diagram illustrating the system configuration of the nineteenth embodiment.

FIG. 50 is a conceptual diagram showing the optimal set search method according to the present invention. Example

[Basic principle]

As described above, the group management algorithm includes a plurality of types of parameters. In order to efficiently manage multiple elevators in groups, it is necessary to search for optimal combinations of parameter values according to traffic conditions. The device for that is an optimum set search device, and FIG. 50 shows the basic principle of the search device according to the present invention. As described above, a combination (sequence) of parameter values is called a “parameter value set” or simply a “set”.

In FIG. 50, the optimal set is searched for by repeatedly generating a new set and selecting a good set. The details are described below. First, an initial setting is performed on the storage unit A2 (A1). For example, a plurality of initial sets prepared in advance are stored in the storage unit (A2).

Next, a new set is generated (A4). Here, a new set is generated by randomly selecting one of the generation methods from numerical exchange (crossover) and new value substitution (mutation). If crossover is selected, two sets (parent set ♦ pairs) are fetched from storage A2, some parameter values are exchanged between the two sets, and two new sets are created. Is generated. If a mutation is selected, one set (parent set) is fetched from memory A2 and some of the parameter values in that set are replaced with new, randomly generated numbers. And one new set is generated.

Note that selection of the generation method, selection of the parent set, and selection of parameters for replacing numerical values are basically performed randomly. It should be noted that selection conditions can be defined for each, and selection probability can be weighted for each selection element.

Next, the generated new set A5 is evaluated (A6) _c, that is, the group management algorithm to which the new set is assigned is executed virtually or actually, and the execution result is obtained. Can be The execution result is obtained as a “group management performance value” indicating the performance of the new set, and a new set having an excellent group management performance value is stored in the storage unit A2 (A8). The bad set is selected without being stored (A9) or deleted after storing (A10). Due to the selection of such excellent sets (A7), only excellent sets are always stored in the storage unit A2.

When more excellent reduction cycle is repeated sequentially, a plurality of sets A3 stored in the storage unit A 2 is carefully gradually, and _c will be more excellent reduction, finally, stored plurality of From the set A3, one best set is extracted as the optimum set (All), and the optimum set is extracted. The group is supplied to the group management algorithm, and group management is performed.

According to the optimal set search method described above, a child set that inherits the good properties of the parent set can be efficiently generated. That is, to increase the probability of generating excellent child set from excellent parent sets, the search _t here can be performed rapidly, it is also possible to employ only one of the crossover and mutation, Desirably, both can be selected at random, so that both convergence and diversity can be appropriately achieved for a plurality of stored sets.

[Example 1]

Description of one configuration

1 to 20 are diagrams showing a first embodiment of an elevator group management system according to the present invention. FIG. 1 shows the overall configuration of the system. This system includes a well-known group management device 1, a well-known simulation device 2, and a search device 10.

The group management device 1 is composed of a microcomputer. In this example, the four elevators installed in a 10-story office building are group-managed. As described above, the group management device 1 has a group management algorithm (see FIG. 9) including a plurality of types of control parameters.

Four car control devices 1A to 1D are connected to the group management device 1 via communication cables. Each of the car control units 1A to 1D is composed of a computer with a microphone, and performs various controls on the elevators in charge. Each of the car control devices 1A to 1D has a car call registration function, an operation control function, a door control function, a display control function, and the like.

Here, the function of registering a car call is a function for registering a car call when it occurs in the storage unit. The operation control function sets the elevator to respond to calls to be answered (car call, assigned hall call). It is a function to control driving, stopping, determining the driving direction, etc. in the evening. The door control function controls the opening / closing timing and door opening time of the elevator door and the landing side door. The display control function is a function that turns on the hall lantern to notify the waiting passengers at the landing in advance of the assigned elevator or that the arrival of the elevator is notified by blinking the hall lantern. The car control devices 1A to 1D transmit to the group management device 1 a signal indicating the operation state (car position, operation direction, door opening / closing state, car call, etc.). On the other hand, the group control device 1 transmits signals representing various commands (a command for allocating a hall call, a reference value DB for full passage, a set value of a door opening time, etc.) to the car control devices 1A to 1D.

The group management device 1 outputs to the search device 10 a search condition signal 1a representing a condition for searching for the optimal set. The search condition signal 1a is used for `` elevator specification data '' required to simulate the group control and elevator system on a computer and `` elevator specification data required for simulating the traffic flow in a building on a computer. '' Traffic specification data "and" search command data "to command the search for the optimal set.

The above-mentioned elevator specification data is composed of, for example, data indicating the number of elevators, speed, capacity, stop floor, door type, and the presence or absence of additional operations such as power saving operation and operation at work. Is done. For example, if the traffic flow in a building is indirectly specified, the data that combines characteristic values such as the total number of passengers per hour and the floor-to-story traffic ratio, and the floor In addition, when the traffic flow in the building is directly specified, the passenger data for all passengers (occurrence data Time, generation floor, destination floor, etc.). The simulation device 2 is composed of a microcomputer,

One two It has the same group management algorithm as the logical device 1. The simulation device 2 receives a simulation condition signal 13a composed of elevator specification data, traffic flow specification data, and a parameter value set. The signal 13a causes the simulation device 2 to virtually operate a plurality of elevators under the same conditions as actual conditions under the group management algorithm. Then, after execution, group management performance data indicating statistical results (average waiting time, long wait rate, etc.) indicating group management performance are output as group management performance value signal 2a. The search device 10 is configured by a microcomputer, and searches for an optimal set as described above.

In such a search device 10, the storage unit 11 stores a plurality of parameter set values, and also stores the group management performance data in association with each set. Note that the output signal 11a from the storage unit 11 is composed of parameter value set data and group management performance data.

The generation unit 12 generates a new set by the “crossover” and the “mutation” described above. The new set is temporarily stored in the generation unit 12 until it is evaluated by the evaluation unit 13 described below. The new set signal 12 a is output from the generation unit 12.

The evaluation unit 13 creates a simulation condition signal 13a based on the search condition signal 1a and the new set signal 12a, and outputs the simulation condition signal 13a to the simulation device 2. After executing the group management simulation, the evaluation unit 13 creates an evaluation result signal 13b based on the group management performance signal 2a output from the simulation device 2, and adds it. Output to Part 15

The additional reference value memory 14 stores an additional reference value for determining whether the new set evaluated as described above is to be additionally registered in the storage unit 11 or to be eliminated. Is stored. The additional reference value signal 14 a is output from the additional reference value memory 14.

Additional portions 1 5, voted result signal 1 3 group management performance data contained in b, additional registration to create a decision performance evaluation value for which the additional reference value and _c performance evaluation value is added the reference value for comparing If it is better, a signal 15a including the new set and its group management performance value is created and output to the storage unit 11. As a result, an excellent new set is additionally registered in the storage unit 11.

The deletion unit 16 obtains a performance evaluation value for deletion determination for each set based on the group management performance data when a predetermined condition regarding the set registration status is satisfied. Then, the deletion unit 16 selects a set having a poor performance evaluation value, and outputs a deletion command signal 16a indicating the set number. As a result, the registration of the specified set from the storage unit 11 is peripherally performed.

The end determination unit 17 determines whether or not to end the search, and outputs a search end signal 17 a to the generation unit 12 when determining that the search is to end. This ends the generation of the new set.

The additional reference value correction unit 18 corrects the additional reference value stored in the additional reference value memory 14 by the correction signal 18a. The modification degree is determined based on the group management performance data of each set in the storage unit 11.

The re-search determination unit 19 monitors the search command signal 1a, and outputs a re-search command signal 19a for re-executing the search for the optimal set when the elevator specification or traffic flow specification changes. . When this signal 19a is output, the search end command by the search end signal 17a is invalidated and the search is started again from the beginning, and the search is started again from the beginning even during the search.

The extraction unit 20 obtains a performance evaluation value for determining an optimal set based on the group management performance data of each set in the storage unit 11, and extracts one set having the best evaluation value. That is, the optimal set is extracted. The extraction unit 20

Two The signal 20a output from is composed of an optimal set, elevator specification data, traffic flow specification data, and search state data.

The initial setting unit 21 includes a plurality of initial set groups. At the start of the search, the initial setting unit 21 selects one of the plurality of initial set groups stored in advance according to the search condition signal 1a or the re-search command signal 19a. The appropriate set group used in the setting is specified and output to the storage unit 11. FIG. 2 shows a hardware configuration of the search device 10 shown in FIG. In FIG. 2, the search device includes a microprocessor 10A, a read-only memory (ROM) 10B, a readable / writable memory (RAM) 10C, an input interface circuit 10D, and an output interface circuit 10 It consists of E and Here, the ROM 10B stores a search program describing the operation procedure of the microprocessor 10A and fixed data. In addition, the RAM 10C stores the operation result (operation data) of the microprocessor 1OA, the contents of the search condition signal 1a and the group management performance value signal 2a (input data) input from outside, and the simulation to output to the outside. It stores the contents (output data) of the operation condition signal 13a and the optimum set signal 20a. FIG. 3 shows the storage contents of the RAM 10C of FIG. 2, and FIG. 4 shows the fixed data portion of the storage contents of the ROM 10B. In Fig. 3, ELS is data indicating elevator specifications, TRS is data indicating traffic flow specifications, and SCM is data indicating search commands. These input data are included in the search condition signal 1a shown in FIG. Fig. 5 shows the specific structure of the elevator specification data ELS. In the example shown in Fig. 5, the number of elevators is four, the speed is 12 OmZ, the capacity is 20 people, the stop floor is the first floor at the lowest floor and the top floor is 10 stops at the 1◦ floor, door width Is set to 100 O mm. Among the priority allocation operations, [2] ride time priority allocation function, [3] power saving allocation function, [4] proximity elevator priority allocation function, and [5] light load elevator priority allocation function are all described as “ "Enable" and [6] Specific elevator priority assignment function is set to "Invalid". Among the various driving patterns, [11], [14] At-work operation, [12], [15] Up-peak operation , [13], [16] Down peak operation, and [17] Distributed standby operation are all set to “valid”, and the other additional operation [18] power saving operation is also set to “valid” (Although not shown, [8] Allocation change operation for long waiting calls, [9] Automatic packed passage function, and [10] Allocation change operation for crowded hall calls And always “valid”).

Figure 6 shows the specific structure of the traffic flow specification data TRS. The example shown in FIG. 6 relates to the time zone of business hours (14: 00 to 15: 00). For example, the total number of passengers per hour is 500, based on the results of actual measurement of the traffic flow in advance using the group control device 1, and the first floor and ground floor (2 to 10 floors) of the total traffic volume Is 80%, the ratio of traffic in the up direction to the total traffic (-up traffic ratio) is 50%, and the traffic in the down direction is 50%. Is set at 50%. The PRF at the upper left of Fig. 3 is data indicating group management performance indicating the excellence of each set, and corresponds to the group management performance signal 2a in Fig. 1.

Fig. 7 shows the specific structure of the group management performance data PRF. In this example, the group management performance data PRF includes the average waiting time AWT, long waiting rate RLW, maximum waiting time MWT, forecast loss rate RPE, forecast change rate RPC, packed passage occurrence rate RBP, average riding time ABT, and maximum riding time. Time MBT, Power consumption PWC, Proximity elevator Evening response rate RNR (When a landing button is operated, the Beta response rate), Light-load elevator response rate RLR (Ratio of light-load elevators assigned when a hall call is registered), and Specific elevator response rate RSR (Registered elevator when a hall call is registered) (Assigned ratio). Returning to FIG. 3, P at the upper right of the figure is data representing the number of sets (also referred to as excellent sets) registered in the storage unit 11, and EPS (1) to EPS (Pmax) are Data representing sets from set numbers 1 to Pmax, and PRE (1) to PRE (Pmax) are data representing group management performance values corresponding to EPS (i) to EPS (Pmax). Here, the number of sets P, the set data EPS (l) to EPS (Pmax), and the group management performance data PRE (l) to PRE (Pmax) correspond to the signal 11a in FIG. Here, Pmax described later is data representing the maximum value of the number of sets that can be registered. _C FIG. 8 shows a specific configuration of a parameter set as an example. In FIG. 8, this set is composed of 25 types of control parameters. That is, each set data EPS (l) to EPS (Pmax) shown in FIG. 3 is configured as shown in FIG. The group management performance data PRE (l) to PRE (Pmax) in FIG. 3 have the same configuration as the group management performance data PRF in FIG. 3 (for the specific configuration, see FIG. 7). In the middle part on the right side of Fig. 3, Pn is data representing the number of new sets generated, and NPS (l) to NPS (Noiax) are data representing new sets with set numbers 1 to Nmax. The number of new sets Pn and the new sets NPS (1) to NPS (Nmax) correspond to the signal 12a in FIG. Here, Nmax described below is data representing the maximum value of the number of new sets that can be generated.

SIM at the upper right of Fig. 3 is output data corresponding to the simulation condition signal 13a in Fig. 1, and includes evaluation set data NPSX and Elevator overnight specification data. Evening ELSX and traffic flow specification data TRSX. The evaluation set data NPSX is data indicating the contents of the new set for which group management performance should be evaluated by simulation, and has the same configuration as the EPS in Fig. 8. Elevator overnight specification data ELSX and traffic flow specification data TRSX are data representing the elevator specification and traffic flow specification at the time of simulation, and are ELS in Fig. 5 and Fig. 6 respectively. It has the same configuration as TRS. RES at the upper left of Fig. 3 is data corresponding to the evaluation result signal 13b in Fig. 1, and consists of the number of evaluations NE, the evaluation set NPSY, and the group management performance data PRFY. Evaluation frequency NE is data representing the cumulative number of evaluations. The evaluation set NPSY is data representing the new set after the group management performance has been evaluated by simulation, and has the same configuration as the EPS in Fig. 8. Further, the group management performance data PRFY is data representing the group management performance value obtained by simulation, and has the same configuration as the PRF in FIG.

BX is data representing an additional reference value for judging whether or not to register an evaluated new set additionally, and is a data equivalent to the additional reference value signal 14a in FIG.

RAP is data corresponding to the additional registration signal 15a in Fig. 1, and consists of the number of additional registrations, the evaluation set NPSZ, and the group management performance data PRFZ. The number of additional registrations NR is data representing the number of times additional registration has been determined. The evaluation set NPSZ is a data set representing an excellent new set registered in the storage unit 11, and has the same configuration as the EPS in FIG. The group management performance data PRFZ is data that represents the group management performance when a group management simulation is performed using the evaluation set NPSZ, and has the same configuration as the PRF in Fig. 7.

Fig. 3 The RP in the middle part on the left side is data indicating the number of the set whose registration is to be deleted as a poor set for the P sets EPS (1) to EPS (P) registered. This is the data corresponding to the delete command signal 16a. FLAG is data (search permission flag) indicating whether to continue the optimal set search or to terminate the search, and is data corresponding to the search end signal 17a in FIG.

CBX is data for newly rewriting the additional reference value BX, and is data corresponding to the correction signal 18a in FIG.

STR is data for instructing re-execution of the optimal set search, and is data corresponding to the re-search command signal 19a in FIG.

The BPD is an output data corresponding to the optimal set signal 20a in FIG. 1, and is composed of an optimal set BPS, elevator specification data ELSY, traffic flow specification data TRSY, and search state data SS. The optimal set BPS is the set with the best performance evaluation value among the registered sets, and has the same configuration as the EPS in Fig. 8. Elevator specification data ELSY and traffic flow specification data TR SY are data representing elevator specifications and traffic flow specifications when a group management simulation was performed using the optimal set BPS. It has the same configuration as the ELS in Fig. 5 and the TRS in Fig. 6. Further, the search state data SS is data representing a search state when the optimum set BPS is selected. In this embodiment, a value indicating the number of evaluations NE is set.

GPS0 in the lower left part of Fig. 3 is data corresponding to the initial setting signal 21a in Fig. 1, and the number of initial sets Pk, multiple initial sets IPS (l) to IPS (Pk), and multiple groups It consists of management performance data PRI (l) to PRI (Pk). The initial set number Pk is data representing the number of sets at the start of the search, and is usually set to the same value as the judgment value Pe for judging the end of deletion. Further, a plurality of initial sets I PS (1) to I PS (Pk) are prepared in advance as a set group at the start of the search, and have the same configuration as the EPS of FIG. Group management performance data PRl (l) to PRl (Pk) are data that represent the group management performance when a group management simulation was performed using the initial set IPSh) to IPS (Pk). It has the same configuration as the PRF in Fig. 7. VPD (l) to VPD (Pmax) on the right side of Fig. 3 are performance evaluation values for deletion VPE-VPE (Pmax) is the performance evaluation value for setting the additional reference value used to modify the additional reference value BX, and VPS (l) -VPS (Pmax) is the optimal set BPS The performance evaluation value for judging the optimal set to be used when performing the evaluation, and the VP in the lower left part of Fig. 3 are the performance evaluation values for the additional judgment used to judge whether to add the evaluation set NPSY. In this embodiment, the average waiting time AWT extracted from the group management performance data is substituted for each performance value as it is.

NP is data representing the set number for which group management performance should be evaluated in the new sets NPS (l) to NPS (Nmax).

WVPE is the data representing the worst value among the performance evaluation values, BVPE is the data representing the best value among the performance evaluation values, and RC is the set number used when searching for the worst value WVPE or the best value BVPE. The search counter for counting, BP, is data representing the number of the registration set having the best value BV PE.

PS1 is the first parent set number that indicates the number of the parent set for creating a new set, PS2 is the same second parent set number, and PX is the target of “crossover” or “mutation” The data represent the number (position) of the parameters, CR represents the crossover selection probability (occurrence rate), and MR represents the mutation selection probability (appearance rate). In FIG. 4, Pmax is data representing the maximum number of sets that can be registered, and Nmax is data representing the maximum value of the number of new sets that can be generated. In the first embodiment, Pmax = 50 , Nmax = 20 are set.

NEa is a search termination judgment value used when judging whether or not the search for the optimal set has converged by the number of searches NE. In the first embodiment, the NEa force is set to q.000 times.

Ps is the deletion start judgment value used when judging whether or not to delete the inferior set based on the set registration number P, and Pe is the end of the inferior set deleting process. This is a deletion end determination value used when determining whether or not Ps = 50 and Pe = 30 in the first embodiment.

AVPE is data indicating the correction value to be added to the worst value WVPE of the performance evaluation value when setting the additional reference value CBX. That is, the additional reference value is corrected as a value obtained by adding the correction value AVPE to the worst value WVPE. In general, a value of 0 second or more is set as the correction value. In the first embodiment, AVPE is set to 1 second. GPS1 to GPS4 are a set of initial settings corresponding to normal driving (business one), commuting driving, up-peak driving, and down-peak driving. Each of the initial setting groups GPS1 to GPS4 has the same configuration as the initial setting group GPS0 in FIG.

-Explanation of operation

Next, the operation of the first embodiment will be described with reference to FIGS. FIG. 9 shows a main configuration of a control program of the group management device 1. This control program includes a group management algorithm, and the group management device 1 performs control based on the control program. The group management algorithm itself is publicly known.

In FIG. 9, in step 221 the hall call registration program is executed. Specifically, the hall call generated when the hall button is operated by the passenger is registered in the memory. If one of the elevators handles the call, the call registration will be deleted.

In step 222, the assignment program is executed. Specifically, the allocation evaluation function of the above equation [1] is used, and the allocation evaluation value is calculated for each elevator. Then, the elevator with the smallest evaluation value is assigned to the call. In this step, in addition to the basic assignment calculation by the above evaluation function, [2] ride time priority assignment function, [3] power saving assignment function,

[4] Priority assignment function for proximity elevators, [5] Priority assignment function for light-load elevators, And [6] also includes processing based on the specific elevator priority assignment function.

In step 2 23, the assignment change program is executed. Specifically, it is detected that the service of the hall call assigned as described above has deteriorated, and an assignment for rescue is performed. This step also includes processing based on [8] Allocation change operation for long waiting calls and [10] Allocation change operation for crowded hall calls.

In step 224, the commute operation program is executed. Specifically, [11] selection of driving at work · selection and cancellation of the driving mode according to the cancellation conditions are performed, and if the driving at work is selected, [14] driving operation at work Operation control is performed.

In step 225, an up-peak operation program is executed. Specifically, [12] selection of up-peak operation ♦ The operation mode is selected and canceled according to the cancellation condition, and when up-peak operation is selected, [15] up-peak operation is performed. Operation control is performed.

In step 222, the down peak operation program is executed. Specifically, the operation mode is selected and canceled according to [13] Down-peak operation selection and cancellation conditions described above, and [16] Down-peak operation operation when down-peak operation is selected. in _c step 2 2 7従Tsuta operation control is carried out, the dispersion stand-by operation program is executed. Specifically, when none of the commuting operation, up-peak operation, and down-peak operation are selected, the decentralized standby operation is selected. When the dispersing waiting operation is selected, the _c Step 2 2 8 that is performed is the operation control in accordance with [17] dispersing waiting operation, the power-saving operation program is executed. Specifically, in order to save power while taking into account the operation service status, [18] operation control is performed while increasing or decreasing the number of operating units in accordance with power saving operation.

In the last step 229, the output program is executed. Specifically, the four car control devices 1 A to 1 connected to the group management device 1 [9]

Three The packed passage reference value DB required for the automatic passage function is sent. Each of the car control devices 1A to 1D determines whether or not the car is full based on the car load and the packed passage reference value DB. If the car is full, the car control devices 1A to 1D automatically pass the call generation floor. Here, the packed passage reference value DB is treated as a control parameter as a search target because it has a large effect on group management performance.

The entire group management program (including the control program shown in FIG. 9 and the search command program shown in FIG. 10 shown below) is repeatedly executed periodically (for example, every 100 milliseconds). FIG. 10 shows a search command program included in the group management device 1. This program issues a search instruction to the search device 10.

Ό o

In FIG. 10, step 231, which is executed when an optimal set has already been determined for any traffic flow, takes in the optimal set signal 20a from the search device 10 and The optimal set BPS is stored in the memory of the group management device 1 in association with the traffic TRS. In this case, the search state data SS included in the optimal set signal 20a is also stored.

In steps 23 and 23, the degree of progress of the search is determined based on the search state data SS. If the search state data SS is equal to NEa (= l, 000), the search has been completed, so in step 234, the traffic flow to search for the optimal set is specified. A new search condition signal 1a consisting of the stream specification data TRS, the elevator specification data ELS, and the search command data SCM set to “1” is newly created and output.

On the other hand, if it is determined in step 2 3 3 that the search state data SS is 1 or more, it indicates that the search has already been started, so in step 2 3 5 the search condition signal 1 a Rewrite the value of search command data SCM in the table to “◦” and output a new search condition signal 1a. In order to search for the optimal set for each traffic flow, four types of traffic flow corresponding to normal driving (business person), driving at work, up-peak driving, and down-peak driving are selected in order.

In step 234, the traffic flow specification data TRS was created based on the actual measurement results of the group management device 1.For example, a well-known traffic measurement device was connected to the group management device 1, and the collected traffic conditions The data (number of passengers, number of calls, etc.) may be aggregated, the data may be input to the group management device 1, and the traffic specification data TRS may be created based on the data.

In addition, if the search device 10 employs a mode in which the optimal set is output at any time during the search, the search condition data SS obtained in step 23 You can determine how reliable the optimal set is. For example, if the search state data SS indicates the initial stage of the search, the set used in the past without using the set input from the search device 10 is used as the set to be actually used. Can be used. In this way, it is possible to prevent a decrease in the group management performance of the system. If the search state data SS indicates the middle or late stage of the search, the optimal set output from the search device 10 can be determined to be highly reliable. However, the group management performance can be improved even before the search is completely completed. FIG. 11 shows a search program (main program) stored in the search device 10. This program is stored in ROM10B.

In FIG. 11, in step 25, a re-search determination program having the function of the re-search section 19 in FIG. 1 is executed. In step 26, the search start determination program is executed to determine whether it is time to restart the search for the optimal set. This re-search judgment method is explained using Fig. 12. explain.

In FIG. 12, the search device 1◦ inputs the search condition signal 1a from the group management device 1 in step 261, and converts the elevator specification data ELS, the traffic flow specification data TRS, and the search command data SCM into RAMI. Store to 0 C. Then, in the next step 262, it is detected that the search command data SCM has changed from “0” to “1”, and if it is detected, the search start flag STR is set to “1” in step 265. Set to. On the other hand, if it is determined that the search command data SCM has not changed from “0” to “1”, in the detection step 263, the elevators that have been searched by the elevator overnight specification data ELS are used. It is determined whether or not it is different from the overnight specification data ELSX, and in step 264, it is determined whether or not the traffic flow specification data TRS is different from the traffic flow specification data TRSX searched so far. If ELS is different from ELSX, or if TRS is different from TRSX, in step 265, the search start flag STR is set to “1”. Otherwise, in step 26, the search start flag STR is set to “0”.

It should be noted that the re-determination of the search in Steps 26 3 to 26 65 is performed when the preconditions for the search have changed compared to before, and it is highly probable that the currently registered optimal set is no longer optimal. In addition, it is necessary to search for the optimal set again. For example, if the traffic flow in the building changes due to a change of tenants, or if a part of the group management algorithm is changed to improve functions, a re-search will be performed.

Returning to FIG. 11, in step 27, it is determined based on the processing result in step 26 whether it is necessary to perform the initial setting again. Here, ST R = “0” means that the search is in progress. On the other hand, STR ≠ “0” means that the search is interrupted in the middle and the search is started again from the beginning, or that the search is started again after the search is completed. Therefore, when STR = “0”, the process proceeds to the generation program 29 as it is, but when STR ≠ “0”,

6 one After the initialization program of step 28 is executed and the initialization of various data is performed, the process proceeds to the generation program 29.

Here, the operation of the initial setting program 28 will be described with reference to FIGS.

In FIG. 13, in step 281, an initial data group suitable for a designated traffic flow is selected from a plurality of types of initial data groups stored in advance. Here, each initial data group is composed of an initial set number Pk, Pk initial sets, and Pk group management performance data.

For example, if the normal time zone is specified by the traffic flow specification data TRS, an initial data group suitable for normal operation is selected from the multiple initial data groups GPS1 to GPS4, and the initial data The group is registered as the initial setting data group GPS0 in Fig. 3. Here, the initial setting data group GPS0 in Fig. 3 is composed of the initial set number Pk, a plurality of initial sets IPS () to IPS (Pk), and group management performance data PRI () to PR1 (Pk). ) And.

The initial set number Pk is set to the same value (= 30) as the deletion end judgment value Pe.

Next, in step 282, the initial set number Pk becomes the set number P, and the initial set IPS (l) to IPS (Pk) become the registered sets EPS (1) to EPS (P). Then, the group management performance data PRI (l) to PRI (Pk) are substituted into the group management performance data PRE (1) to PRE (P), respectively. That is, as shown in FIG. 50, the initial setting A1 of the storage unit A2 is executed.

Then, in step 283, the number of evaluations NE is set to 0, the number of additional registrations is set to 0, the evaluation target set number NP is set to 0, the search permission flag FLAG is set to 1, and the crossover probability CR is set to 1.0. , Mutation probability MR is initialized to 0.01, and this program ends. Returning to FIG. 11, in step 29, the production corresponding to the generation unit 12 in FIG. The configuration program is executed. First, in step 30, it is determined whether to continue the search. If the search permission flag FLAG is "0", the process returns to the search start determination program in step 26. On the other hand, if the search permission flag FLAG is "1", the process proceeds to the new set generation program in step 31. And proceed. The new set generation program will be described below with reference to FIG.

In FIG. 14, first, in step 311, it is determined whether or not a new set that has not been evaluated remains. If the evaluation target set number NP is less than the maximum value Nmax, there is a remaining new set that has not been evaluated, and the program 31 is immediately exited in order to evaluate the new set. . On the other hand, when the evaluation target set number NP is equal to or greater than the maximum value Nmax, that is, when the evaluation of all new sets has been completed, the process proceeds to step 312, where the number of generated sets Pn is initialized to 0. Set.

In the next step 3 13, the average waiting time AWT (1) ~ P (P) from the group management performance data PRE (P) ~ PRE (P) for P registration sets EPS (1) ~ EPS (P) The AWT (P) is extracted, and these are substituted into the performance evaluation values VPS (1) to VPS (P) for determining the optimum value. The probability (appearance rate) that each set is selected as the parent set is set based on the reciprocals of these performance evaluation values VPS (1) to VPS (P). O

Then, the processing of steps 314 to 324 described below is repeated to generate a new set up to the maximum value Nmax.

First, in step 314, the number of generated sets Pn is increased by one. In the next step 3 15, a random number with a value between 0 and [CR + MR] (the sum of the crossover rate and the mutation rate) is generated, and the value of the random number is less than CR (= 1. 0). Is a "crossover" if the value is CR, and a "mutation" if the value is CR (-1.0) or more. So, randomly select the generation method. However, the ratio between the crossover rate (probability that crossover is selected) CR and the mutation rate (probability that mutations are selected) MR can be modified according to a predetermined standard.

By the way, if the "crossover" is determined in step 316, the process proceeds to step 317. Here, first, the reciprocal of the performance evaluation value VPS is given as a weighting value for each set. The magnitude of the weight value indicates the magnitude of the probability of being selected. Then, two random numbers are generated within a range where [0] is the lower limit and [the sum of the reciprocals of the performance evaluation values VPS (1) to VPS (P)] is the upper limit. Then, two sets are selected according to the value of the generated random number. Here, suppose that the two parent sets (cross-set pairs) are the first set EPS (PSl) and the second set EPS (PS2) of PS1.

Further, in the next step 318, one random number having a value between 0 and 25 is generated, and a parameter number で specified by the value of the random number is selected. The parameter number is a number that is common to the two parent sets, and specifies the parameter position where numerical value exchange is performed.

Then, in step 319, the 数値 th numerical value in the parent set EPS (PS1) and the PXth numerical value in the parent set EPS (PU2) are exchanged with each other. This creates two new sets. Then, the generated two sets are set as the [Pn] th new set NPS (Pn) and the [Pn + 1] th new set NPS (Pn + l).

Finally, in step 320, the value of the number of generated new sets Pn is increased by one.

If the result is “mutation” in step 3 16, proceed to step 3 21, where the reciprocal of the performance evaluation value VPS is assigned to each set as a weight. I do. The magnitude of the weight value indicates the probability of being selected. And [0] is the lower bound, One random number is generated within the range up to [the performance evaluation value VPS (1) to the sum of the reciprocals of VPS (P)]. Then, one set is selected according to the value of the generated random number.

Further, in step 3 22, as in step 3 18, a random number is generated, and the parameter number PX to be mutated is selected.

Then, in step 32, one random number is generated between the [minimum value] and the [maximum value] that the parameter specified by the number PX can take. Replace the PX-th numerical value of the PS1's excellent set EPS (PSl) with a random value. The generated set is set as the Pn-th new set NPS (Pn).

In the next step 324, it is determined whether the required number of new sets has been created. Here, the generation of a new set ends when Pn + 2> Nmax, considering that two new sets are generated at once due to “crossover”. Until then, the processing of steps 3 14 to 3 24 is repeated to generate up to Nmax new sets. When the generation of the new set is completed, the set number of the new set to be evaluated first, that is, the evaluation target set number NP is initialized to 1 in step 325. “Crossover” is a search method with convergence of the solution, while “mutation” is a search method with a variety of solutions. In other words, if only crossover is used, the direction of the search becomes local and the possibility of losing the optimal solution is increased. . In that sense, they are complementary. There is also the danger that the optimal solution that has been searched for may be destroyed due to sudden mutation. In that sense, they are competing.

Therefore, in this Example 1, the mutation rate MR is much higher than the crossover rate CR in order to minimize the risk of competition while utilizing the complementarity. Is set to a small value.

In this embodiment, after generating 20 (= Nmax) new sets, the group management performance of each new set is evaluated. However, other techniques can be applied.

For example, by setting the maximum value Nmax to “1”, each time “crossover” or “mutation” is performed, the evaluation of the group management performance for the new set is performed, and this may be repeated. it can.

By the way, when the value of the maximum value Nmax becomes large, a new set can be generated at a time, so that the calculation time can be reduced. However, in that case, it is necessary to secure a large capacity of RAMIOC. It is desirable to appropriately determine the value of the maximum value Nmax in consideration of the calculation time and the amount of memory. Returning to FIG. 11, the evaluation program of step 33 will now be described. This evaluation program is equivalent to the evaluation unit 13 in FIG. 1, and gives a new set to the simulation device, executes the group management algorithm, and obtains the execution result. About this evaluation program,

This will be described using FIG. In Fig. 15, first, in step 331, simulation condition data SIM consisting of the evaluation set NPSX, the overnight specification data ELSX, and the traffic flow specification data TRSX is created. In other words, a new set NPS (NP) is set in the evaluation set NPSX, and the elevator specification data ELSX and the traffic specification data TRSX contain the elevators included in the elevator specification signal 1a. Evening data ELS and traffic flow data TRS are set. Then, in the next step 332, the simulation condition data SIM is output to the simulation device 2 as the simulation condition signal 13a, and the virtual group management operation is performed in the simulation device 2. Is performed. And step 3 3

One four At 3, wait for the end of the simulation.

The simulation device 2 performs a simulation in accordance with the simulation condition signal 13 a, and outputs the group management performance signal 2 a to the search device 10 when the simulation ends. In step 33, when the group management performance value signal 2a is received, it is determined that the simulation has been completed, and in step 3334, the group management performance data PRF included in the group management performance value signal 2a is determined. Store it in RAMI 0 C, and proceed to the next step 335.

In step 3 35, the evaluation count NE is incremented by 1 and updated, and the evaluation result data consisting of the evaluation count NE, the evaluation set NPSY (= NPSX), and the group management performance data PRFY (= PRF) Create a RES. Then, in step 336, the value of the evaluation target set number NP for the new set is increased by 1 and updated. Returning to FIG. 11, the additional program in step 34 corresponds to the additional section 15 in FIG. 1, and determines whether or not to register a new set (unit number NP). This additional program will be described with reference to FIG. In FIG. 16, first, in step 341, the average waiting time AWT is extracted from the group management performance data PRFY, and is set as a performance evaluation value VPN for determining additional registration. Then, in step 342, the performance evaluation value VPN is compared with the additional reference value BX to determine whether or not to register the value in the storage unit. If VP N ≥BX, do not allow registration and immediately end this program 34 o

—If the VPN is BX, go to step 3 4 3 where the value of the number of additional registrations NR is increased by 1 and updated, and the number of additional registrations NR and the evaluation set NP SZ (= NPSY) Create additional registration data RAP consisting of, and group management performance data PRFZ (= PRFY). Then, in step 3 4 4, additional registration is performed as the [P + 1] -th set, and at the same time, the value of the number of registered sets P is set to 1 Increase and update. Returning to FIG. 11, the deletion program in step 35 corresponds to the deletion unit 16 in FIG. 1, and deletes a set having a poor performance evaluation value. The deletion program will be described with reference to FIG.

In FIG. 17, first, in step 351, the number P of registered sets is compared with the deletion start determination value Ps to determine whether it is time to delete a set. If P <Ps, it is determined that it is not time to delete the excellent set, and the program immediately exits the program 35. If P ≥ Ps, it is determined that it is time to delete the excellent set, and the following steps 352-259 are repeated to set the number of sets P to the deletion end determination value Pe. Reduce bad sets as much as possible.

In step 35, the average waiting times AWT (1) to AWT (P) are extracted from the group management performance data PRE (1) to PRE (P), respectively, and the performance evaluation values VPD ( 1) Set as VPD (P) respectively. Then, in step 353, initialization is performed to detect the bad set to be deleted. That is, the search counter RC is set to 1, the worst value WVPE of the performance evaluation value is set to 0, and the deletion set number RP is set to 0.

Then, the set having the worst performance evaluation value (set number RP) is specified by repeating the processing of steps 354 to 357. That is, in step 354, every time a set whose performance evaluation value VPD (RC) is worse than the worst value WVPE up to that point is detected, the performance evaluation value VPD (RC) is calculated in step 355. Worst value Set to WVPE. Also, the value of the search power center RC is set to the deletion set number RP. In step 356, the search counter RC is incremented by one, and in step 357, it is determined whether the search for all sets has been completed.

In step 3 58, the set with the worst value WVPE (deletion set number RP) Remove the registration, together, the registration of the group management performance data PRE (RP) also deleted _£ further updating is reduced by one the values of the registered sets the number P. Then, the remaining sets are assigned a set number again from 1 and stored again, and the processing of step 358 is completed.

In step 359, it is determined whether the number P of sets after the deletion is equal to or less than the deletion end determination value Pe. Here, if the following conditions are not satisfied, the processing of the above steps 352 to 358 is repeated. Then, when P≤Pe, the execution of the deletion program ends.

In step 351, in the present embodiment, the deletion start determination value Ps is set to 50 and the deletion end determination value Pe is set to 30. However, the present invention is not limited to this.

The deletion start determination value Ps may be set within a range that does not exceed the maximum number Pmax of sets that can be stored in RAMIOC. In addition, if the deletion start judgment value Ps is set so that Ps = Pe + 1, a method of `` deleting one substitute set for each additional set registered '' is realized. it can. This method is useful when the storage capacity of the RAM 10 C is not enough.

The deletion end determination value Pe in step 359 means the number of remaining parent sets. If the deletion end judgment value Pe is small, it becomes difficult to maintain the diversity of the new set to be generated, and the probability of generating an excellent new set decreases. Conversely, if the deletion end judgment value Pe is large, the diversity of the new set to be generated can be secured, and as a result, the probability of generating a better new set can be increased. However, since the amount of calculation for generation increases, it is not desirable to make the deletion end determination value Pe too large from the viewpoint of efficient search.

Therefore, it is desirable to set the deletion end judgment value Pe by trial and error in some cases according to the number of combinations of the two sets that perform crossover, the type and number of control parameters, and the like. In the first embodiment, Pe = 30. Therefore, 30 X 29 + 2 = 435 combinations are secured. Returning to FIG. 11, the additional reference value correction program in step 36 is equivalent to the additional reference value correction section 18 in FIG. 1, and according to the set registration status of the storage section 11, the additional reference value correction program is added. Modify the value BX. This additional reference value correction program will be described with reference to FIG.

In Fig. 18, first, in step 361, average waiting times AWT (1) to AWT (P) are extracted from the group management performance data PRE (l) to PRE (P), respectively, Substitute the performance evaluation values VPE (1) to VPE (P) for value setting. Next, in step 362, calculation is performed to specify the worst value WVPE among the performance evaluation values VPE (1) to VPE (P) for setting the reference value. This calculation is the same as the processing of steps 353 to 3557 in FIG. Then, in step 365, [worst value WVPE-correction value AVPE] is calculated to obtain a correction value CBX, and in step 365, the correction value CBX is substituted into the additional reference value BX to correct the value. I do.

In this embodiment, the correction value AVPE is fixed at 1 second from the beginning to the end of the search. That is, the additional reference value BX, which indicates the average waiting time, is set so as to decrease by 1 second. Force, and other values can be used.

Correction value When the value of AVPE is increased, the conditions for additional registration gradually become stricter.If the emphasis is on obtaining as many excellent sets as possible within a limited number of evaluations, a very large value Should not be. Conversely, if the correction value AVPE is set to 0 seconds, the possibility that additional sets with similar characteristics with almost no difference in performance will be added will increase. Therefore, it is necessary to determine the value appropriately according to the search conditions. Returning to Fig. 11, the search end determination program of step 37 is This is equivalent to the search end determination unit 17 and determines whether or not the search for the optimal set has been completed. This will be described with reference to FIG. 19. c In FIG. 19, in step 371, it is determined whether or not to end the search based on the evaluation count NE and the search end determination value NEa. If NE <NE a, it is determined that the search has not been performed yet, and in step 372, the search permission flag FLAG is set to “1” to continue the search. If it is determined that NE≥NEa and the search has been sufficiently performed, the search permission flag FLAG is set to "0" in step 37 to terminate the search.

In this embodiment, the search end determination value NEa is set to 1,000 times. The judgment value NEa is not limited to this.

In general, it is difficult to unambiguously determine how many evaluations should be set. This is because the degree of convergence of the search greatly depends on the search conditions such as the type and number of control parameters, the contents of the initial set, the method of generating a new set, and the conditions for additional registration.

In order to obtain a large number of excellent sets exhibiting even slightly superior group management performance, the search end judgment value NEa should be set to a value as large as possible. However, if the cumulative value NE of the number of searches increases to a certain extent, it takes a long time to complete the search, which causes a decrease in search efficiency. Therefore, in order to efficiently obtain a good set, it is necessary to appropriately set the search end judgment value NEa according to the search conditions.

Returning to Fig. 11, the optimal set extraction program in step 38 is equivalent to the extraction unit 20 in Fig. 1, and extracts one optimal set from among the multiple sets. is there. This will be described with reference to FIG. In Fig. 20, first, in step 381, the average waiting times AWT (1) to AWT (P) are extracted from the group management performance data PRE (l) to PRE (P), and these are the optimal values. Substituted into the performance evaluation values VPS (1) to VPS (P) for judgment. So Then, in step 382, the initial settings for detecting the optimal set are made. That is, the search counter RC is set to 1, the best value of the performance evaluation value BVPE is set to 9,999, and the set number BP is set to 0.

Next, the optimum set (set number BP) having the best performance evaluation value is specified by repeating the processing of steps 3883 to 3886. That is, in step 38, the performance evaluation value VPS (RC) is compared with the highest value BVPE checked so far. If a performance evaluation value VPS (RC) that is better than the maximum value BVPE is detected, in step 384, the performance evaluation value VPS (RC) is substituted for the maximum value BVPE, and the set number BP is searched for. Set the value of the counter RC. In step 385, the search counter RC is advanced by one, and in step 386, it is determined whether or not the investigation has been completed for all the sets. Then, in step 388, the optimum set data BPD including the optimum set BPS, the elevator specification data ELSY, the traffic flow specification data TRSY, and the search state data SS is created. In other words, the contents of the set having the highest value BV PE are substituted into the optimal set BPS, and the elevator specification data ELSY and the traffic flow specification data TRSY are replaced with the elevator specification data ELSX in the simulation condition data SIM and Set the same contents as the traffic flow specification data TRSX. In the search state data SS, set the value of the number of evaluations NE up to that time NE

Finally, in step 388, the optimal set signal 20a including the optimal set data BPD is output to the group management device 1. Returning to FIG. 11, as described above, when the search for the optimum set is completed, the process returns to step 26 again, the search permission flag FLAG is reset to “0” by the search end determination program, and the search ends. Steps 26, 27, 30 to 38 are repeatedly executed until the judgment is made. If there is a change in the contents of the elevator specification data or traffic flow specification data during the search, re-search is executed. Is done. That is, the search permission flag FLAG is changed to “1”, and each step from step 31 is executed. Advantage of one embodiment 1

As described above, according to the first embodiment, an excellent set can be efficiently generated, and an optimum set can be efficiently searched. Further, since the group management simulation device 2 separate from the group management device 1 is used, it is possible to search for the optimum set without hindering the original group management operation.

Further, in the first embodiment, since a new set can be generated by two types of generation methods, “crossover” and “mutation”, the characteristics of both can be fully exhibited. In other words, for a new set to be generated, appropriate diversity and appropriate convergence can be achieved at the same time, and the optimal set can be searched early by combining wide area search and local search. It is.

Further, in the first embodiment, the selection of the parent set is performed after weighting the selection probability based on the magnitude of the performance evaluation value with respect to each parent set. The probability of selection can be increased, in other words, the probability of producing a good new set that inherits the good qualities of the parent can be increased.

In the first embodiment, the additional reference value is modified based on the worst value among the plurality of performance evaluation values, and the additional reference value becomes gradually stricter. Unnecessary processing can be avoided.c If the search termination is determined, the parent set selection condition is corrected, and the parameter selection condition is corrected based on the number of additional registrations, appropriate Processing can be realized.

Further, in the first embodiment, the number of registered sets can be suppressed to a fixed value by the deletion processing, so that a reasonable set can be registered in consideration of the storage capacity. As a result, better out of as many sets as possible Can be sorted out.

Also, in the first embodiment, the sets with the lowest performance evaluation values are deleted in order, so that only the sets with the highest performance evaluation values can be left. When generating a new set, the superior set is always used as the parent. It can be a set.

In the first embodiment, the extraction unit searches for the optimum set at each point in time during the search, and outputs the optimum set. Therefore, in the group management device 1, an optimal set can be obtained even during the search without waiting for the search to be completed, and can be used. Furthermore, in the course search can best so obtained to set the appended search state data (number of evaluations NE), for determining the use value of the optimum set Bok the search middle Te odors are output to accurately _c Further, in the first embodiment, the search is continued until the number of searches reaches the predetermined number, so that it is possible to prevent the search from being completed before the search is sufficiently performed.

In addition, since the first embodiment has a re-search function, even when the search is completed, if any one of the elevator specification data or the traffic flow specification data changes, an excellent quality is automatically obtained. Resume the set search. Therefore, even if the search start command from the group management device is delayed for some reason, the search can be started early. Therefore, the optimal set corresponding to the latest group management conditions can be obtained at an early stage. In addition, the search function automatically restarts the search from the beginning under the new group management conditions when the elevator specification data and traffic flow specification data change even if the search is in progress. be able to.

In the first embodiment, an initial set group corresponding to each traffic flow specification is prepared in advance. Then, when starting the search, the initial set group that best matches the traffic flow at that time can be selected and the initial settings can be made. Thus, a set that is somewhat excellent from the beginning can be made the parent set, and a quick search can be performed. During the search, the optimal set is output. When using a set that has been set, some good group management performance can be obtained even at the beginning of the search.

In the first embodiment, the group management performance data PRF obtained by the simulation device 2 is stored in the storage unit 11. Here, the group management performance data PRF is composed of a plurality of data as shown in FIG. Then, the performance evaluation value VPN for additional registration, the performance evaluation value VPE (1) to VPE (P) for setting the reference value, the performance evaluation value VPDG for deletion judgment to VPD (P), and the performance for optimal value judgment Any of the data included in the group management performance data PRF is substituted for the evaluation values VPS (1) to VPS (P). Therefore, it is not necessary to perform simulation every time it becomes necessary to obtain each performance evaluation value. Of course, when each performance evaluation value is composed of common data (for example, average waiting time AWT), only the common data need be recorded as the group management performance data PRF.

[General theory of generation method and selection method]

There are roughly two ways to generate n + 1 generation new individuals (children) from n generations of populations (parents). The first method is to use the newly created individual (child) as a parent to generate the next new individual (child) of the same generation (n + 1 generation). The second method is not used.

That is,

G n (Mn): n-th generation population with Mn population size

G n * (j): A new population consisting of j new individuals generated based on at least the population G n (Mn) with a population size of j

g n (i): The i-th individual of the n generation is g n (i)

n * (j): the j-th new individual of the new population G n; j)

Then, there are the following two methods for generating and adding a new new individual gn: Kj + l) to generate a new individual Gn: j + l). [Generation method A]: A method in which only the current generation population Gn (Mn) is used as a parent, and a new individual gn j + l) is generated as a child by crossover or mutation.

[Generation method B]: Use the current generation population Gn (Mn) and all or part of the new population Gn = j) as a parent, and generate a new individual gntU + l) as a child by crossover or mutation. how to

However,

Gn (Mn) = {gn (l), gn (2),…, gn (Mn)}

Gn * (j) = {gn * (l), gn * (2), "', gn * (j)}

Gn = {gn * (l), gn * (2), ~, n * (j), gn * (j + l)} As a modification of [Generation method B], a new population Gn * (j ), There is a method (hereinafter referred to as [generation method Ba]) that limits only those who have the qualifications appropriate for the parent. On the other hand, the method for selecting the next-generation population can be classified according to whether the current-generation individuals are left as the next-generation individuals or not. That is,

Gn * (Mn *): New population with a population size of Mn

Gn + l (Mn + l): next-generation population with a population size of Mn + 1

When

[Selection method A]: Select only Mn + 1 new individuals _gn + l (i) (…, Mn + 1) from the new population Gn Mn: | Method to use as population Gn + 1 (Mn + 1) (Current generation population Gn (Mn) is not left as a next generation individual at all)

[Selection method B]: Mn + 1 new individuals gn + l (i) (i, based on a predetermined criterion, from the current generation population Gn (Mn) and all or part of the new population Gn Mni = l ", Mn + 1) and use it as the next generation population Gn + 1 (Mn + 1)

However,

Gn * (Mn *) = {gn (1), gn 2),…, gn * (Mn *)} Gn + l (Mn + l) = {gn + l (l). Gn + l (2),..., Gn + l (Mn + l)} As a modification of [Selection method B], Among generation populations Gn (Mn), there is also a method (hereinafter referred to as [selection method B a]) in which individuals that could not become parents of the new population GnWMn :! :) are not left as next-generation individuals. In addition, there is a new population Gn (a method in which a new individual who does not qualify as a parent among Mn is not left as a next-generation individual (hereinafter referred to as [selection method Bb]).

[Example of combination of generation method and selection method]

(A) When implementing the optimal set search method combining the above [Generation method B] and [Selection method B], the following is performed.

That is, the storage unit 11 is divided into two areas: a set group area of the current generation and a set group area to be newly registered. Then, the generation unit 12 generates a new set group using the set group of the current generation and the additionally registered set group. Then, the adding unit 15 selects a new set from among the new set group based on predetermined criteria (including a case where all new sets are selected unconditionally), and additionally registers. On the other hand, every time the number of additional registrations reaches a predetermined value (for example, (Ps—Pe + 1)) by the deletion unit 16, a predetermined criterion is selected from the current generation set group and the additionally registered set group. Use to select a certain number of sets, and set the set group again as the current generation set group. And repeat these.

Therefore, in the case of such a combination, the generation unit 12 and the addition unit 15 have the function of [generation method B], and the addition unit 15 and the deletion unit 16 have the function of [selection method B].

In the first embodiment, [generation method Ba] which is a modification of [generation method B] and [selection method Bb] which is a modification of [selection method B] are used. Additional part of this embodiment 1 15 《Additional registration of only those that have the qualifications suitable as a parent among the new sets, as one of the parents of the next new set

2 one Focusing on the point of use, it can be said that the adding unit 15 has a part of the function of the [generating method B a] together with the generating unit 12. At the same time, paying attention to the fact that the additional part 15 of the first embodiment is registered as a candidate for a next-generation parent set, only those that have a qualification suitable as a parent are added and registered. It can be said that it plays a part of the function of [Sorting method B b] together with 16.

(B) When [Generation method A] and [Selection method A] are combined, do as follows. That is, the storage unit 11 is divided into two areas: an area for the current generation set group and an area for the newly added set group. Then, the generation unit 12 generates a new set group from the excellent set groups of the current generation. Then, the adding unit 15 selects a plurality of sets from a new set group based on a predetermined standard and additionally registers them. On the other hand, every time the number of additional registrations reaches a predetermined value (for example, Pe), the deletion unit 16 deletes all sets of the current generation and moves the additionally registered sets there. Update the generation and repeat.

Therefore, in the case of such a combination, the generation unit 12 of the first embodiment has the function of [generation method A], and the addition unit 15 and the deletion unit 16 have the function of [selection method A].

(C) When combining [Generation Method A] and [Selection Method B (or B b)], do as follows.

That is, the storage unit 11 is divided into two areas, one for the current generation set group and the other for the newly added set group, and the generation unit 12 creates a new set from the current generation set group. Generate a group of groups. Then, the adding unit 15 selects a new set group from the new set group based on a predetermined standard and additionally registers the new set group. On the other hand, the deletion unit 16 sets the number of additional registrations to a predetermined value (for example, (Ps -Pe +1)), a set group is selected based on a predetermined criterion from the current generation excellent set group and the additionally registered excellent set group, and the set group is assigned to the current generation excellent set. Set as a group. And repeat these.

Therefore, in the case of such a combination, the generation unit 12 of the first embodiment is provided with the function of [generation method A], and the addition unit 15 and the deletion unit 16 are provided with the selection method B (or Bb). ] Function.

(D) When combining [Generation method B (especially B a)] and [Selection method A], do as follows.

That is, the storage unit 11 is divided into two areas: an area for the current generation set group and an area for the newly added set group. Then, the generation unit 12 generates a new set group from the current generation excellent set group and the additionally registered excellent set group. The adding unit 15 selects a new excellent set group from the new set group based on a predetermined criterion, and performs additional registration. On the other hand, every time the number of additional registrations reaches a predetermined value (for example, Pe), the deletion unit 16 deletes all the excellent set groups of the current generation and moves the additionally registered set groups there. Let me do it. And repeat these.

Therefore, in the case of such a combination, the generation unit 12 and the addition unit 15 of the first embodiment have the function of [generation method B (particularly, B a)], and the addition unit 15 and the deletion unit 16 have Have the function of the sorting method A].

[Performance evaluation value]

For reference, for each performance evaluation value used in the above examples, the means of use and the purpose of use are summarized below.

[19] Performance evaluation value for determining the optimal set: VPS (1) to VPS (P)

• Use: Extraction unit 20

• Intended use: Used to select the optimal set [20] Performance evaluation value for determining additional registration: VPN

• How to use: Additional part 1 5

• Purpose of use: Used to determine whether to add a new set

[21] Performance evaluation value for deletion judgment: VPD (1) to VPD (P)

• Usage: Deleted part 1 6

• Intended use: Used to determine the deletion of a registered set

[22] Performance evaluation values for setting additional reference values: VPE (1) to VPE (P)

• Usage: Reference value correction unit 1 8

• Intended use: Additional reference value Used as a reference when modifying BX

[23] Performance evaluation value for setting parent selection probability: VPS (1) to VPS (P)

• Usage: Generation unit 1 2

• Intended use: Used to select a parent set. Here, the superiority of each set can be judged from various angles. C For example, how much the control target required for the group control device can be satisfied Can be determined from the viewpoint of In this case, the group management performance value (see Fig. 7) directly related to the requested control target can be used as it is as the performance evaluation value.

Regarding crossover, the superiority of each set as a parent can be judged from the viewpoint of the diversity in the storage unit. This is because crossing sets with different characteristics increases the probability that a better set will be generated.

A “distribution index” can be used as a performance evaluation value indicating diversity. The distribution index can be defined as, for example, the number of other sets whose distance between sets is within a predetermined value, with each set being the center. Here, the distance between sets is defined in a multidimensional space defined by a plurality of set components. The distribution index indicates the similarity between sets, and the The lower the value, the higher the parent's performance.

The distribution index can be defined as the sum of the distances from other sets. (In this case, the larger the value, the higher the performance as a parent. In this case, the larger the value, the higher the performance as a parent.

(Hereinafter, margin)

6 one Next, a method of obtaining the performance evaluation value will be described in detail.

In the first embodiment, all the performance evaluation values are configured by the average waiting time AWT. However, the content of each performance evaluation value can be different depending on the purpose of use. For example, each performance evaluation value E may be calculated using a different performance evaluation function.

Generally, the performance evaluation function related to group management performance is expressed as follows. Here, F (X) indicates that it is a function of X.

E = F (XI, X2, ~, Xi, ~, Xn.Tl, T2, —, Ti,, Tn)… [24]

Where, n, 'number of evaluation items for group management performance

Xi: Performance value for evaluation item i (i = i, 2 ^ ', n)

Ti _: Performance standard value of evaluation item i (i = 1.2,…,! 1)

Here, the performance reference value Ti may represent a “target value” that should ultimately be reached as group management performance, or a “limit value” that must be satisfied at a minimum. Further, the “limit value” includes an “upper limit value” and a “lower limit value”. Whether the performance reference value is given as a "target value", "upper limit value" or "lower limit value" depends on where the purpose of group management control is to be set.o

In the performance evaluation function [24], when the performance reference value means “target value”, the smaller the value of I Xi —Ti l, the better the performance. When the performance reference value means “upper limit value”, the larger the (Ti-Xi), the better the performance. When the performance reference value means “lower limit value”, the larger the (Xi—Ti), the better the performance. Noh is good.

In short, the above [19 :! The performance evaluation value shown in [23] is determined by the performance evaluation function, the evaluation item and the performance reference value in the function.

Hereinafter, the calculation method of the performance evaluation value will be described with some specific examples.

[25] Example 1 of performance evaluation function (for the first embodiment) Control purpose "Make the average waiting time as small as possible" Performance reference value T1-0 seconds (T1: "Target value" of average waiting time) Performance evaluation function E = I AVT -T1 I = AWT (AWT: Average waiting time) Determination of additional registration E <BX (BX: additional reference value, eg, BX-15 seconds)

[26] Example 2 of performance evaluation function (in case of Example 2 described later)

Control purpose "Make average waiting time as close as possible to the specified value" Performance reference value Tl = 20 seconds (T1: "Target value" of average waiting time) Performance evaluation function Ε = I AWT -T1 I (AVT: Average waiting time) )

Determination of additional registration E <BX (BX: additional reference value, for example, ΒΧ = 3 seconds)

[27] Performance evaluation function example 3

• Control objectives: "Average waiting time AWT, long waiting rate RLW, and missed forecast rate

Keep the RPE as close as possible to their respective targets. ''

• Performance standard value: Tl = 15 seconds, Τ2 = 2%, Τ3 = 3%.

(However, T1: “Target value” for average waiting time, Τ2: “Target value” for long waiting rate, Τ3: “Target value” for missed forecast rate)

• Performance evaluation function:

E = Apx I AVT -Tl I + Aqx I RLV -T2 I + Arx I RP E -T3 I (However, Ap, Aq, and Ar are weighting factors) • Judgment of additional registration: E <BY (BY: Overall (Evaluation reference value, for example, BY = 10)

[28] Performance evaluation function example 4

Control purpose: "AWT, AWT, RLW, and missed forecast rate

The RPE should be as small as possible as a whole. ”Performance standard values: Tl = 0 seconds, Τ2 = 0%, Τ3 = 0% ο

(However, Tl: "target value" of average waiting time, T2: "target value" of long wait rate, T3: "target value" of missed forecast rate) Performance evaluation function: E = Apx AVT + AqxRLW + ArxRPE (However, Ap, Aq, and Ar are weighting factors)

• Judgment of additional registration: E <BY (BY: Overall evaluation standard value. For example, BY = 10

00)

[29] Example 5 of performance evaluation function

RPE should be as many as possible within their respective tolerances. "

• Performance standard values: Tla = 15 seconds, Tib = 3 seconds, T2 = 2%, Τ3 = 3%.

(However, Tla: “target value” of average waiting time, Tib: allowable range of average waiting time deviation, T2: “upper limit value” of long waiting ratio, T3: “upper limit value” of missed forecast ratio)

• Performance evaluation function: E = f (I AWT one Tla I -Tib) + f (RL -T2)

+ f (RPE-T3) (however, f (X) represents a function that becomes f (X) = 1 when X ≥ 0 and f (X) = 0 when X <0) • Judgment of additional registration : E <BY (BY: Overall evaluation standard value, for example, BY = 1)

[30] Performance evaluation function example 6

Control purpose "Average waiting time AWT is the best value (minimum value)

Make sure that the long wait rate RLW is as small as possible within the specified tolerance from BVPE. ”Performance reference values Tla = BVPE (seconds), Tib = 2 seconds, T2 = 0%.

(However, Tla: “target value” of average waiting time, Tib: allowable range of average waiting time deviation, T2: “target value” of long waiting ratio)

Performance evaluation function E = (100-RLW) f (Tib- I AWT -BVPE I)

(However, f (X) represents a function that becomes f (X) = 1 when X ≥ 0 and f (X) = 0 when X 0 0) • Judgment of the optimal value: Max {E} In the above [25] performance evaluation function example 1 and [26] performance evaluation function example 2, the evaluation item is one for the average waiting time, and the control purpose is also simple. Therefore, the evaluation function can be configured relatively easily. Therefore, it is very easy to use the evaluation items as shown in Fig. 7 (and of course, evaluation items other than those shown in Fig. 7) instead of the average waiting time.

On the other hand, if there are multiple evaluation items, the control objectives will also be diversified, and the configuration of the performance evaluation function will be complicated. In the case of [27] Example 3 of performance evaluation function and [28] Example 4 of performance evaluation function, the performance evaluation function was composed of the weighted sum of the deviation between the evaluation item and the target value. As described above, a method of comprehensively evaluating the degree of achievement of a target by taking into account the priority of each evaluation item is generally used. This is a very convenient method, especially when evaluation items with conflicting control goals are included.

Also, as shown in [29] Example 5 of performance evaluation function, for each evaluation item, within the set allowable range (for example, the deviation from the upper limit value or lower limit value or the target value The group management performance can be evaluated based on the number of evaluation items whose group management performance value falls within the range (eg, within the value, etc.).

In addition, evaluation values can be calculated by two or more different performance evaluation functions, and addition registration judgment, deletion judgment, optimal set judgment, and the like can be performed based on a combination of the plurality of evaluation values.

Also, as shown in [30] Example 6 of the performance evaluation function, the average waiting time must be within the predetermined allowable range from the minimum value (E l = I AWT-BVPE I, E l ≤Tlb), and the long waiting time If the ratio is the smallest among them (E 2 = RLW, Min {E 2}), first select the optimal value candidate using the performance evaluation function E i Then, the optimal value can be finally selected by the performance evaluation function E 2. Also, as shown in [30] Example 6 of performance evaluation function, The two performance evaluation functions and judgment conditions are combined and written into one performance evaluation function and judgment condition, and the optimal set can be selected based on the performance evaluation function and judgment conditions.

In the first embodiment, the selection probability of the parent set was set based on the performance evaluation value. However, the present invention is not limited to this, and the selection probabilities can be equalized. In that case, a new set having similar characteristics is likely to be generated, so that the diversity of the sets in the storage unit 11 may be lost. Loss of diversity causes problems converging to a local solution from the beginning of the search. Therefore, if there is a need to avoid the problem of initial convergence, or if there is a need to reduce the amount of computation, the selection probability of the parent set can be set evenly.

[Example 2]

Next, a second embodiment will be described with reference to FIGS. In the description of the second embodiment, a description will be given mainly of portions different from the first embodiment. FIG. 21 is a diagram corresponding to FIG. 1 and shows the overall configuration of the second embodiment. In FIG. 21, the performance reference value setting device 3 is constituted by a personal computer, and outputs a reference value signal 3 a to the group management device 1. Here, the reference value signal 3a includes a “performance reference value” for group management performance and a “control reference value” for controlling the search device. In this embodiment, the performance reference value is a “target value” of the average waiting time, and the control reference value is a “designated value” given to the additional reference value BX. Note that the reference value signal 3a may be directly input to the search device 10 (evaluation means 13, reference value updating means 18, re-searching means 19, initial setting means 21).

FIG. 22 corresponds to FIG. 3 of the first embodiment, and FIG. 22 shows the stored contents of the RAMI 0C. In Fig. 22, TGT is one of the data that constitutes the search condition signal 1a, and the data that represents the "target value" of the average waiting time AWT. Evening TAW (waiting time target value) and data TCB (additional reference specified value) indicating the value specified for the additional reference value BX. For example, the waiting time target value TAW is set to 5 seconds, and the additional reference specified value TCB is set to 3 seconds. TAWX is data in which the input waiting time target value TAW is transcribed, and is used to calculate the performance evaluation value.

When the reference value signal 3a is input, the group control device 1 sets the performance reference value (waiting time target value TAW) and the control reference value (additional reference specification value TCB) included in the reference value signal 3a. And take out. Then, similarly to the operation of step 2 34 in FIG. 10, the elevator specification data ELS, the traffic flow specification data TRS, the search command data SCM, the waiting time target value TAW, the additional reference specification value TCB, and The search condition signal 1 a composed of the search condition signal 1 a is output to the search device 10. When the search is started in the search device 10 (see FIG. 11), generation, evaluation, addition, deletion, and addition reference value correction are performed in order, as in the first embodiment. When the search condition signal 1a is input based on the search start determination program 26 (see FIG. 25), as shown in FIG. 22, the elevator specification data ELS, traffic flow specification data The TRS, search command data SCM, waiting time target value TAW, and additional reference specified value TCB are stored in RAMI0C.

The second embodiment is characterized in the contents of the addition program 34 and the deletion program 35.

In the second embodiment, the performance evaluation value E is calculated based on the performance evaluation function (E = I AWT -Tl I) shown in [26] Example 2 of the performance evaluation function described above, and the performance evaluation value E and The additional reference value BX and are compared. Then, if the performance evaluation value E is smaller than the additional reference value BX, additional registration is determined. That is, the difference between the actual value and the target value is compared with the specified value, and if the difference is smaller than the specified value, an additional registration of a new set is performed. Therefore, in step 34 1 of the additional program 34 (see Fig. 16), the operation [VPN — I AWT-TAVX I] is performed, and the performance evaluation value VPN for determining additional registration is calculated. Is set.

In addition, in step 3 52 of the deletion program 35 (see FIG. 17), the operation EVPD (P) *-IAWT (P) -TAWX I] is performed, and the performance evaluation value VPD (1 ) To VPD (P) are set.

FIG. 23 shows the contents of the additional reference value correction program 36 in the second embodiment.

In FIG. 23, first, in step 401, the additional reference specified value TCB is read out, and is substituted for the corrected value CBX. Then, in the next step 402, the additional reference value BX is rewritten with the correction value CBX. In step 403, the waiting time target value TAW is read, and the performance standard 基 (waiting time target value TAWX) currently used is rewritten by this value.

When the correction of the additional reference value is completed in this manner, the search end determination and the optimal set extraction processing are performed as in the first embodiment (see FIG. 11). However, in the optimum set extraction program 38, the method of calculating the performance evaluation value is different from that of the first embodiment. That is, in step 381, of the optimal set extraction program 38 (see FIG. 20), the performance evaluation values VPS (1) to VPS (P) for determining the optimal set are expressed as [VPS (l) —I AWT (l)-TAWX I,…, VPS (P) —I AWT (P)-TAVX I 3

After the processing of the optimum set extraction program 38 described above, the process proceeds to the processing of the re-search determination program 25. The operation procedure of the search start determination program 26 in the re-search determination program 25 will be described with reference to FIG. FIG. 24 corresponds to FIG. 12 of the first embodiment.

In FIG. 24, in step 261, the search condition signal 1a is input from the group management device 1, and the elevator specification data ELS, the traffic flow specification data TRS, the search command data SCM, the waiting time target value TAW, the additional reference The specified value TCB is stored in RAM10C. Then, in Steps 26 2 to 26 4, as in the search start determination program 26 of Embodiment 1 (see FIG. 12), the search command data A change in SCM from “0” to “1”, a change in the elevator specification ELS, or a change in the traffic flow specification TRS is detected. In step 2 65, if at least one change occurs, the search start flag STR is set to “1”, and the search is started after the initialization of mode 1 described later is performed. That

I B -p O

If none of the search command data SCM, the elevator specification data ELS, and the traffic flow specification data TRS have changed, it is determined in step 267 whether the target waiting time value TAW has changed. That is, the waiting time target value TAW is compared with the currently set waiting time target value TAWX to determine whether there is any change. If TAW ≠ TAWX, it is determined that “change has occurred”, and in step 265, the search start flag STR is set to “1”. On the other hand, if TAW = TA WX, it is determined that there is no change, and step 269 is executed.

In step 269, it is determined whether the additional reference specified value TCB has changed. That is, the additional reference specified value TCB is compared with the currently set additional reference value BX to determine whether there is any change. If BX = TCB, it is determined that there is no change, and in step 26 6 the search start flag STR is set to 0, and if the optimal set search is in progress, the continuation state is maintained. If the search is in the end state, the end state is maintained. On the other hand, if BX ≠ TCB, it is determined that “change has occurred”, and in step 268, the search start flag STR is set to “2”, and a command is issued to start the search after initializing the mode 2. I do.

Here, the initial setting of mode 1 refers to a process of performing an initial setting in the storage unit and a general initial setting (setting of initial values such as the number of evaluations NE and the number of additional registrations NR). In addition, the initial setting of the mode 2 is a process of performing only the general initial setting without performing the initial setting to the storage unit.

In step 269 above, the initial setting of mode 2 when the additional reference value BX is changed is that the additional reference value BX is more strict than before. Even if the value is changed to (small), there is no problem if the search proceeds on an extension of the previous search. Rather, from the point of view of convergence, it is better to restart the search using the previously generated and sorted set as the initial set. In particular, when the performance evaluation function includes only a single evaluation item and only the additional reference value BX is changed, the initial setting of mode 2 is appropriate. As described above, as a result of determining whether the search for the optimal set is to be performed in the current state (continuous state or terminated state) or whether to start over from the beginning, the search start flag STR = Suppose "0" is set. In this case, in FIG. 11, the process proceeds from step 27 to step 30, where it is determined whether the search is in progress or the search is completed based on the value of the search permission flag FLAG. If the search is in progress, FLAG = “1”, so the process proceeds to step 31 again, the search end flag FLAG is reset to “0” by the search end determination program 37, and the search end is determined in step 27. Repeat steps 3 1 to 3 8 and steps 26 and 27 until the

When the search is completed, the search permission flag FLAG is set to “1” by the search end determination program 37, so that the procedure of steps 26 → 2 7 → 30 → 26 is repeated this time. Wait for re-search.

If, during or after the search, the Elevator overnight specification data ELS, traffic flow specification data TRS, performance reference value TAW, and control reference value TCB are changed, the search is started by the search start determination program 26. The start flag STR is set to "1" or "2", and the process proceeds from step 27 to the initialization program 28. FIG. 25 shows an initialization program 28. FIG. 25 corresponds to FIG. 13 of the first embodiment. In FIG. 25, in step 284, it is determined whether or not the initial setting of the storage unit is necessary according to the value of the search start flag STR. If the initial setting of the mode 1 is specified, that is, if the search start flag STR is “1”, as in the initial setting program 28 of the first embodiment (see FIG. 13), step 28 1 is executed. Read the initial set group and group management performance data corresponding to the traffic flow specification data TRS. Then, in step 282, initial setting is performed using the initial set group and the group management performance data.

Then, in step 285, the additional reference value correction program is started. The additional reference value correction program in step 285 has the same contents as the program 36 shown in FIG. In this step 285, a new additional reference value (BX) and a performance reference value (wait time target value TAWX) are set. Then, similar to the initial setting program 28 of the first embodiment (see FIG. 13), general initial settings are performed in step 283. The above is the initial setting of mode 1.

On the other hand, in step 284, when the search start flag STR is "2", that is, if the initial setting of mode 2 is specified, only the general initial setting of step 283 is executed. . That is, the processes of steps 281 and 282 are not performed, and the set group registered in the storage unit before the start of the current search is used as the initial set group. That is, at the end of the previous search, the excellent sets EPS (1) to EPS (P) remaining in the storage unit 11 and the group management performance data PRE (1) to PRE (P) are Used. The above is the initial setting of mode 2. As described above, in the second embodiment, the target value TAWX of the average waiting time and the designated value BX of the additional reference value can be externally input by the performance reference value setting device 3. Therefore, the optimal set that matches the desired group management control policy and search policy can be searched. Also, in the second embodiment, a change in the reference value TGT (waiting time target value TAW, additional reference value BX) can be detected during the search or after the search is completed, and the search can be executed again. Therefore, when the control policy of group management is changed artificially or the reference value TGT is changed artificially, re-search can be executed automatically and the optimal set can be obtained quickly. be able to.

In the second embodiment, since the initial setting of the mode 1 and the initial setting of the mode 2 can be selected, appropriate initial settings can be made according to the situation at the start of the search. For example, if the additional reference value BX is fine-tuned, the re-search can be terminated early by the initial setting of mode 2. Of course, if the change amount of the additional reference value BX is large, it is desirable to apply the initial setting of mode 1. Conversely, in the second embodiment, when a change in the performance evaluation function (evaluation item, performance reference value, configuration) is detected, the initial setting of mode 1 is applied, but the amount of change is small. In this case, the default setting of mode 2 can be applied.

In short, the question of mode selection, whether to apply the default settings of Mode 1 or the default settings of Mode 2, depends on the set of sets obtained up to that point under the new search conditions and the initial settings. The question is which of the GPS0 sets can be used for efficient search. However, that judgment requires a great deal of computation time and computational complexity, and is not a practical method. Therefore, as described above, it is desirable to select a mode according to the change item and the change amount.

The above-described performance evaluation function is an example, and [24] performance evaluation function and performance evaluation function examples 1 to 6 ([25:] to [30]) can also be used.

In addition, the performance evaluation values described above are not limited to the performance evaluation values for the additional registration determination described above. [19] Performance evaluation values for the optimal set determination, [21] performance evaluation values for the deletion determination, [22] ] It may be a performance evaluation value or the like for determining a reference value. [Embodiment 3 (Other embodiments of the deletion unit)]

Next, another embodiment of the deletion unit 16 will be described with reference to FIG. The following description focuses on the differences from the first embodiment.

FIG. 26 shows a deletion program 35 of the third embodiment. This flowchart corresponds to FIG. 16 of the first embodiment.

First, in step 411, the equation NRH = NR-NRX is executed to calculate the number of new registrations NRH. The number of new registrations, NRH, is the number of times a new set has been newly registered after the previous deletion process. The number of registrations NRX at the time of the previous determination is the value indicated by the number of additional registrations NR when the previous deletion processing was executed.

Next, in step 412, the number of new registrations NRH is compared with the deletion start determination value NRa to determine whether it is time to delete. For example, the deletion start judgment value NRa is set to 10 times.

If NRH is less than NRa, it is determined that it is not time to delete the set, and the program immediately exits the program 35. On the other hand, if NRH ≥NRa, it is determined that it is time to delete, and in step 413, the value of the number of additional registrations NR at the present time is set to the number of registrations NRX at the previous determination, and updating is performed.

Then, steps 414 to 422 are repeated, and the set is deleted until the unit number P reaches the deletion end determination value Pe. Note that the initial value of NRX is initially set to 0 in step 283 of the initial setting program 28 (see FIG. 13) (not shown).

In step 414, the distance DST (i.j) between the sets is calculated (however, i.j = 1,2, —, P, i ≠ j) o

This distance is DST (i.j), the norm for two sets i and j as follows.

DST (i, j) = II EPU (i) -EPU (j) II… [31]

Where i, j = 1.2 ,, ', Ρ, i ≠ j Here, the parameter values of each set EPS (i) (i = 1,2 ", 2) are normalized for each parameter. That is, each parameter value is It is expressed as a percentage of the maximum value that each parameter can take, and can be any value between 0 and 100.

For example, assuming that the maximum value of the packed evaluation coefficient Ca, which is one of the parameters, can be 50,000, the normalized value of the packed evaluation coefficient Ca parameter value for the set EPS (l) in Fig. 8 Is calculated as (10,000 + 50,000) X 100-20. Similarly, when the parameter value of the forecast deviation coefficient Cb (maximum value: 1,600) is normalized, it is calculated as (400 + 1,600) x 100 "25.

Therefore, there are a total of 25 control parameters, so that 0≤DST (i,; j) ≤500. The normalized parameter values are converted back to the original values when creating the optimal set data BPD in step 387 of the optimal set extraction program 38 (see Fig. 20) (for example, In the case of a packed evaluation factor Ca, it is re-converted to 20 X 50,000 = 10,000).

Next, in step 415, a set pair Pdl, Pd2 that minimizes the distance DST (i,; j) is selected. Then, in step 416, it is determined whether or not the two sets Pd 1 and Pd2 have similar characteristics. That is, the distance DST (Pdl, Pd2) is compared with the determination value DSTa, and similarity is determined from the result. As the determination value DSTA, if _c example 25 is set, DST (Pdl, Pd2) if ≤ DSTA, the process proceeds to step 417, where the two sets Pdl, Pd2 group management performance data PRE of ( The average waiting time AWT (Pdl) and AWT (Pd2) are extracted from Pdl) and PRE (Pd2), respectively, and these are set as performance evaluation values VPD1 and VPD2 for deletion.

In step 418, the performance evaluation values VPD1 and VPD2 are compared to determine a set to be deleted. If VPDKVPD2, it is determined that the set number to be deleted is set number Pd2, and the process proceeds to step 419. Then, the registration of the set EPS (Pd2) and the group management performance data PRE (Pd2) is deleted. In addition, registered The value of the unit number P is also reduced by one. The set number is re-assigned to the remaining set, and the processing in step 4 19 ends.

On the other hand, in step 418, if VPD1≥VPD2, it is determined that the deletion target is set number Pdl, and the flow proceeds to step 420. Then, the registration of the set EPS (Pdl) and the group management performance data PRE (Pdl) is deleted. <In addition, the value of the registered unit number P is also reduced by one. Then, the set number is re-assigned to the remaining set, and the processing in step 420 ends.

If DST (Pdl, Pd2)> DSTa in step 4 16, it is determined that the characteristics are not similar, and the process proceeds to step 4 21, where all unit numbers 1 to P The performance evaluation values VPD (1) to VPD (P) are determined for the deletion judgment, and the set with the worst value is specified from these, and the set number is set as Pdl. Note that this step 421 is the same as steps 352 to 357 in the deletion program 35 of the first embodiment (see FIG. 17), and a description thereof will be omitted. The deletion set number RP in Fig. 17 corresponds to the set number Pdl.

In step 4 22, it is determined whether or not the number P of sets after deletion has become equal to or less than the deletion end determination value Pe. If not, the processing in steps 4 14 to 4 2 2 described above is repeated, and When ≤ Pe, the processing of the deletion program 35 ends. As described above, in the third embodiment, a pair having similar characteristics is specified based on the distance DST between sets, and one of the pairs is deleted. A plurality of sets different from each other can be left in the storage unit 11, and diversity in the storage unit 11 can be secured. Furthermore, when selecting pairs, priority is given to the pair with the highest similarity (closest distance), so that multiple sets with different characteristics can be left as much as possible. Monkey

Further, in the third embodiment, when one of the pairs is deleted, the set with the better performance evaluation value is left and the set with the worse performance evaluation value is deleted, so that a plurality of sets in the storage unit 11 are deleted. The group management performance can be maintained at a high level as a whole.

In the third embodiment, when there are no more similar pairs to be deleted, the sets with the worst performance evaluation values are deleted in order. Thus, unnecessary sets can always be deleted up to the deletion end judgment value Pe.

[Example 4 (Another example of search end determination)]

Hereinafter, another embodiment of the search end determination unit will be described with reference to FIG. The description of the fourth embodiment focuses on the differences from the first embodiment.

In step 371 of FIG. 19, the number of additional registrations NR is used instead of the number of evaluations NE, and the search end determination value NRb is used instead of the search end determination value NEa. The number of additional registrations NR is set to, for example, 200 times.

That is, when NR is less than NRb, the search permission flag FLAG is set to "1" in step 372 to continue the search, and when NR≥NRb, the search permission flag FLAG is set to "0" in step 3732. To end the search.

Here, in general, when the number of searches NE becomes a large value and the search converges to some extent, the ratio of the number of additional registrations to the number of searches decreases. Therefore, in order to perform a more efficient search, it is necessary to determine the end of the search in consideration of the result of the search process, that is, the number of additionally registered sets. For this reason, in the fourth embodiment, the search is continued until the number of additional registrations NR indicating the number of additional registrations reaches the search end determination value NRb.

One seven As described above, according to the fourth embodiment, it is possible to prevent the search from being completed before the search is sufficiently performed.

[Example 5 (Another example of search end determination)]

Next, another embodiment of the search end determination will be described with reference to FIG. In the description of the fifth embodiment, a description will be given focusing on portions different from the first embodiment. FIG. 27 shows a search end determination program 37 of the fifth embodiment, and corresponds to FIG. 19 of the first embodiment.

In step 431, the elapsed evaluation frequency NEH is calculated by the expression NEH = NE-NEX. Here, the number of elapsed evaluations NEH is the number of new evaluations since the previous end determination was made. The number of evaluations NEX at the previous judgment indicates the value of the number of evaluations NE at the time of the previous end judgment. Then, in step 432, it is determined whether or not the evaluation has been performed at least a predetermined number of times since the last time. If the elapsed evaluation frequency NEH is less than the predetermined value NEb, in step 43, the search permission flag FLAG is set to “1” to continue the search. The number of progress evaluations NEH is set to, for example, 20 times. If the elapsed evaluation frequency NEH is equal to or greater than the predetermined value NEb in step 432, the process proceeds to step 434, where the calculation of the success index RSC is performed. First, the number of new registrations NRH is calculated by the formula NRH = NR-NRX. Here, the number of new registrations NRH indicates the number of times a new set has been additionally registered since the previous end determination was made. Then, the success index RSC is calculated by the following equation: RSC = NRH ÷ NEH. Here, the number of registrations NRX at the previous determination represents the value of the number of additional registrations NR at the time of performing the previous end determination. In the next step 435, the evaluation count NEX at the previous judgment and the registration count NRX at the previous judgment are updated based on the current evaluation count NE and the additional registration count NR. In step 436, it is determined whether to end the search based on the number of evaluations NE, the search end judgment value NEc, the success index RSC, and the search end judgment value RSCa. You. Here, the search late decision value NEC, also _c for example 600 times is set as the search termination determination value RSCA, for example, 0. 05 is set.

If it is NE, NEc or RSC RSCa, it is determined that the search has not been performed yet, and in step 43, the search permission flag FLAG is set to “1” to continue the search. If NE≥NEc and RSC and RSCa, it is determined that the search has been performed sufficiently, and in step 434, the search permission flag FLAG is set to “0” to end the search.

In step 436, the condition regarding the number of evaluations NE was included in the initial stage of the search, because the success index RSC was reduced due to the initial set GPS0, crossover rate CR, and mutation rate MR. This is to prevent the search from ending without a sufficient number of evaluations being performed. If such a problem does not exist, the condition for the number of evaluations NE is not required for the search termination judgment condition, and the condition for the success indicator RSC is sufficient.

As described above, in the fifth embodiment, the end of the search is determined based on the success index RSC based on the number of evaluations and the number of additional registrations. Therefore, it is possible to accurately determine whether the search has sufficiently converged. Therefore, there is no need to repeat the search needlessly, and the optimum set can be efficiently searched.

Further, in the fifth embodiment, it is detected that the search is in the initial period based on the number of evaluations NE.In this initial period, the search is not terminated even if the success index RSC becomes less than the search end determination value RSCa. As a result, there is no possibility that the search will be terminated without performing a sufficient number of evaluations.

[Example 6 (Another example of search end determination)]

Another embodiment of the search end determination unit 17 will be described with reference to FIG. In the description of the sixth embodiment, the description focuses on the differences from the first embodiment.

FIG. 28 shows a search end determination program 37 of the sixth embodiment. This corresponds to Figure 19.

In step 451, the distance DST (i,; j) between each set is calculated. This distance DST (i,; j) is calculated according to the above equation [31]. This calculation is the same as step 4 14 of the deletion program 35 (see FIG. 26) of the third embodiment.

In step 45, based on the distance DST (i, j) obtained as described above, the number of similar sets, that is, the number of sets in which DST (i, j) ≤ DSTa, NDST (similar set) Count). Note that DSTa is a determination value for determining whether or not sets are similar, and is set to 25 in the sixth embodiment as in the third embodiment.

In the next step 453, the search end judgment value NDSTa is calculated. In general, as the search converges, many sets tend to concentrate around the optimal set. As one of the indicators to grasp this tendency, the number of similar sets NDST described above is used, and the convergence of the search is determined based on the ratio of the number of similar sets NDST to the number of combinations of all sets. Assuming that the judgment criterion for that ratio is 80%, when the number of registered sets is P, the number of all combinations is {PX (P-l) ÷ 2}, so the search end judgment value NDSTa is NDST a = {P x (P-l) ÷ 2} X 0.8

Here, when the number of additionally registered sets reaches the maximum value Pmax, unnecessary sets are deleted according to a predetermined criterion.However, at this time, the number of similar sets NDST is Change. Similarly, the number of similar units NDST changes depending on the set generation method and the additional registration method. Therefore, the above criterion is not limited to 80% and needs to be modified as appropriate based on other factors.

Φ o

Thus, in step 454, it is determined whether or not to end the search based on the number of similar units NDST and the search end determination value NDSTa. If NDST is close to NDSTa, it is determined that the search has not been performed yet, and In step 4 5 5, the search permission flag FLAG is set to “1” to continue the search, and the processing of the search end determination program 37 ends. If it is determined that NDST ≥ NDSTa and the search has been sufficiently performed, the search permission flag FLAG is set to “0” to end the search in step 456, and the search end determination program 3 The process of 7 ends.

As described above, in the sixth embodiment, since the end of the search is determined based on the inter-set distance DST, the convergence of the search can be accurately detected. Therefore, there is no need to repeat the search needlessly, and an efficient search can be performed. Note that the search termination determination condition is not limited to the above, and other conditions may be employed.

[Example 7 (Another example of optimal set extraction)]

Next, another embodiment of the extraction unit 20 will be described with reference to FIG. In the description of the seventh embodiment, the description focuses on the differences from the first embodiment.

FIG. 29 shows the optimum set extraction program 38 of the seventh embodiment, and corresponds to FIG. 20 of the first embodiment.

In step 471, the average waiting times AWT (1) to AWT (P) are extracted from the group management performance data PRE (P) to PRE (P), respectively, and these are taken as the first performance evaluation value VP Sl (l ) To VPS1 (P). In step 472, the long wait rates RLW (1) to RLW (P) are extracted from the group management performance data PRE (1) to PRE (P), and these are taken as the second performance evaluation values. VPS2 (1) to VPS2 (P). In step 473, the minimum value of the first performance evaluation values VPS1 (1) to VPS1 (P) is obtained and set as the best value BVPE.

In step 4 7 4, the optimal set BP is selected based on the performance evaluation value _c. That is, a plurality of sets i satisfying [(VPSl (i) -BVPE) ≤BZ] are found from the storage unit 11. . Furthermore, among them, the second performance evaluation value VPS2 (i)

7 The smallest set is selected as the optimal set BP. In other words, two-stage selection is performed. BZ is a reference value indicating an allowable range from the best value BVPE, and is set to 2 seconds in this embodiment.

Then, similarly to the first embodiment, the optimal set data BPD is created, and the optimal set data BPD is output to the group management device 1 in the next step 388. As described above, in the seventh embodiment, the two-stage selection is applied to the optimal set extraction. Therefore, when the priority is defined in the two evaluation items, the priority is determined according to the priority. Extraction can be realized. Of course, more than two levels of selection can be applied. Note that the seventh embodiment is equivalent to the above [30] Example 6 of the performance evaluation function.

[Example 8 (Other methods of obtaining group management performance value)]

Instead of the simulation device 2 of the first embodiment, as described in Japanese Patent Application Laid-Open No. 57-57168, an actual group management device 1 is used to determine the group management performance for a new set. You can also. This will be described with reference to FIGS. 30 and 31.

FIG. 30 shows a system configuration of the eighth embodiment, which corresponds to FIG. 1 of the first embodiment. FIG. 31 is a diagram illustrating the operation of the group management device 1, and corresponds to FIG. 9 of the first embodiment. The eighth embodiment will be described below focusing on the differences from the first embodiment.

In FIG. 30, when the search condition signal 1 a is input from the group management device 1 to the search device 10 and a search command is issued, the search device 10 executes the evaluation program of step 33 in the arithmetic program (FIG. 15). The simulation condition data is created and output as the simulation condition signal 13a as in the first embodiment.

In this embodiment, the signal 13a is output to the group management device 1, as shown in FIG. On the other hand, the group control device 1 receives the signal 13a. The test run mode. This operation will be described with reference to the flowchart of FIG.

In FIG. 31, at step 491, it is determined whether or not trial operation is being performed, and at step 492, it is determined whether or not trial operation is to be started. When the trial operation flag FLG is “0” and there is no instruction to start the trial operation in the signal 13a, the normal group management operation is executed according to steps 21 to 22.

On the other hand, in step 492, when the start of the trial operation mode is detected based on the content of the signal 13a, the trial operation flag FLG is set to "1" in step 493, and in step 494, , Save the currently used parameter value set temporarily. Then, instead, the evaluation set (new set) NPSX included in the signal 13a is written.

Then, in the same manner as the normal group management operation, the group management operation is performed in Steps 2 21 to 2 29. Also, during the trial operation, since the trial operation flag FLG is “1”, the group management operation is performed in steps 221 to 229 via steps 491 to 4995 through step 491.

In this way, when the group management operation is performed for a predetermined period (for example, 1 hour), the end of the trial operation is detected in step 495, and the trial operation flag FLG is reset to “0” in step 496. Then, the parameter overnight set saved in step 497 is restored, and at the same time, the group management performance data overnight PRF (average wait time, long wait rate, etc.) for this trial operation is calculated. Then, in step 498, the group management performance data PRF is output to the search device 10 as the group management performance value signal 2a. After that, return to the normal state and repeat Step 2 2 1-2

Perform group management operation according to 29.

As described above, when the group management performance signal 2a is obtained by the trial operation,

The search device 10 of 30 obtains the performance evaluation value VPN based on the group management performance value signal 2a and compares it with the evaluation reference value BX, as in the first embodiment, and Judge whether to register additional rule sets.

As described above, in Example 8, since the evaluation of the new set was performed on the actual machine, the time required to obtain the optimum set was long. This has the advantage that the system can be configured at low cost.

[Embodiment 9 (Other embodiments for creating a new set)]

In Example 1, the crossover rate CR and the mutation rate MR were fixed. The ninth embodiment is characterized in that CR and MR are modified according to a search situation.

This embodiment will be described with reference to FIGS. The following description focuses on the differences from the first embodiment and the second embodiment.

FIG. 32 shows the overall configuration of the ninth embodiment, and corresponds to FIG. 1 of the first embodiment. In Fig. 32, the appearance rate correction unit 4 corrects the crossover rate CR and the mutation rate MR (selection rate of each generation method) according to the search situation.

FIG. 33 is a diagram showing an arithmetic program of the ninth embodiment, and corresponds to FIG. 11 of the first embodiment. Note that, except that an appearance rate correction program corresponding to the function of the appearance rate correction unit 4 in FIG. 32 is added to step 50, the processing is the same as the arithmetic program 100 in FIG.

FIG. 34 is a diagram showing the appearance rate correction program. In step 501, the elapsed evaluation frequency NEH, which indicates the number of new evaluations since the previous end determination was made, is calculated by the expression NEH = NE-NEX. Here, the same value as the number of evaluations NE at the time of the previous end determination is set to the number of evaluations NEX at the previous determination. In step 502, it is determined whether or not a predetermined number of evaluations have been performed since the last time. If the number of progress evaluations NEH is less than a predetermined value NEb (for example, 20 times), the program for correcting the appearance rate is ended. On the other hand, if the elapsed evaluation frequency NEH is equal to or more than the predetermined value NEb in step 502, the process proceeds to step 503, where the success index RSC is calculated.

First, the number of new registrations, NRH, which indicates the number of times a new set has been additionally registered since the previous end determination was made, is calculated using the formula NRH = NR-NRX. Then, the success index RSC is calculated by the following equation: RSC = NRH ÷ NEH. Here, the same value as the number of additional registrations NR at the time of the previous end determination is set to the number of registrations NRX at the previous determination. In the next step 504, based on the evaluation number NE and the additional registration number NR at the current time, the evaluation number N X at the previous judgment and the registration number NRX at the previous judgment are updated.

In steps 505 to 510, the number of evaluations NE, the first judgment value NEdl, the second judgment value NEd2, and the success index! ? Correct the crossover rate CR and the mutation rate MR based on SC, the success rate judgment value RSCb, and the success rate judgment value RSCc. For example, NEdl = 500, NEd2 = 800, RSCb = 0.10, RSCb = 0.05.

If NE≤NEdl and RSC ≤RSCb, that is, if the success index is lower than expected in the first half of the search, the currently set crossover rate CR and mutation rate MR Judge as inappropriate, go to step 505 → 507 → 509, reduce the crossover rate CR by a little (for example, 0.001) from the current value, and make the mutation rate MR less than the current value (for example, 0.001) Just make it bigger.

In step 509, contrary to the above correction, the crossover rate CR is set to be slightly larger (for example, 0.001) than the current value, and the mutation rate MR is set to be slightly smaller (for example, 0.001) than the current value. You can also. Either method can be basically applied for breaking down in a sluggish state. In other words, if the crossover rate CR is too large and is not working well, it is sufficient to reduce the crossover rate CR and increase the mutation rate MR. On the other hand, if the crossover rate CR is too small to be successful, the crossover rate CR should be increased and the mutation rate MR should be reduced. After all, if the search is sluggish, the ratio of the selection probabilities of each generation method may be changed to overcome the sluggish state.

If NE NEd2 and RSC ≤RSCc, that is, if the success index is lower than expected at the end of the search, it is determined that the search for the excellent set is almost converging. Steps 5 0 5 → 5 0 6 → 5 0 8 → 5 10, make the crossover rate CR slightly larger than the current value (for example, 0.001), and make the mutation rate MR smaller than the current value (for example, , 0.001).

If the above requirements are not satisfied, the currently set crossover rate CR and mutation rate MR are determined to be appropriate, and the program is terminated without modifying those values. . As described above, in the ninth embodiment, the progress of the search is determined based on the number of evaluations NE and the success index RSC, and the crossover rate CR and the sudden mutation rate MR can be corrected accordingly. Therefore, as compared with the case where the crossover rate CR and the mutation rate MR are fixedly set, an excellent set can be found quickly, and the search time can be shortened. As a result, search efficiency can be improved.

In Example 9, especially when the success index is lower than the expected value in the initial stage (first half) of the search, the crossover rate CR is lower than the current value, and the mutation rate MR is lower than the current value. Since the setting is set to a high value, the possibility of generating a set having better group management performance can be improved by emphasizing a broad search by mutation. Moreover, further _c can be overcome stagnation state of the search, in Example 9, the final stage of the search (late), the success index becomes lower than expected value, the current value of the crossover rate CR High, sudden mutation rate Since the MR is set lower than the current value, local search can be emphasized and the search can be converged at an early stage. In addition, the search can be overcome. [Example 10 (Other example of appearance rate correction)]

Another embodiment of the appearance rate correction unit 4 will be described with reference to FIG. FIG. 35 is a diagram showing an appearance rate correction program, which is a partial modification of the appearance rate correction program of Example 9 (see FIG. 34).

In FIG. 35, in step 502, it is determined that a certain number of evaluations NEb (for example, 40 times) has been reached. In step 506, it is determined that it is the end of the search (that is, NE≥NEd2). Finally, at step 510, the crossover rate CR is set slightly higher (eg, 0.001) than the current value, and the mutation rate MR is set slightly lower (eg, 0.001) than the current value. Thus, in the end of the search, gradually the value of the crossover rate CR As the search progresses increases, the value of the mutation rate MR is still _c gradually reduced summer Conversely, operation other than the above, Example 9 It is exactly the same. As described above, in the tenth embodiment, the generation is switched from the emphasis on mutation to the generation mainly based on crossover in accordance with the progress of the search. On the other hand, the search can be converged early in the latter half of the search. As a result, a set having better group management performance can be efficiently searched.

[Example 11 (Other example of cross pair selection)]

Next, Example 11 will be described with reference to FIGS. 36 to 39. The following description focuses on the differences from the first and second embodiments.

FIG. 36 shows a system configuration of the eleventh embodiment. In this embodiment, the distance between sets is used as a condition for selecting a pair of sets to be crossed (crossed pair). Also, in this embodiment, the condition for the distance between sets (hereinafter, distance condition) can be modified.

One eight Has been established. The parent selection condition correction unit 5 in FIG. 36 corrects the distance condition according to the search situation.

FIG. 37 is a diagram illustrating the calculation program of the eleventh embodiment, and corresponds to FIG. 11 of the first embodiment. In FIG. 37, a new set is generated in step 31. The operation is as described with reference to FIG. 14 relating to the first embodiment. In the first embodiment 11, the process of selecting the crossover pairs PS1 and PS2 (step 3 17) is largely different from that of the first embodiment. different. In step 52, a parent selection condition correction program corresponding to the parent selection condition correction unit 5 (see FIG. 36) is added, and the distance condition described above is corrected. Other configurations are the same as in the first embodiment.

FIG. 38 is a diagram showing the operation of step 3 17 included in the new set generation program 31 (see FIG. 14). In FIG. 38, in step 317a, the value of the counter RC is initialized to ◦. In this embodiment, the counter RC counts the number of times a pair is selected using the set distance condition. The distance between sets is a-in the above equation [31].

In the next step 317b, as in the first embodiment, the selection probability is weighted based on the magnitude of the performance evaluation value, and the two sets PS1 and PS2 to be paired are randomly selected. . In the next step 317c, it is determined whether the set selection using the distance condition has been performed a predetermined number of times (for example, 10 times) or more.

Here, if two sets PS1 and PS2 that satisfy the distance conditions cannot be selected even after repeating steps 3 17 d to 3 17 h more than a predetermined number of times, the selection is made in step 3 17 b. The two sets PS1 and PS2 are determined as a pair that performs crossover.

If the number of set selections using the distance condition is less than the predetermined number (RC less than 10), go to step 3 17 c → 3 17 d, and first, counter RC ^ l Only increase. Then, in step 317e, the distance DST between the two sets PS1 and PS2 selected in step 317e is calculated. Ie

[31] Similarly,

DST = II EPS (PS1) -EPS (PS2) II

Is calculated. Each parameter value is normalized to a value between 0 and 100, as described above.

When the distance DST is calculated in step 317e, it is determined in steps 317f to 317h whether or not the distance condition which is one of the selection conditions of the crossed pair is satisfied.

If the condition selection flag SELS is set to 1 and the first selection condition is specified as the distance condition, the process proceeds to step 317f → 317g, where the distance DST is set to the first selection condition. Judge whether the value is equal to or greater than the reference value DSTbi. If DST <DSTbl, the two sets PS1 and PS2 that do not satisfy the distance condition are discarded, the process returns to step 317b again, and the same process is repeated from the beginning.

If the condition selection flag SELS is set to 2 and the second selection condition is specified as the distance condition, the distance DST is set to the second selection criterion in step 317h as described above. It is determined whether the value is equal to or less than the value DSTb2. If DST> DSTb2, the selection condition is not satisfied, and the process returns to step 317b to start over from the beginning.

In this way, steps 317 b to 317 h are performed until the force that determines the distance condition for a predetermined number of times (10 times) or more, or until two sets that satisfy the distance condition are found by then. Is repeated. On the way, if two sets satisfying DST ≥ DSTb 1 or DST ≤ DSTb2 are generated, these two sets are used as normal cross pairs PSi and PS2, and the process proceeds to the next step 318. The procedure after step 318 is the same as in the first embodiment.

As described above, it is possible to select crossover pairs using the distance condition. You. The distance condition is corrected by the selection condition correction program 52 according to the search situation.

Fig. 39 shows the details of the selection condition modification program of step 52 (see Fig. 37).

In Steps 5 2 1 to 5 2 4, the same processing as Steps 5 0 1 to 5 4 of the appearance rate correction program 50 (see FIG. 3) of Embodiment 9 is performed, and the success indicator RSC is calculated. . This success metric is conceptually defined as the number of additional registrations evaluated.

Steps 5 2 5 to 5 3 6 use the evaluation count NE, the first judgment value NEdl, the second judgment value NEd2 success index RSC, and the success rate judgment values RSCd, RSCe, and RSCf to select the selection reference values DSTbl and DSTb2. Make corrections. Here, for example, NEdl = 500, NEd2 = 800, RSCd = 0.10, RSCe = 0.05, RSCf = 0.05. In step 283 of the initial setting program 28 (FIG. 13), for example, SELS = 1, DSTbl = 250, and DSTb2 = 250.

By the way, if it is determined in step 5 25 that NE ≤ NEdl, that is, the search is still in the “first half”, in step 5 27, SELS is set to 1 and the first selection condition is specified as the cross pair selection condition. . Then, in step 529, the success index RSC is compared with the success rate judgment value RSCd. When the success index RSC is lower than the success rate judgment value RSCd (RSC ≤RSCd), the currently set first selection criterion value DSTbl is too strict, and the number of sets that meet the conditions is small and the success index It is determined that the RSC does not become high, and in step 531, the value of the first selection reference value DSTbl is set to be slightly smaller (for example, 5%).

—If NE≥NEd2, that is, if the search is “final”, proceed to step 5 2 5 → 5 2 6 → 5 2 8 and set 2 to SELS, and select 2nd as the crossover pair selection condition Specify conditions. Then, in step 529, the success index RSC is compared with the success rate judgment value RSCf. If the success index RSC is lower than the success rate judgment value RSCf (RSC ≤RSCf), the currently set second selection criterion value It is determined that the number of crossover sets that meet the conditions is small and the success indicator RSC is not high because the value of DSTb2 is too severe. In step 53, the value of the second selection criterion value DSTb2 is slightly reduced (for example, 5 %).

If the success index RSC is higher than the success rate judgment value RSCf, it is determined that the selection criterion value of the currently selected cross pair selection condition is not a problem, and the processing of the selection condition correction program 52 is performed as it is. To end.

If it is determined in steps 525 and 526 that NEdK NE is less than NEd2, that is, “middle”, the success indicator RSC is compared with the success determination value RSCe in step 533. If the success index RSC is lower than the success criterion value RSCe (RSC ≤RSCe), it is considered that the current selection condition is inappropriate, and the selection condition must be switched. Therefore, after determining whether the current selection condition is the first or second in step 53, in step 53 35, the selection condition is changed to the second selection condition (SELS = 2) or step 5 At 36, the selection condition is changed to the first selection condition (SELS = 1).

When the success index RSC is higher than the success determination value RSCe, it is determined that the currently selected selection condition is not a problem, and the processing of the selection condition modification program 52 is terminated as it is. As described above, in Example 11 above, the distance between sets indicating the similarity between two sets was used as the selection criterion of the crossed pair, so that the properties of both the global search and the local search were Can be used to generate new sets o

In other words, if a pair whose distance between sets is equal to or greater than the first selection reference value DSTbl is preferentially selected and crossover is performed between two sets with different characteristics as much as possible, many hits and misses will converge. However, the probability of generating a set with better group management performance can be increased. On the other hand, the pair whose distance between sets is equal to or less than the second selection reference value DSTb2 is preferentially selected and

8 If crossover is performed between two sets with similar characteristics as much as possible, it is unlikely that a set with very good group management performance will be generated. Power <, Moderately good group management The probability that a new set with high performance will be generated can be increased.

In Example 11 above, particularly, the initial and final stages of the search are determined based on the number of evaluations, and in the early stage of the search, a pair in which the distance between sets is equal to or greater than the first selection reference value DSTbl is preferentially selected. At the end of the search, on the other hand, at the end of the search, the pair with the inter-set distance equal to or smaller than the second selection criterion value DSTb2 was preferentially selected to prioritize the convergence of the search. Can be improved.

Further, in Example 11 described above, particularly when the search is performed using the cross pair selection condition of preferentially selecting a pair in which the distance between sets is equal to or greater than the first selection reference value DSTbl, In the early stage of the search, if the success indicator RSC is lower than the expected value, the first selection criterion value DSTbl is set to be smaller than the current value so that the cross pair selection condition becomes milder than the current value. When the search is sluggish due to the inappropriate first selection criterion value DSTbl, the appropriate first selection criterion value DSTbl can be automatically set to overcome the sluggish state.

Further, in Example 11 above, particularly, in the case where the search is performed using the cross pair selection condition of preferentially selecting a pair in which the distance between sets is equal to or less than the second selection criterion value DS Tb2, At the end of the search, if the value of the success index RSC becomes lower than expected, the second selection criterion DSTb2 is set to be larger than the current value so that the cross pair selection condition becomes milder than the current When the search is sluggish because the set second selection criterion value DSTb2 is inappropriate, the appropriate second selection criterion value DSTb2 is automatically set to overcome the sluggish state. it can.

In Example 11 above, the value of the success indicator RSC is expected especially when the search is performed using the first crossover pair selection condition in the middle stage of the search. If the search condition is lower than the value, the selection condition is changed to the second crossover pair selection condition.If the selection condition does not adapt to the current situation and the search is sluggish, it can be automatically switched to the appropriate selection condition. it can.

Also, in Example 11 above, especially when the search is performed using the second crossover pair selection condition at the end of the search, if the value of the success indicator RSC becomes lower than the expected value, Since the selection condition is changed to the first crossover pair selection condition, if the selection condition does not adapt to the current situation and the search is sluggish, it is possible to automatically switch to the appropriate selection condition.

[Example 1 2 (Other example of cross pair selection)]

Another embodiment of the selection condition correcting unit 5 will be described with reference to FIG. FIG. 40 shows a selection condition modification program, which is a partial modification of the selection condition modification program 52 (see FIG. 39) of the embodiment 11.

In FIG. 40, in step 522, it is determined that the number of evaluations is a certain number NEb (for example, 50). In step 5 26, it is determined whether it is the second period (end stage). If it is not the second period (NE <NEd2), the first condition is specified as the cross pair selection condition in step 5 27, In step 51, the first selection reference value DSTbl is set slightly smaller (for example, 2%). On the other hand, if it is determined in step 5 2 6 that the second period (ie, NE≥NEd2), then in step 5 2 8 the second condition is specified as the cross-set selection condition, and step 5 3 In step 2, the second selection reference value DSTb2 is set slightly smaller (eg, 2%). As described above, in Embodiment 12, in the period in which the cross pair selection condition using the first selection reference value DSTbl is used, the cross pair selection condition can be switched according to the search progress. That is, the value of the first selection reference value DSTbl at the beginning of the period is compared with the value at the end of the period. Is set to be large, the diversity of group management performance can be emphasized at the beginning of the period, and the convergence of search can be emphasized at the end of the period.

Similarly, in the embodiment 12, in the period in which the cross pair selection condition using the second selection reference value DSTb2 is used, the cross pair selection condition can be switched according to the search progress. That is, since the value of the second selection criterion DSTb2 at the beginning of the period is set to be smaller than the value at the end of the period, the diversity of group management performance is emphasized at the beginning of the period, and at the end of the period In, the convergence of the search can be emphasized.

Therefore, the search efficiency can be further improved by switching the selection conditions finely in this way.

[Example 13 (Other example of cross parameter selection)]

In the first embodiment, the crossover parameters (parameter positions) at which the parameter values are exchanged are randomly selected for the two parent sets. The generator of Example 13 is characterized in that the parameter value difference (parameter deviation) is used as a parameter overnight selection condition, and furthermore, the parameter selection condition is modified according to a search situation. And

The generation unit will be described with reference to FIGS. It is to be noted that Embodiment 13 will be mainly described with respect to differences from Embodiment 11. FIG. 41 shows the contents of step 3 18 in the new set generation program 31 (see FIG. 14) of the embodiment 13. In FIG. 41, in step 318a, the value of the counter RC is initialized to 0. In this embodiment, the count RC is used to count the number of times the parameter deviation condition, which is one of the cross parameter overnight selection conditions, is determined. In step 318b, one random number is generated between 0 and 25, and the parameter number PX is specified by the value of the random number. This is the same as in the first embodiment. In the next step 318c, the number of times the above parameter deviation condition is determined is compared with a predetermined number. If the crossover parameter PX that satisfies the parameter deviation condition cannot be found even after repeating step 3 18 d to 3 18 h for a predetermined number of times (for example, 10 times), the selection considering the parameter deviation condition is performed. It is determined that the selection has been made satisfactorily, and the processing of this step 318 is terminated, and the parameter number PX selected in step 318b is determined as the number of the crossover parameter.

If the number of times that the parameter deviation condition is judged is less than the specified number (RC minus 10) in step 3 18 c, go to step 3 18 c → 3 18 d, where the counter RC is set to 1 Only increase. In the next step 3 18 e, the difference between the PX-th numerical values of the two selected sets PS1 and PS2 [I EPS (PS1) <PX> -EPS (PS2) <PX> I] , And this is the distance DSTP. It is assumed that each parameter value is converted into a numerical value between 0 and 100 and normalized. Also, in step 387 of the optimal set extraction program 38 (see FIG. 20), when the optimal set data BPD is created, the value is returned to a value that can be used by the group management device 1.

As described above, when the difference DSTP between the PX-th parameter values of the two sets PS1 and PS2 is calculated in step 318e, then in steps 318f to 318h It is determined whether or not the selected and specified deviation condition is satisfied. If the first selection condition is specified as the parameter overnight deviation condition (SELS-1), go to step 3 18 f → 3 18 g, where the deviation DSTP is greater than or equal to the first selection reference value DSTcl. Is determined. In the case of DSTP or DSTcl, the crossover parameter PX that does not satisfy the parameter deviation condition is discarded, and the process returns to step 318b and the same processing is repeated from the beginning. Similarly, when the second selection condition is specified as the parameter deviation condition (SELS = 2), in step 3 18 h, it is determined whether the deviation DSTP is equal to or less than the second selection reference value DSTc 2 and the selection condition is determined. If you are not satisfied (DSTP> DSTc2) Return to step 318b and start over.

The operations in steps 318b to 318h are repeated until the parameter deviation condition is determined a predetermined number of times (10 times) or more, or until a cross parameter that satisfies the parameter deviation condition is found. If a crossover parameter PX satisfying DSTP DSTcl or DSTP≤DSTc2 is detected on the way, it is determined to be a normal crossover parameter PX, and the process proceeds to the next step 319. Since the procedure after step 319 is the same as that in the first embodiment, the description is omitted. A method of modifying the above selection conditions according to the search situation will be described with reference to FIG.

FIG. 42 is a diagram showing a selection condition correcting program in step 52 of the arithmetic program 100 (see FIG. 37).

The selection condition modification program in the thirteenth embodiment is the same as the selection condition modification program 52 (FIG. 39) in the eleventh embodiment except for steps 531 and 532. Therefore, the operation of steps 529 to 532 for correcting the selection reference values DSTcl and DSTc2 will be mainly described. It is assumed that, in step 283 of the initial setting program 28 (see FIG. 13), the specified data SELS = 1, the selection reference value DSTcl = 50, and the selection reference value DSTc2 = 50.

At the beginning of the search (NE≤NEdl), if the first selection condition is selected (SELS = 1) and the success indicator RSC is lower than the expected value RSCd (RSC ≤RSCd), the currently set first Since the value of the selection criterion value DSTcl is too strict and inappropriate, it is determined that the number of crossover parameters that meet the conditions is small and the success indicator RSC does not become high, and the process proceeds from step 529 to 531. The value of the value DSTcl is set a little smaller (for example, 5%), and the processing of the selection condition modifying program 52 ends. At the end of the search (NE≥NEd2), if the second selection condition is currently selected (SELS = 2) and the success indicator RSC is lower than the expected value RSCf (RSC ≤RS Cf), the current setting Since the value of the second selection criterion DSTC2 is too strict and inappropriate, it is determined that the number of crossover parameters that meet the conditions is small and the success indicator RSC does not become high. Proceed to 2 and set the value of the second selection reference value DSTC2 slightly (eg, 5%) slightly larger. The value expected by the success indicator RSC! If it is higher than SCd, it is judged that there is no problem with the selection reference value of the currently selected selection condition, and the processing of the selection condition correction program 52 is terminated as it is. As described above, in Embodiment 13 above, the parameter deviation indicating the similarity between the two selected sets is obtained, and the crossover parameter selection condition is set based on the parameter deviation. Generation can be performed in consideration of both diversity and convergence of search.

In other words, if control parameters whose parameter deviation is equal to or greater than the first selection reference value DSTcl are selected preferentially and parameters that have as far apart characteristics as possible are crossed over, the convergence of the search is poor due to many hits and misses However, it is possible to increase the possibility that a new set having extremely excellent group management performance is generated. In addition, if a control parameter with the above parameter deviation equal to or less than the second selection criterion DSTC2 is preferentially selected and parameters with similar characteristics are crossed as much as possible, a new set with extremely excellent group management performance will be generated. The likelihood of a new set with reasonably good group management performance is increased, albeit less likely.

Also, in Example 13 above, the initial and final stages of the search are determined based on the number of evaluations. At the beginning of the search, the control parameters whose parameter deviation is equal to or greater than the first selection reference value DST c 1 are preferentially selected and grouped. A search can be conducted with emphasis on the diversity of management performance.On the other hand, at the end of the search,

one (2) A search parameter with priority given to the convergence of the search can be performed by preferentially selecting control parameters that are equal to or less than the selection reference value DST c 2. Therefore, it is possible to generate a new set in consideration of both the diversity of group management performance and the convergence of search according to the search time.

In Example 13 above, when a search is performed using the cross parameter selection condition using the first selection criterion value DSTcl, the value of the success indicator RSC is expected at the beginning of the search. If the value is lower than the set value, the first selection criterion value DSTcl is set to be smaller than the current value so that the above selection condition becomes milder than the current value, so the setting value of the first selection criterion value DSTcl is inappropriate. If the search is sluggish, the value can be automatically corrected to an appropriate value.

In Example 13 above, when a search is performed using the cross parameter selection condition using the second selection criterion value DSTC2, the value of the success indicator RSC is expected at the end of the search. If the value is lower than the set value, the second selection criterion value DSTC2 is set to be larger than the current value so that the above selection condition becomes milder than the current value, so the setting value of the second selection criterion value DSTC2 is inappropriate. If the search is sluggish, the value can be automatically corrected to an appropriate value.

Further, in Example 13 above, in the middle stage of the search, when the search is performed using the first crossover parameter overnight selection condition using the first selection reference value DS Tel, the value of the success index RSC is obtained. If the value becomes lower than the expected value, the condition is changed to the second crossover parameter selection condition using the second selection criterion value DSTC2, so the first crossover parameter overnight selection condition is inappropriate for the current situation. If your search is sluggish, you can automatically switch to the appropriate selection conditions.

In Example 13 above, in the middle stage of the search, the search is performed using the second crossover parameter selection condition using the second selection reference value DS Tc2. In this case, if the value of the success indicator RSC becomes lower than the expected value, the condition is changed to the first crossover parameter selection condition using the first selection criterion value DSTcl. If the search is inadequate for the current situation and the search is sluggish, it can be automatically switched to the appropriate selection conditions.

As described above, by changing the selection criterion value and switching the selection condition according to the search situation, it is possible to appropriately perform a break from the search sluggish state and improve the search efficiency.

[Example 14 (Other example of modification of selection condition)]

Next, another embodiment of the selection condition correcting unit 5 will be described with reference to FIG. FIG. 43 is a diagram showing the operation procedure of the selection condition modification program 52, which is a modification of a part of the selection condition modification program 52 (see FIG. 42) of the embodiment 13.

In FIG. 43, in step 522, it is determined that the number of times of evaluation has reached a predetermined value NEb (for example, 50 times). In step 526, it is determined whether or not it is the second period (end stage). If it is not the second period (NE <NEd2), the first condition is specified as the crossover parameter selection condition in step 527, and the step At 531, the first selection criterion value DSTcl is reduced by a small amount (for example, 2%) to 8 channels.

On the other hand, if it is determined in step 5 2 6 that the second period (ie, NE≥NEd2) is satisfied, in step 5 2 8 the second condition is specified as the crossover parameter selection condition, and in step 5 3 2 Set the second selection criterion value DSTC2 slightly smaller (for example, 2%).

As described above, the condition regarding the parameter deviation is changed according to the search situation. The operation other than the above is exactly the same as that of the embodiment 13.

one As described above, in Example 14 above, in the period in which the cross parameter selection condition using the first selection criterion value DSTcl is used, the value of the first selection criterion value DSTcl at the beginning of the period is used in the end stage. The value was set to be larger than the value, and the above selection conditions were set so that the initial stage of the period was stricter than the end stage, so the diversity of group management performance could be emphasized at the beginning of the period, and the search was conducted at the end Can be emphasized.

In Example 14 above, in the period in which the cross parameter selection condition using the second selection criterion value DSTC2 is used, the value of the second selection criterion value DSTC2 at the end of the period is larger than the initial value. Since the above conditions were set so that they were set smaller and the end of the period was stricter at the end of the period, the diversity of group management performance could be emphasized at the beginning of the period, and the convergence of the search at the end of the period Can be emphasized.

In Examples 9 to 14, the first half of the search, the second half of the search, or the early end of the search were determined according to the number of evaluations NE. The number of additional registrations may be used in place of the number of evaluations NE to determine the first half / second half of the search, or the initial end, etc.

[Example 15 (Other example of parameter selection)]

Next, another embodiment of the generator will be described with reference to FIGS. 44 and 45. FIG. In the embodiment 15, the selection probability (appearance rate) for each parameter is set based on both the degree of association with the traffic flow characteristics and the degree of association with the evaluation item of the group management performance. The basic configuration of the fifteenth embodiment is the same as that of the second embodiment, and therefore, the description will focus on the differences from the second embodiment.

FIG. 44 shows the contents of step 3 18 in the new set generation program 31 (see FIG. 14).

In Fig. 44, in steps 318j to 318q, the parameter appearance rates RPA (1) to RPA (25) for 25 parameters are determined according to the traffic flow specifications. Set. Specifically, first, in steps 318 j to 3181, the traffic flow is determined based on the contents of the traffic flow specification data TRS, such as the total number of passengers, the entrance floor traffic ratio, the up traffic ratio, and the down traffic ratio. Determine the type. That is, it is determined whether the current traffic conditions are in the work hours, up-peak, down-peak, and normal hours.

If it is determined to be a normal time zone, in Step 318m, the appearance rates RPA1 (1) to RPA1 (25) for each parameter prepared in advance for the normal time zone will be used as parameter overnight appearance rates RPA (1) to RPA Set as (25). Similarly, during work hours, in step 318 η, RPA2 (1) to RPA2 (25) are set as parameter overnight occurrence rates RPA (1) to RPA (25). In step 318p, RPA3 (1) to RPA3 (25) are set as the parameter appearance rates RPA (1) to RPA (25) .If the peak is down, RPA4 (1) to RPA4 ( 25) Parameter appearance rate RPA (l) ~! ? Set as PA (25).

At 10B in Fig. 45, the parameter appearance rates RPA1 (1) to RPA1 (25), RPA2 (1) to RPA2 (25), RPA3 (1) to! ? PA3 (25) and RPA4 (1) ~! ? PA4 (25) is shown.

In Fig. 45, when the characteristics of the traffic flow are in the normal time zone, the parameter appearance rates RPA1 (1) to RPA1 (25) are highly relevant to the traffic flow (parameters = 1 to 9, For 22 to 25), the appearance rate is set to 10, and for almost irrelevant parameters (parameter number = 18 to 21), the appearance rate is set to 0. The appearance rate is set to 5 for control parameters that have a moderate degree of relevance (parameter numbers = 10 to 17).

When the characteristics of the traffic flow are during work hours, the parameter appearance rates RPA2 (1) to RPA2 (25) are related to the control parameters that are highly relevant to the traffic flow (parameter number = 1 to 9, For 18 to 21), the appearance rate is set to 10, and almost unrelated control parameters (parameter no. For 7), the appearance rate is set to ◦. The appearance rate is set to 5 for control parameters with moderate relevance (parameter numbers = 22 to 25).

When the characteristics of the traffic flow are up-peak, the parameter occurrence rate RP A3 (i) to RPA3 (25) is the control parameter that is highly relevant to the traffic flow (parameter number = 1 to 13). For, the appearance rate is set to 10, and for almost irrelevant control parameters (parameter number = 14 to 21), the appearance rate is set to 0. In addition, the appearance rate is set to 5 for control parameters that have a moderate degree of relevance (parameter numbers = 22 to 25).

Parameter appearance rate RPA4 (1) ~ when traffic flow characteristics are down-peak ? For PA4 (25), the appearance rate is set to 10 for control parameters that are highly relevant to the traffic flow (parameter numbers = 1 to 9, 14 to 17), and there is almost no control The appearance rate is set to 0 for the parameters (parameter number = 18 to 21). The appearance rate is set to 5 for moderately related control parameters (parameter numbers = 22 to 25).

As described above, in steps 318 j to 318 q, parameter appearance rates RPAh) to RPA (25) for 25 control parameters are set according to the traffic flow specifications. In addition, the appearance rate of each traffic flow RPAl (l) to RPA1 (25), RPA2 (1) to RPA2 (25), RPA3 (1) to RPA3 (25), and RPA4 (1) to RPA4 (25) The values are not limited to those shown in FIG. Any value may be set as long as it relatively represents the degree of association with each traffic flow characteristic. In addition, the appearance rate may be finely differentiated between control parameters overnight.

Returning to FIG. 44, next, in step 318r, the above parameters are obtained by using correction values RPAA (l) to RPAA (25) proportional to the degree of association with the evaluation item (for example, average waiting time) of the group management performance. Appearance rate RPA h)-RPA (25) is corrected. Here, the correction values RPAA (l) to RPAA (25) are set by the performance reference value setting device 3 described above. Here, as described in the second embodiment, the performance reference value setting device 3 sets the “target value” for the average waiting time and the “specified value” for the evaluation reference value BX, as in the second embodiment. In the second embodiment, the “degree of association” with the average waiting time, which is an evaluation item, is output as a correction value.

Therefore, the reference value data TGT in the search condition signal 1a input from the group management device 1 to the search device 10 includes the waiting time target value TAW, the additional reference specification value TCB, and the correction value RPAA (l ) To RPAA (25) are included. RPAA (l) to RPAA (25) in Fig. 45 indicate correction values for the target value of the average waiting time.

Fig. 4 Correction value of 5 RPAA (l) ~! ? In PAA (25), the correction value is set to 10 for control parameters (parameter number = 8, 22, 23) that are highly relevant to the average waiting time as an evaluation item, and there is almost no correlation The correction value is set to 0 for control parameters (parameter number = 10 to 21). The appearance rate is set to 5 for moderately related control parameters (parameter number = 1 to 7, 9, 24, 25).

On the other hand, if the evaluation item is related to power saving rather than the average waiting time, the control values that are highly relevant in the correction values RPAA (l) to RPAA (25) (parameter number = 4 to 7, 2 The correction value is set to 10 for 2 to 25), and the correction value is set to 0 for almost unrelated control parameters (parameter numbers = 9 to 21). In addition, the correction value is set to 5 for control parameters that have a moderate degree of relevance (parameter numbers == 1 to 3, 8).

Similarly, in the case of evaluation items other than the above, the correction value RPAA (1) to! ? PAA (25) is set. Note that the correction values RPAA (l) to RPAA (25) are set to any values as long as they relatively represent the degree of association with the evaluation items. You may. Further, the correction value may be set with a finer difference between the control parameters.

Next, in step 318 s, one random number having a value between [0] and [the sum of the parameter appearance rates RPA (1) to RPA (25)] is generated, and the value of the random number is used. Determine the number PX of the parameter where the crossover or mutation is performed _c. Then proceed to the next step 319. Note that the procedure from step 319 is the same as that of the first embodiment, and the description is omitted. As described above, in Embodiment 15 described above, since the degree of association between the parameter and the traffic flow characteristic is used as the parameter selection condition, the value of the parameter closely related to the specific traffic flow characteristic is preferentially changed. This can increase the possibility that a new set with excellent group management performance will be generated. In Embodiment 15 described above, an appearance rate proportional to the degree of association with the traffic flow characteristics is set for each parameter, and parameters are selected according to the appearance rate. The parameters that are more closely related and more likely to affect group management performance are easier to select, and can increase the possibility of generating a new set with better group management performance.

In Embodiment 15 described above, for control parameters that are not related to the traffic flow characteristics when performing group management, the appearance rate is set to 0 so as not to be selected. It is possible to completely prevent crossover and mutation from being applied.

Also, in Embodiment 15 described above, since the degree of association between the parameter and the evaluation item to be evaluated is set as the parameter selection condition, the value of the parameter closely related to the evaluation item is preferentially changed. As a result, the possibility of generating a new set having excellent group management performance can be increased.

In Example 15 described above, an appearance rate proportional to the degree of association with the evaluation item to be evaluated is set for each parameter, and the parameter is set in accordance with the appearance rate. Since the meter is selected, parameters that easily affect the group management performance are more easily selected, and the possibility of generating a new set having better group management performance can be increased.

In Embodiment 15 described above, the appearance rate is set to 0 for control parameters that are not related to the evaluation item to be evaluated, so that they are not selected. Mutation can be completely prevented from being applied.

Further, in Example 15 described above, since the degree of association between the parameter and the evaluation item to be evaluated, and the degree of association between the control parameter and the traffic flow characteristics were combined as the parameter condition, a more excellent group was obtained. The possibility of generating a new set with management performance can be increased.

As described above, according to the fifteenth embodiment, useless generation of new sets, evaluation, additional registration determination, and the like can be reduced, and search can be performed efficiently.

When a plurality of evaluation items are included in the evaluation function of the group management performance, a weight or a correction value of the appearance rate may be set according to the importance of each evaluation item.

[Example 16 (Other example of cross pair selection)]

Next, another embodiment of the generator will be described. In this embodiment, an intersection set is selected based on the number of similar sets. Hereinafter, the embodiment 16 will be described focusing on parts different from the embodiment 1.

In the new set generation program 31 (see FIG. 11), in this embodiment 16, the process of selecting the crossover pairs PS1 and PS2 in step 3 17 (see FIG. 14) is largely different from that of the first embodiment. No, it is explained using Fig. 46. In FIG. 46, first, in step 3 17 j, the distance DST (i.j) between the sets according to the above equation [31] (where i.j-1, 2,..., Ρ , I ≠ j) ₉ This operation is the same as the deletion program 35 of the third embodiment (FIG. 26). This is the same operation as in step 414 of FIG. It is assumed that the value of each parameter is normalized to a value between 0 and 100.

Then, in step 317k, set number i is initially set to 1, and in step 317p, for all sets (i = 1, 2, ..., P), the appearance rates RSA (l) to RSA ( Until it is detected that the setting of P) has been completed, the processing of steps 317 1 to 317 n is repeated. In step 3171, the number of sets MDST (i) (similar number of sets) for which DST (i, j) ≤DSTa is obtained for all j = 1,2, ..., P where j ≠ i. Here, DSTa is a judgment value for judging whether or not the two sets are similar to each other. In the sixteenth embodiment, 25 is set as in the third embodiment.

Next, in step 317m, based on the number of similar sets MDST (i), the appearance rate RSA (i) of the set i is calculated by the following equation: RSA (i) = l ÷ {MDST (i) + l} Calculate. That is, the smaller the number of similar sets, the higher the appearance rate. Then, in order to calculate the appearance rate for the next set, the set number i is increased by 1 in step 317 η.

In this way, the appearance rates RSA (1) to RSA (P) of all sets are set. Finally, in step 317 q, the sum of the appearance rates RSA (1) to RSA (P) is reduced from [0]. Value], and select two parent sets PS1 and PS2 according to the value of each random number and the appearance rate RMh) to RUA (P).

Thus, the process of step 317 is completed, the two sets PS1 and PS2 are determined as a normal crossover pair, and the process proceeds to the next step 318. _c The procedure after step 318 is the same as in the first embodiment. Omit the explanation o

As described above, in the sixteenth embodiment, since the crossed pairs are selected based on the number of similar sets, the probability of crossing pairs having different characteristics from each other can be improved. Therefore, the probability of generating a new set having extremely excellent group management performance can be improved. In each of the above embodiments, only one crossover parameter was selected (this method of selecting a parameter is generally referred to as one point crossover). Absent. It is also possible to adopt a method of selecting two or more crossover parameters at the same time (multipoint crossover). In addition, a bit string (mask) of the same length as the number of parameters is prepared in advance, and depending on the value of each bit specified in the mask, the gene of either parent (parameter value) is determined. It is also possible to adopt a method called uniform crossover, which determines whether the child inherits the. This is the same for "mutation".

Although the above set was composed of 25 parameters, the number and content are merely examples, and the present invention can be applied to various parameter sets used in the group management algorithm. .

[Example 17 (Other system configuration example)]

In each of the above embodiments, the group management device 1 and the search device 1◦ are provided in the elevator machine room of the building, and the optimal set is obtained online. The power is not limited to this.

For example, in the above Examples 1 to 6 and 8 to 15, as shown in FIG. 47, the search device 10 and the simulation device 2 are installed in the monitoring center of the elevator maintenance company, and the search device The communication device 4A and the communication device 4B can be used to connect between 10 and the group management device 1 via a telephone line. In this case, the communication device 4A can perform data communication with another building having the communication device 4B. With this configuration, one set of search device and simulation device can be shared by a plurality of group management devices. In addition, it is also possible to install the search device 10, the simulation device 2, and the communication device 4A in the manager's office or disaster prevention center of the building. According to this embodiment 17, the system can be configured inexpensively by sharing the expensive search device 1 ° and the simulation device 2.

By the way, in Example 8 described above, as shown in Fig. 48, the search device 10 was installed in the building management room or the monitoring center of the Yerebe maintenance company, and the search device 10 and the group management were installed. It is also possible to connect the device 1 with a telephone line using the communication device 4A and the communication device 4B.

The search device 10 can also be used for developing a group management algorithm, that is, when selecting an optimal group management algorithm plan from a plurality of group management algorithm plans. Normally, when developing a new group management algorithm, simulation is performed using a simulation device, and based on the group management performance data PRF obtained at that time, the performance of the group management algorithm is evaluated or optimized. _c in this case it is performed to or seeking set, as shown in FIG. 4 9, connects the seeker 1 0 the simulation apparatus 2.

The search device 10 is used when searching for and registering the optimal set by the simulation device 2 in FIG. 49 when the group management device 1 is shipped from the factory, or when the initial setting set is used. Available when registering groups GPS1 ~ GPS4. The search device 10 and the simulation device 2 can be realized by a single microcomputer instead of being configured separately, and the group management device 1, the search device 10 and the simulation device 2 can be realized by a single microcomputer. It can be composed of a computer.

Claims

The scope of the claims

1. A system for group management of multiple elevators according to a group management algorithm including multiple parameters,

A search device for searching for an optimal set in a set, which is a combination of parameter values given to the group management algorithm,

The search device,

Storage means for storing a plurality of sets;

Generating means for selecting one or more sets as a parent from the storage means and generating one or more new sets partially inheriting the properties of the parent;

An elevator group management system, comprising: an extraction unit that extracts the optimum set based on the group management performance value from a plurality of sets stored and improved in the storage unit. .

2. In the system according to claim 1,

The generation means,

Numerical exchange means for generating two new sets by exchanging some numerical values with each other between the two sets selected from the storage means, and selected from the storage means A new value replacement means for generating one new set by replacing some parameter values in one set with new random numbers; and

Generation method selection means for stochastically selecting the exchange and the replacement, An elevator group management system, comprising:

3. In the system of claim 1,

The generation means,

Parent selection means for selecting one or two sets from the storage means; and parameter selection means for selecting, for the one or two sets, parameters for which numerical values are exchanged or new values are replaced.

Numerical exchange means for generating two new sets by exchanging the values of the parameters selected by the parameter selection means between the two sets selected by the parent selection means. When,

One new set is generated by replacing the value of the parameter selected by the parameter selection means in one set selected by the parent selection means with a new randomly generated numerical value. New value replacement means,

And a generation method selecting means for stochastically selecting the exchange and the replacement.

4. The system according to claim 3,

The system according to claim 1, wherein said parent selecting means selects a parent based on parent selection criterion information for increasing a probability that an excellent new set is generated.

5. In the system according to claim 4,

The parent selection criterion information is a distance between sets,

The parent selecting means calculates the distance between sets, and randomly selects a pair of sets whose set distance satisfies a predetermined condition from the storage means. Group management system.

6. In the system according to claim 4,

The parent selection criterion information is the group management performance value,

The parent selection means includes a selection probability of each set according to the group management performance value.

0 4 An elevator group management system, wherein weighting is performed and one or two sets are randomly selected from the storage means.

7. The system of claim 4,

The parent selection criterion information is the number of similar sets,

The parent selection means calculates the number of similar sets for each set, weights the selection probabilities to each set based on the number of similar sets, and outputs one or two from the storage means. An all-you-can-eat group management system characterized by randomly selecting a set.

8. In the system according to claim 3,

An elevator group management system including means for correcting a parent selection condition according to a progress of a search.

9. In the system according to claim 3,

An elevator group management system according to claim 1, wherein said parameter selection means selects a parameter based on parameter selection criterion information for increasing a probability of generating an excellent new set.

10. The system of claim 9, wherein:

The parameter selection criterion information is a difference between two parameter values to be exchanged between the two sets, and

The elevator group management system according to claim 1, wherein the parameter selecting means calculates the difference and randomly selects a parameter whose difference satisfies a predetermined condition.

11. The system according to claim 9, wherein:

The parameter selection criterion information is the degree of association between the use status of the elevator and each parameter,

The elevator group management system, wherein the parameter selecting unit weights a selection probability for each parameter according to the degree of association, and randomly selects the parameter.

0

1 2. In the system according to claim 9,

The parameter selection criterion information is a degree of association between the content of the performance evaluation value and each parameter,

The elevator group management system, wherein the parameter selection unit weights a selection probability to each parameter according to the degree of association, and randomly selects the parameter.

1 3. In the system according to claim 3,

An elevator group management system including means for correcting a parameter selection condition in accordance with the progress of a search.

1 4. In the system according to claim 2,

An elevator group management system comprising a probability correction means for correcting the selection probability of each generation method according to the progress of the search.

1 5. In the system according to claim 14,

The probability correcting means includes:

A success index is calculated from a ratio of the number of evaluated sets to the number of sets added to the storage means, and the selection probability is corrected based on the success index. Management system.

1 6. A system that manages multiple evaluators according to a group management algorithm that includes multiple parameters,

The search device,

Storage means for storing a plurality of sets;

By exchanging some parameter values between the two sets selected as parents from the storage means, two new sets that partially inherit the properties of the parents are generated. Numerical exchange means,

Some parameters in one set selected as parents from the storage means

0 6 A new value substitution means for replacing a meter value with a new, randomly generated value to generate a new set partially inheriting the properties of its parent;

An elevator group management system comprising: an extraction unit that extracts the optimum set from a plurality of sets stored and improved in the storage unit based on the group management performance value. .

1 7. The system according to claim 16,

An elevator group management system including means for correcting the additional condition.

1 8. In the system of claim 17,

The elevator group management system according to claim 1, wherein the additional condition is determined based on a group management performance value of each set, and the additional condition is gradually set stricter.

1 9. In the system according to claim 16,

The elevator group management system, wherein the deletion unit deletes a set whose performance evaluation value is poor.

20. In the system of claim 19,

The deletion group management system according to claim 1, wherein the deletion unit deletes a set similar to another set based on a distance between sets.

2 1. In the system according to claim 16,

An elevator group management system comprising an initial setting means for performing an initial setting of a search.

2 2. In the system according to claim 21,

The initial setting means includes a first initial setting mode and a second initial setting mode,

In the first mode, initial setting is performed using a plurality of sets prepared in advance,

In the second mode, the initial setting is performed using a plurality of sets improved in the previous search,

An elevator group management system, wherein a first initial setting mode and a second initial setting mode are selected according to a search start condition.

2 3. In the system according to claim 16,

An elevator group management system including an end determination unit that determines the end of a search according to a search situation.

24. In the system according to claim 23,

The elevator group management system according to claim 1, wherein the end determination unit determines the end of the search based on the number of evaluated sets.

2 5. In the system according to claim 23,

The elevator group management system, wherein the end determination means determines the end of the search based on the number of added sets.

2 6. In the system according to claim 23,

The elevator group management system, wherein the end determination means determines the end of the search based on a success index which is a ratio of the number of evaluated sets to the number of added sets.

27. In the system of claim 23,

The end determination unit is configured to determine a plurality of sets stored in the storage unit. An elevator group management system characterized in that the distance between sets is calculated by using the distance between sets, and the end of the search is determined based on the distance between sets.

2 8. In the system according to claim 16,

An elevator group management system comprising: means for determining re-search based on a change in a precondition given at the start of a search.

2 9. In the system according to claim 16,

The elevator group management system, wherein the storage means stores the group management performance value in association with each set.

30. In the system according to claim 16,

An elevator group management system, wherein a target value setting device is connected to the search device for setting a target value related to search.

3 1. In the system according to claim 16,

In the search device,

A group management device including the group management algorithm and controlling operation of a plurality of elevators;

A simulation device including the same group management algorithm as the group management algorithm included in the group management device;

Is connected,

The evaluation means, wherein the execution result of the simulation device is set as a group management performance value.

3 2. In the system according to claim 31,

The search device and the simulation device are remotely located with respect to the group management device;

An elevator management system, wherein the search device and the group management device are interconnected by a communication line.

3 3. In the system according to claim 16,

The search device includes a plurality of elevators including the group management algorithm.

0 9 one Group management device that controls the operation of

The evaluator group management system, wherein the evaluation unit sets an execution result when the group management algorithm is simulated by the group management device as the group management performance value.

3 4. The system according to claim 3,

An elevator group management system, wherein the search device is remotely located with respect to the group management device, and is interconnected between them by a communication line.

3 5. A simulation device including a group management algorithm for group management of a plurality of elevators,

A search device connected to the simulation device and searching for an optimal set of parameter values for the group management algorithm;

Including

The search device,

Storage means for storing a plurality of sets;

By exchanging some parameter values between the two sets selected as parents from the storage means, two new sets that partially inherit the properties of the parent are generated. Numerical exchange means,

Only good new sets that meet the specified additional conditions are added to the storage means. Additional means for storing,

A system comprising: an extraction unit that extracts the optimal set based on the group management performance value from a plurality of sets stored and improved in the storage unit.

3 6. A system for group management of multiple elevators according to a group management algorithm including multiple parameters,

The search device,

Storage means for storing a plurality of sets;

A crossover type that generates two new sets that partially inherit the properties of the parent by exchanging some numerical values between the two sets as parents selected from the storage means. Generating means;

Evaluation means for obtaining, as a group management performance value, an execution result when the group management algorithm is executed using the new set;

Selecting means for improving a plurality of sets stored in the storage means by using the addition of the new set to the storage means and the deletion of a bad set from the storage means; and

Extracting means for extracting an optimal set having the best group management performance value from a plurality of sets stored and improved in the storage means;

An elevator group management system, comprising:

3 7. A system for group management of multiple elevators according to a group management algorithm including multiple parameters,

A search device for searching for an optimal set in a set which is a combination of parameter values given to the group management algorithm, The search device,

Storage means for storing a plurality of sets;

By replacing some parameter values in one set as a parent selected from the storage means with new random numbers, the properties of the parent were partially inherited1 A mutant generation means for generating two new sets;

Extracting means for extracting an optimal set having the best group management performance value from among a plurality of sets stored and improved in the storage means;

An elevator group management system characterized by including: