CN1044219C - Elevator group control system - Google Patents

Abstract

The present invention relates to a system which uses a group control algorism containing a plurality of parameters to integrally control a plurality of elevator cars. The system comprises a searching device for searching the optimal set of parameters, wherein the optimal set is obtained from the group of parameter values supplied to the group control algorithm. Some new sets are generated through crossover or variation during the operation of the system, superior sets are additionally stored in a memorizer by using additional records, and defective sets are deleted out from the memorizer, so that superior sets are accumulated in the memorizer. Since the optimal set is selected from the accumulated sets, the system can effectively find the optimal set.

Description

Elevator car group control system

Background

1. Field of the invention

The present invention relates to elevator car group control systems, and more particularly, to an apparatus for efficiently finding an optimal combination of control parameter values.

2. Description of the related art

A group control system for elevator cars is a system for efficiently operating a plurality of elevator cars based on various traffic conditions within a building. The group control devices of the system control operation, such as the configuration of the elevator cars, according to a group control algorithm. Wherein the group control algorithm is used to perform and operate various functions and activities related to the operation of the elevator cars, such as configuration control of the elevator cars, etc.

The group control algorithm includes various types of control parameters. For efficient operation, it is necessary to substitute appropriate values for these parameters according to various transport conditions in the building and the like.

In response to the assignment control of floor calls (calls from elevator rooms) as one of the basic group control functions, when one floor is newly registered, an evaluation value Em of each elevator car (elevator room) is calculated according to an assignment evaluation function described below, so that various service conditions such as waiting time and estimation and the like are evaluated based on the new floor call and the floor calls already registered. Then, the elevator car whose estimated value Em is the smallest is selected as the assigned elevator car. Elevator door lights or similar devices located at the floors are turned on before the car arrives, thereby indicating the assigned elevator car and guiding the waiting passengers (this type of operation is referred to as predictive operation).

For example, the following equation [1]]A function for determining the above-mentioned allocation estimate Em is exemplified, where i is the floor call number; m is the elevator car number. <math> <mrow> <mrow> <mi>Em</mi> <mo>=</mo> <munder> <mi>&Sigma;</mi> <mi>i</mi> </munder> <mo>{</mo> <mi>W</mi> <msup> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <mi>Ca</mi> <mo>&times;</mo> <mi>M</mi> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>+</mo> <mi>Cb</mi> <mo>&times;</mo> <mi>Y</mi> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>}</mo> </mrow> <mo>-</mo> <mi>Pm</mi> <mo>-</mo> <mi>Bm</mi> <mo>&lsqb;</mo> <mn>1</mn> <mo>&rsqb;</mo> </mrow> </math> Wherein,

em represents the assigned estimate when a new floor call is assigned to elevator car m; w (i) represents the estimated waiting time for a floor call i when a new floor call is assigned to elevator car m; m (i) represents the probability of a full passenger status for a floor call i when a new floor call is assigned to elevator car m (0. ltoreq. M (i) ≦ 1); y (i) represents the error rate forecast for the floor call i when a new floor call is assigned to elevator car m (0 ≦ y (i) ≦ 1) (the forecast error refers to the situation when another elevator car arrives at the floor first, instead of the forecasted elevator car); pm denotes the Penalty (Penalty) when a new floor call is allocated to elevator car m; bm indicates the prize (bonus) when a new floor call is assigned to elevator car m; ca represents a full-state estimation coefficient; cb represents the prediction error estimation coefficient.

The full-length state estimation coefficient Ca is a relative latency estimation value W (i)²Is the coefficient of the full-length state estimated value M (i) plus. If a larger value is assigned to the coefficient Ca, the system can weight the run through that floor during full load conditions, rather than the run for wait time.

The prediction error estimation coefficient Cb is a relative latency estimation value W (i)²A coefficient weighted by the prediction error estimate y (i). If a larger value is assigned to the coefficient Cb, the system can assign cars to prevent prediction errors from being weighted instead of waiting times.

The priority allocation function using the penalty Pm in equation [1] includes, for example, a travel time priority allocation function [2], and a power saving priority allocation function [3], as described below.

[2] The travel time priority assignment function is a function that attempts to ensure that an elevator car that has received many floor calls rarely receives another call assignment from any other floor. For example, the value (Pa) calculated by the travel time priority multiplied by the number of calls (Nm) is supplied to the penalty function Pm.

[3] The power saving priority assignment function is a function that ensures that an elevator car that is not parked rarely accepts new calls to be assigned. For example, the value represented by the power saving priority Pb is given to the penalty function Pm of the elevator car which is not used for suspension, and the penalty function Pm of each of the other elevator cars is set to zero.

The priority assignment function using the winning function Bm of equation [1] includes, for example, a neighboring car priority assignment function [4], a light-load car priority assignment function [5], and a specific car priority assignment function [6 ].

[4] The adjacent car priority assignment function is a function that ensures that an elevator car located in the vicinity (adjacent elevator car) can be assigned promptly. For example, a value Ba expressed as the priority of the adjacent car is set to the prize function Bm of the adjacent elevator car, and zero is set to the prize functions Bm of the other elevator cars.

[5] The light-load car priority assignment function is a function that enables an idle elevator car or a light-load elevator car (light-load elevator car) to be rapidly assigned. For example, the value Bb indicating the priority of the lightly loaded car is set as the prize function Bm of the lightly loaded elevator car, and zero is set as the prize function Bm of each of the other elevator cars.

[6] The car-specific priority assignment function is a function that enables a specific elevator car to be assigned quickly. For example, a value Bc, which is expressed as a priority of a particular car, is set to a prize function Bm of an elevator car, e.g., being run to the floor, roof, observation floor, etc., and zero is set to the prize function Bm of each of the other elevator cars.

As mentioned above, Ca, Cb, Pa, Pb, Ba, Bb, and Bc are all group control parameters associated with the allocation estimation function [1 ].

Even after the system has performed the allocation function using the allocation estimation function [1], long periods of waiting may still occur in connection with unexpected calls. Therefore, the group control system also has an additional distribution function [7] and a distribution modification function [8 ].

[7] An additional allocation function for long waiting periods is an additional allocation function in which the elevator can perform various services before requiring intervention of the currently allocated elevator.

[8] The assignment modification function for long waits is a function that communicates the long wait call assignment (and forecast) to the intervening elevator car. In order to detect a long wait, the judgment reference value DL is set.

[9] Each elevator car of the group control system has a self-passing function during a full load state. If the load of the elevator car exceeds the reference value DB, it can pass the floor on which the allocation has been performed without stopping. The intervention run is performed by the assignment modification function for calls in which the elevator car has passed by itself during a full load state.

[10] An allocation modification function adapted to the calls already in motion during full load conditions, transmits the allocation and prediction of a floor call to another elevator car in intervention. The forecast modification for the newly assigned elevator car is referred to as forecast modification. As mentioned above, DL and DB are also group control parameters.

In addition to operations related to floor calls, other operations are also performed using various types of control parameters. For example, various types of control parameters are used as conditions for selecting the following operation manner and canceling such selection.

[11] Selecting and canceling rush hour operation

Selecting peak hour operation when the starting time of the peak hour has passed and when the number of calls recorded by the elevator car leaving the main floor first is equal to or greater than a reference value DIUPC; on the other hand, peak hour operation is cancelled when the end time of the peak hour has passed.

[12] Selecting and cancelling an up-peak operation

During DUPT, the top run is selected when the elevator car leaves and the number of main floor passengers exceeds a first reference value DUP 1; on the other hand, the overhead run is cancelled when there is no elevator car departure and the number of main floor passengers exceeds the second judgment reference value DUP 2.

[13] Selecting and cancelling bottom-peak (down-peak) operation

During DDPT, when the elevator car descends and the number of passengers in the elevator car exceeds a first reference value DDR1, a lower bottom operation is selected; on the other hand, when there is no elevator car descending with the number of passengers equal to or greater than the second reference value DDR2, the lower bottom run is cancelled.

Each mode of operation includes the following controls, as well as a control parameter.

[14] Peak hour operation

During peak hours of operation, the elevator cars whose numbers are assigned with the sequential car numbers DIUPN are lined up at the main floor.

In the departure adjustment operation, even if a call is issued to the elevator car which leaves the main floor first and the departure is delayed because the car door is still open, a departure time equal to the door opening time is set to the reference value DIUPT.

The elevator car whose number is designated as open-door waiting car number DIUPW must wait open the door, while the other elevator cars wait closed the door.

[15] Run on top of

In the top run, the elevator cars whose numbers are assigned to the sequential car numbers DUPN are lined up at the main floor.

[16] Lower sole operation

In the bottom run, the calculated forecast waiting time is significantly greater by an amount corresponding to the priority DDPE, when calculated, than the floor call in the direction of the main floor.

[17] Decentralized wait-for-run

A scatter-wait operation is an operation that attempts to reduce latency. When there is a free elevator car, the elevator car is evacuated beforehand on the floor where the next call is likely to occur. Peak hour operation is not selected when such operation is selected.

The scatter-wait run is performed when there are a number of free elevator cars equal to or greater than the conventional number DOHN and when this situation persists for at least one standard time DOHT.

In a decentralized waiting run, the floor or floors on which the elevator car is waiting (waiting floors), the number of elevator cars subject to waiting, and other factors can be used as control parameters.

Furthermore, the following additional control parameters are also used to control the number of cars in operation.

[18] Power saving operation

Power-saving operation attempts to save power by automatically reducing the number of operating cars based on service conditions. When the average waiting time of the last five minutes is equal to or less than the first service reference value DESW1, the system decreases the number of currently operated cars by one, and when the average waiting time is equal to or greater than the second service reference value DESW2, the system increases the number of currently operated cars by one.

As described above, the group control algorithm includes a number of parameters. These parameters serve various control purposes such as shortening the waiting time, improving the prediction accuracy, improving the passenger comfort, saving power, and the like. However, since the purpose of each parameter is different, such group control is generally affected by the combination of values substituted into each parameter.

In other words, it is required to quickly find an optimum combination of parameters in order to realize an efficient group control operation according to various traffic conditions of a building, which vary from time to time, and various desires of passengers.

Note that the combination (series) of parameter values described below is referred to as a "parameter value set" or simply a "set".

In the conventional optimum set finding method, there is a sophisticated calculation method in which each possible combination (i.e., each set) of parameter values is checked (see, for example, japanese patent publication No. 4-51,475 or japanese laid-open patent publication No. 57-57,168).

There are few problems when only a few parameters are used.

However, if the different types of parameters are increased, the number of combinations of the various parameter values to be checked is greatly increased, so that it becomes quite difficult to select the optimal set by checking all possible combinations. The following is a specific description of such a method.

In a sophisticated calculation method, it is assumed that the number of parameters is M and the number of probability values of the parameters is L. For example, if M =3, L =6, the total number of simulations is 216Sub (= L)^M). Therefore, when the number of parameters or the number of probability values is not a small number, it takes a considerable time to determine the optimum combination of parameter values even if simulation is performed or the elevator car is experimentally run by an actual group control device, and thus it is impractical.

A system disclosed in japanese patent publication No. 5-24,067 suggests reducing the number of simulations. That is, for example, when two parameters are employed, the system first finds the optimum value of the first parameter, and then finds the optimum value of the second parameter while fixing the value of the first parameter to the optimum value. Such a process is called a sequential process.

According to such a sequential method, if M =3 and L =6, the total number of simulations is 18(= L × M), which has been greatly reduced as compared with the above-described precision calculation method.

However, the sequential method is effective only when various parameters have no correlation with each other, and is not applicable when the system control includes a large number of highly correlated parameters such as a parameter group for group control.

Summary of The Invention

The present invention is directed to a control system that solves the problems of the conventional systems described above, and that efficiently searches an optimal set for a parameter group having a strong correlation, even if the number of parameter sets is large.

Another object of the present invention is to provide a control system that can implement a dedicated lookup method. To find an optimal set according to a general technique called "genetic algorithm".

(1) In order to achieve the above object, a control system having a search device includes:

a storage device for storing a plurality of sets,

generating means for selecting at least one set at a time from the storage means as a parent set, and generating at least one new set at a time that inherits properties of a portion of the parent set,

an estimating means for taking a result of the execution as a group control performance value each time the group control algorithm is executed with a new set,

a selection means for improving the plurality of sets stored in the storage means by adding new sets to the storage means and deleting defective sets from the storage means, an

And an extracting means for extracting an optimum set based on the group control performance value of the improved set stored in the storage means.

According to the above structure, along with the genetically generated set and the selected good set, the probability of generating a good set becomes large, and only the sub-set (new set) that inherits the good quality of the parent set is stored in the memory. That is, by repeating one cycle, a plurality of sets stored in the storage device are sequentially updated and thereby improved. The best set is finally extracted from the storage means according to the group control performance value of each set. Each value comprising the optimal set is substituted into a corresponding parameter in the group control algorithm so that the system can perform group control such as elevator car assignment.

Thus, an optimal collection or a collection having similar content to the optimal collection can be efficiently found according to the present invention. The amount of operations and the number of simulations can be reduced, thereby allowing the system to find them quickly.

(2) According to the invention, the generating means comprise:

numerical value exchange means for generating two new sets by exchanging the numerical value parts of the two sets selected from the storage means,

new value replacement means for generating a new set by replacing some parameter values of a set selected from the storage means with new values generated in a random manner, and

and the generation method selection device selects between value exchange and new value replacement according to the respective probabilities.

In this configuration, a new set is generated by randomly selecting a "cross" (cross) in the value exchange means and a "mutation" (mutation) in the new value exchange means.

In short, the crossover converges on the solution, while the variation brings changes to the solution. Thus, the crossover is able to converge the contents of a set combination stored in the storage device, but in turn, changes in the set combination do not occur at their early stages, so the system may not go through partial solutions and lose the inherent (best) solution. In this case, the variation may take the system out of the scope of the local solution. In this sense, crossover and variation are in complementary relationship.

On the other hand, mutation has a tendency to invalidate the good solutions found by crossover. In this sense, crossover and mutation are competing relationships. Therefore, it is required to set the ratio between the crossover rate at the selective crossover and the variation rate at the selective variation to an appropriate value. By properly utilizing both crossover and mutation in any way, the system can effectively exploit the advantages of both and increase the probability of generating a good new set.

In addition, in some cases, the system may also find the best set using crossover or mutation.

(3) According to the invention, the generating means comprise:

a parent set selection means for selecting at least one parent set from the storage means, an

Parameter selection means for selecting various parameters by exchanging values or replacing values.

With this configuration, the parent set selection means selects two parent sets (set pairs) from the storage means when the selection cross is made, and selects one parent set from the storage means when the selection change is made. The parameter selection means selects the position (crossing position or variation position) of each parameter at which the parameter value is replaced with the crossing or variation.

(4) The parent set selection device selects a parent set according to the parent set selection reference information to improve the probability of generating a good new set. By selecting reference information using the parent set, the system can make the probability of generating a good new set higher.

(5) If the distance between sets is used as the parent set selection reference information, the similarity between sets can also be used as a selection criterion. This method allows the system to provide priority to changes or convergence (convergence) of the new set.

(6) If the group control performance value is used as the parent set selection reference information, the parent set is selected according to the degree of superiority of each set. The method increases the probability that a good parent set will be selected, and as a result, the system improves the probability of generating a good new set.

(7) If the same number of sets is used as the parent set selection reference information, each original set becomes the selection reference. The method increases the probability of selecting pairs of sets having mutually different characteristics, thereby ensuring the generation of new set changes.

(8) Preferably, the conditions for the mother set selection are modified by the mother set selection means in accordance with the progress of the search. For example, the system may prepare a variety of parent set selection conditions and swap them according to the progress of the search, or change the reference values of the parent set selection conditions according to the progress of the search. Thus, when so configured, the system can increase the probability of generating a good new set.

(9) The parameter selection device selects parameters according to the parameter selection reference information so as to improve the probability of generating a good new set. Therefore, the probability of generating a good new set becomes higher.

(10) If the difference between the parameter values is used as parameter selection reference information, the similarity between the parameters becomes a selection reference. This approach provides priority to the ability to generate new set changes or convergence.

(11) If the degree of correlation is used as the parameter selection reference information in connection with the use of the elevator car, the system can increase the probability of selecting a parameter having a greater degree of correlation with the use of the elevator car, thereby increasing the probability of generating a better update set.

(12) If the correlation is used as parameter selection reference information in conjunction with the content of the performance estimate, the system can increase the probability of selecting a parameter having a greater correlation with the performance estimate, thereby increasing the probability of generating a better update set.

(13) Preferably, the parameter selection condition is modified by the parameter selection means in accordance with the progress of the search. For example, the system may prepare and exchange various parameter selection conditions according to the progress of the search, or change the reference values of the parameter selection conditions according to various situations. The system may increase the probability of generating a good new set.

(14) The selection probability of each generation method is preferably modified by probability modification means in dependence on the search progress. This approach allows the system to take advantage of the search process by mutation and the local search process by interleaving, thereby improving search efficiency.

(15) When the system is so constituted, it is desirable that the selection probability is modified by the probability modifying means based on, for example, the success index. According to this structure, the system can determine the progress of the search from the convergence degree of the search, and thus, the system can set the selection probability suitable for the process.

(16) In another aspect of the invention, a lookup apparatus according to the invention comprises:

a storage device for storing a plurality of sets,

numerical value exchanging means for generating a new set in which two parts inherit the properties of the parent set thereof by exchanging partial parameter values between the two sets selected as the parent set from the storage means,

new value replacement means for generating a new set having a part that inherits the properties of its parent set by replacing the values of the partial parameters of a set selected from the storage means as the parent set with new values generated by a random method,

a generation method selection device for selecting between the numerical value exchange method and the new value replacement method in combination with the objective probabilities,

an estimating means for searching an execution result as a group control performance value each time the group control algorithm is executed by using the new set,

adding means for additionally storing only a good new set satisfying a certain additional condition in the storage means,

deletion means for deleting a damaged set satisfying a certain deletion condition of the storage means, and

extracting means for extracting the optimum set based on the group control performance values between the plurality of sets which have been improved and stored in the storage means.

According to this structure, a new set is generated by selecting a generated set of exchanged values (crossover) in a random manner and replacing the generated set with a new value (mutation). Only new sets that satisfy some additional conditions between the generated new sets are stored in the storage device, and damaged sets are deleted from the storage device.

When this cycle is repeated in order, the storage device only stores good sets, so that the best set can be extracted therefrom. Each value of the optimal set is then placed within a corresponding parameter of the group control algorithm.

According to the system of the present invention, it is thereby possible to efficiently find an optimal collection or a collection whose content is very close to the content of the optimal collection. That is, the system can reduce the number of runs and simulations, thereby enabling rapid lookups.

(17) In a preferred feature of the system, the look-up means further comprises additional condition modifying means.

(18) Additional conditions are determined, for example, from the group control performance values of each set stored in the storage device and modified according to the search progress. By doing so, the system can always accumulate only good new sets in the storage device, reducing unnecessary processes and thereby increasing the probability of generating good new sets.

(19) The deleting means deletes the group control data based on the group control performance value. By doing so, the system leaves only the good aggregate, thereby optimizing the parent aggregate of the plurality of stores as a whole.

(20) The deleting means deletes, for example, according to the distance between sets. By doing so, the system can avoid the existence of duplicate sets of similar states with respect to multiple sets within the storage device and can ensure that the sets change.

(21) The finding means according to the invention further comprise in a preferred feature thereof initialization means. The system may reduce the seek time if the seek condition is encountered as close as possible with initialization of the initialization cluster.

(22) The initialization means preferably comprise a first and a second mode. In the first mode, a plurality of sets prepared in advance are used as an initial cluster; in the second mode, multiple sets that were refined in the last lookup process are used as the initial cluster. If the appropriate mode is selected based on the conditions when the system starts the search, the system can speed up the convergence of the search.

(23) The finding means according to the present invention further comprises in its preferred features end-of-finding determining means. The apparatus determines the end of the search process when the system has entered a state where the system can expect to effectively improve the set during the search. If the lookup process is still insufficient and any lookup process is invalid, the system may cancel the end of the lookup process.

(24) The number of sets to be estimated is related to the execution time of the improvement loop and can be used as a criterion for the end judgment.

(25) The number of sets to be added represents the degree of improvement of the storage device and can be used as a criterion for the termination.

(26) The success index is the ratio of the number of sets to be appended to the number of sets to be estimated, and can be used as a reference for the end judgment since it indirectly represents the convergence capability of the search process.

(27) The distance between sets may be used as a criterion for the ending judgment, which indicates the similarity of sets in the storage device as a whole.

(28) The finding means according to the present invention preferably further comprises re-finding determining means for determining a re-finding on the basis of a change in the various premises at the start of the finding. The apparatus allows the system to automatically find the best set under new conditions. These preconditions include, for example, elevator car specifications, traffic flow specifications, ratio of performance reference values to control reference values, etc.

(29) According to the system of the present invention, it is even possible to store the group control performance values in the storage means.

(30) The finding means may be connected to the target value setting means to set a target value associated with the finding process. In the case where the control target is freely set, the system can search for the optimum set according to the set target.

(31) In the present invention, the new set is evaluated using dedicated simulation means, and the finding means are connected to the simulator in addition to the group control means. The simulator contains a group control algorithm identical to the group control algorithm contained in the group control device. The estimation means sets the execution result of the simulation as a group control performance value. When the simulator is employed, the system can evaluate the new set without interrupting the group control.

(32) In the present invention, the cluster control device is installed in, for example, the same building as the search device (and simulator). Where it is desired to install the look-up device (and simulator) separately from the cluster tool, a communication link is used to connect the cluster tool to the look-up device. If a plurality of group control devices share one searching device (and simulator), the cost of the system is reduced.

(33) In the present invention, a group control device connected to a search device as a utility device may be used to perform the simulation. Therefore, the simulator may not be used and the cost of the system may be reduced.

(34) In the present invention, the group control device and the search device are connected by a communication line when the search device is located at a position far from the group control device, and if a plurality of group control devices share one search device, the cost of the system can be reduced.

(35) The invention can be extended for the purpose of estimating the group control algorithm by connecting the simulator and the finding means.

The relationship between the present invention and GA:

genetic algorithms have been described in various documents and literature (see, for example, "Current genetic Algorithm and problem existence" published in "Measurement and Control" vol. 32, vol. 1, 1993, 1). Basic genetic algorithms typically include a series of initialization cycles, mother set selection, interleaving, mutation, and generation.

In the "genetic algorithm solution to the problem of elevator group control car assignment for fixed calls" published in the 34 th automated control association lecture, which was held from 20 to 22 months 11 and 1991, a system has been described in which a genetic algorithm is applied to elevator car group control.

Such conventional systems make it possible to optimally assign elevator cars for "call replacement" using genetic algorithms. Therefore, the present invention is the same as the conventional system in terms of the system concerning genetic algorithms, but their objects to be achieved are different, and they are also greatly different in terms of the basic structure and the like.

In short, the present invention not only employs genetic algorithms, but also provides a new search technique with an optimal set of parameter values. This feature of the invention will be embodied in the specific structure of the invention derived from a specific set of property parameter values.

Drawings

FIG. 1 is a block diagram showing the general structure of a first embodiment of a system according to the invention;

FIG. 2 is a diagram showing the construction of a search apparatus including a microcomputer;

FIG. 3 is a schematic diagram showing the internal structure of RAM10C of FIG. 2;

FIG. 4 is a schematic diagram showing an internal structure of the ROM10B of FIG. 2;

fig. 5 is a schematic diagram showing the structure of elevator car specification data (ELS);

fig. 6 is a schematic diagram showing a structure of traffic flow specification data (TRS);

FIG. 7 is a diagram illustrating the structure of group control performance data (PRF);

FIG. 8 is a schematic diagram showing the structure of a parameter value set (EPS);

FIG. 9 is a flowchart showing the contents of a control routine of the first embodiment;

FIG. 10 is a flowchart showing a search command routine of the first embodiment;

FIG. 11 is a flowchart showing the search main routine of the first embodiment;

FIG. 12 is a flowchart showing a search initiation determination routine of the first embodiment;

FIG. 13 is a flowchart showing an initialization procedure of the first embodiment;

FIG. 14 is a flowchart showing a new set generation procedure of the first embodiment;

FIG. 15 is a flowchart showing an estimation procedure of the first embodiment;

FIG. 16 is a flowchart showing an addition procedure of the first embodiment;

FIG. 17 is a flowchart showing a deletion procedure of the first embodiment;

FIG. 18 is a flowchart showing an additional reference value modification procedure of the first embodiment;

fig. 19 is a flowchart showing a search end judgment routine of the first embodiment;

FIG. 20 is a flowchart showing the optimal set extractor of the first embodiment;

FIG. 21 is a block diagram showing a second embodiment of the present invention;

FIG. 22 is a diagram showing a RAM in a second embodiment;

FIG. 23 is a flowchart showing an additional reference value modification procedure of the second embodiment;

FIG. 24 is a flowchart showing a search initiation determination routine of the second embodiment;

FIG. 25 is a flowchart showing an initialization procedure of the second embodiment;

FIG. 26 is a flowchart showing a deletion procedure of the third embodiment;

fig. 27 is a flowchart showing a seek end judgment routine of the fifth embodiment;

FIG. 28 is a flowchart showing a seek end determination routine of the sixth embodiment;

FIG. 29 is a flowchart showing an optimum value extracting routine of the seventh embodiment;

FIG. 30 is a block diagram showing an eighth embodiment;

FIG. 31 is a flowchart showing a main program of search in the eighth embodiment;

FIG. 32 is a block diagram showing a ninth embodiment;

FIG. 33 is a flowchart showing a main program for search in the ninth embodiment;

FIG. 34 is a flowchart showing an occurrence (emergence) modification procedure of the ninth embodiment;

FIG. 35 is a flowchart showing an appearance rate modification routine of the tenth embodiment;

FIG. 36 is a block diagram showing an eleventh embodiment;

FIG. 37 is a flowchart showing a main routine of the operation of the eleventh embodiment;

FIG. 38 is a flowchart showing a part of a new set generating program of the eleventh embodiment;

FIG. 39 is a flowchart showing a selection condition modifying procedure of the eleventh embodiment;

FIG. 40 is a flowchart showing a selection condition modifying procedure of the twelfth embodiment;

FIG. 41 is a flowchart showing a part of the new set generating program of the thirteenth embodiment;

FIG. 42 is a flowchart showing a selection condition modifying procedure of the thirteenth embodiment;

FIG. 43 is a flowchart showing a selection condition modifying procedure of the fourteenth embodiment;

FIG. 44 is a flowchart showing a part of a new set generating program of the fifteenth embodiment;

FIG. 45 is a diagram showing the occurrence rates of the respective parameters of the fifteenth embodiment;

FIG. 46 is a flowchart showing a part of a new set generating program of the sixteenth embodiment;

FIG. 47 is a schematic view showing a seventeenth embodiment;

FIG. 48 is a schematic view showing an eighteenth embodiment;

FIG. 49 is a schematic view showing a nineteenth embodiment;

FIG. 50 is a diagram illustrating an optimal set lookup method according to the present invention.

Detailed description of the preferred embodiments [ basic principles ]

As described above, the group control algorithm includes various types of parameters. In order to effectively group control a plurality of elevator cars, the parameter values of the optimal combination need to be searched according to the traffic conditions. The means for this purpose is the best set finding means and fig. 50 shows the basic principle of the finding means according to the invention. As described above, the combination (sequence) of parameter values is referred to as a "set of parameter values" or simply a "set".

As shown in fig. 50, the best set is found by iteratively generating new sets and selecting good sets. It is described specifically as follows:

first, the memory a2 is initialized (a 1). For example, a plurality of initial sets prepared in advance are stored in a memory (a 2).

Then, a new set is generated (a 4). The new set is generated by randomly selecting value swapping (interleaving) or new value permutation (mutation). When the interleaving method is selected, two sets (parent set pairs) are fetched from the memory a2 to exchange a portion of the parameter values between the two sets to generate two new sets. When the mutation method is selected, a set (mother set) is taken out from the memory a2, and a part of parameter values in the set is replaced with new values generated in a random manner to generate a new set.

Note that the selection generation manner, the mother set, and the respective parameters in which values are exchanged are basically performed randomly, and each selection condition may be arbitrarily determined, and each selection element may be weighted according to its selection probability.

The new set a5 generated is next evaluated. That is, the group control algorithm that has been loaded into each new set is virtually or actually executed to obtain the execution result. The obtained execution result indicates the performance of the new set as a "group control performance value", and the new set having a good group control performance value is stored in the memory a2 (A8). Damaged collections are no longer stored but discarded (A9) or deleted after storage (A10). This good set selection method (a7) always accumulates only good sets in memory a 2.

With such a perfecting cycle repeated, the plurality of sets A3 accumulated in the memory a2 can be gradually differentiated and improved. Since the best set can be finally extracted as the best set (a11) among the stored sets A3, the best set is fed to the group control algorithm for group control.

According to the method for searching the optimal set, the sub-set which inherits the good characteristics of the parent set can be effectively generated. That is, the system can increase the probability that a good subset is generated from a good parent, thereby enabling them to be located quickly.

Either of crossover and mutation methods may be used, but preferably a random selection between the two is made so that the multiple cumulative sets can satisfy convergence and variance appropriately.

[ first embodiment ]

Description of the construction

Fig. 1 to 20 show a first embodiment of an elevator car group control system according to the invention. Fig. 1 shows the whole system, which comprises a known group control device 1, a known simulator 2 and a finding device 10.

The group control 1 comprises a microcomputer which in this embodiment controls a group of four elevator cars installed in a 10-floor office building. As described above, the group control apparatus 1 includes a group control algorithm including a plurality of control parameters (see fig. 9).

The group control device 1 is connected to the four car controllers 1A to 1D through communication cables. The car controllers 1A to 1D comprise respective microcomputers, which control the respective elevator cars in a varying manner. Each of the controllers 1A to 1D has various functions such as call recording, operation control, door control, and display control.

The recording function to call the car is to record the call in memory when it occurs. The run control function is the decision to control the travel, stop and direction of movement of the elevator car so that the elevator car responds to calls (car calls and assigned floor calls) that must be responded to. The door control function is to open and close doors provided on the elevator cage and the floors. The display control function is to notify the waiting passenger of the assigned elevator car by illuminating a floor signal lamp and to notify the passenger that the elevator car has arrived by flashing the floor lamp.

The controllers 1A to 1D transmit signals indicating the running states (e.g., car position, traveling direction, open/close states of doors, car call, etc.) to the group control device 1. In turn, the group control device 1 sends signals representing various commands (an assignment command for a floor call, a reference value DB for passage when the car is fully loaded, a set value of a door opening period, and the like) to the cabin (cabin) controllers 1A to 1D.

The cluster control means 1 delivers a look-up condition signal 1a to the look-up means 10 indicating the conditions under which the best set is found. The search condition signal 1a includes "elevator car specification data" for which the elevator car group control system needs to be simulated on the computer, "traffic flow specification data" for which the traffic flow in the building needs to be simulated on the computer, and "search instruction data" for instructing to search for the optimum set. The elevator car specification data comprise e.g. data indicating its number, speed, passenger limit, floor to be stopped and type of elevator car door, as well as additional operations, such as power saving operations and peak hour operations, which are serviced or not serviced. The traffic flow specification data, for example, when the traffic flow in the building is indirectly represented, includes data for combining various characteristic values such as the total number of passengers per hour and the floor-to-floor traffic flow rate, and data for combining various characteristic values such as the number of passengers who get on the elevator per floor and per direction per unit time, and the like, and on the other hand, when the traffic flow in the building is directly represented, it includes passenger data for all passengers (such as the occurrence time, the occurrence floor, the destination floor, and the like).

The simulator 2 includes a microcomputer having the same group control algorithm as the group control apparatus 1. The simulator 2 receives a simulated condition signal 13a comprising elevator car specification data, traffic flow specification data and a set of parameter values. The simulator 2 operates on the signals 13a of the plurality of elevator cars under virtually the same conditions as the actual conditions by means of a group control algorithm. After execution, the simulator 2 provides group control performance data as a group control performance value signal 2a, which represents statistical results (such as average latency, longest latency, etc.) that describe the group control performance.

The finding means 10 comprise a microcomputer and find the best set as described above.

In the lookup means 10, a memory 11 stores a plurality of parameter value sets and stores group control performance data relating to the respective sets. The output signal 11a from the memory 11 comprises a set of parameter values and group control performance data.

The generator 12 generates a new set by the "cross method" and the "mutation method" described above. The new set is temporarily stored in the generator 12 before being estimated by the estimator 13 as described below. The generator 12 transmits a new aggregate signal 12 a.

The simulator 13 generates a simulation signal 13a from the search condition signal 1a and the new aggregate signal 12a and transmits it to the simulator 2. The estimator 13 generates an estimation result signal 13b based on the group control performance value signal 2a output from the simulator 2 after the simulator 2 performs the group control simulation, and transmits it to the adding unit 15.

The additional reference value memory 14 stores an additional reference value for determining whether the new set thus estimated is additionally registered in the memory 11 or discarded. The additional reference value memory 14 delivers an additional reference value signal 14a at its output.

The adding unit 15 generates a performance estimation value for an additional record determined by the group control performance data contained in the estimation result signal 13b and compares the value with an additional reference value. When the performance estimation value is better than the additional reference value, the signal 15a generated by the adding unit 15 comprises the new set and its group control performance value and is transmitted to the memory 11. As a result of this operation, a new set is additionally recorded in the memory 11.

After certain conditions regarding the recording of the sets are met, the deletion unit 16 calculates performance estimates from the group control performance data and makes a deletion decision for each set. The deleting unit 16 selects a set whose performance estimation value is poor, and transmits a deletion instruction signal 16a indicating a poor set number. As a result of this operation, records of the specified set are deleted from memory 11.

The end determination unit 17 determines whether the seek has ended and transmits a seek end signal 17a to the generator 12 when the end of the seek has been determined. As a result of this operation, the lookup means stops generating new sets.

The additional reference value modifier 18 modifies the additional reference value stored in the additional reference value memory 14 by means of a modification signal 18 a. The degree of modification depends on the group control performance data for each set in the memory 11.

The re-search decision unit 19 monitors the search traffic signal 1a and provides a re-search command signal 19a for re-searching the optimum set when the elevator car specification or traffic flow specification changes. If the signal 19a has been output, the seek end instruction is cancelled if the seek end signal 17a is received, and then the seek is started again from the beginning even in the middle of the seek.

The extractor 20 calculates a performance estimation value for determining the best set according to the group control performance data of each set in the memory 11, and extracts the set with the best group control performance data. That is, the extractor 20 extracts the optimal set. The signal 20a at the output of the extractor 20 comprises the best set, elevator car specification data, traffic flow specification data and lookup status data.

The initialization unit 21 includes a plurality of initial clusters and discriminators, and performs initialization with an appropriate cluster among a plurality of initial clusters stored in advance at the start of search to send it to the memory 11, according to the search condition signal 1a or the re-search instruction signal 19 a.

Fig. 2 shows a hardware configuration of the lookup apparatus 10 shown in fig. 1. In fig. 2, the lookup apparatus 10 includes a microprocessor 10A, a Read Only Memory (ROM)10B, a Random Access Memory (RAM)10C, and an input interface circuit 10D and an output interface circuit 10E. In this case, the ROM10B stores a search program indicating the operation steps and the solidification data of the microprocessor 10A. The RAM10C stores the arithmetic operation result (operation data) of the microprocessor 10A, the contents (input data) of the seek condition signal 1a and the group control performance value signal 2a which are externally input, and the contents (output data) of the analog signal 13a and the optimum set signal 20A which are to be externally transmitted.

FIG. 3 illustrates the memory contents of RAM10C shown in FIG. 2; fig. 4 shows the solidified data portion of the ROM10B storage content.

In fig. 3, ELS is data indicating the car specification; TRS is data representing traffic flow specifications; the SCM is data representing a lookup instruction. These input data are contained in the search condition signal 1a shown in fig. 1.

Fig. 5 shows a specific structure of the elevator car specification data ELS. In the example shown in fig. 5, the predetermined number of cars is 4; the speed is 120 m/min; passenger capacity is 20 persons; the elevator car stops at a10 th floor, wherein the 1 st floor is the bottommost floor, and the 10 th floor is the highest floor; the width of the door is 1,000 mm. Regarding the priority assignment operation, [2] running time priority assignment function, [3] power saving priority assignment function, [4] adjacent car priority assignment function, and [5] light-load car priority assignment function are all set to "active", and specific car priority assignment function [6] is set to "inactive". Regarding the sum operation, [11,14] peak time operation, [12,15] top-up (up-peak) operation, [13,16] bottom-down (down-peak) operation, and [17] dispersed wait operation are all set to "active", and [18] power saving operation is also set to "active" as another additional operation. Although not shown in fig. 5, [8] is run in response to allocation modification of long waiting calls, [9] the automatic passage function when fully loaded and [10] the allocation change function are essentially always put "active".

Fig. 6 shows traffic flow specification data TRS of a specific configuration. The example shown in fig. 6 is for a range of business hours (14:00 to 15: 00). For example, the number of lambertian guests is 500 persons per hour, based on the result of the traffic flow actually measured in advance by the group control device 1; the ratio of the traffic volume between the floor and any other floors (floor 2 to floor 10) to the entire traffic volume is 80% (floor traffic rate); the ratio of the uplink traffic volume to the entire traffic volume (uplink traffic rate) is 50%; the ratio of the amount of downlink traffic to the total amount of traffic (downlink traffic rate) was 50%.

The PRF appearing on the upper left side of fig. 3 refers to data indicating group control performance, which indicates the excellence of the respective sets, and is equal to the group control performance value signal 2a shown in fig. 1.

Fig. 7 shows a specific structure of the group control performance data PRF. In this example, the group control performance data PRF includes an average waiting time AWT, a long waiting time RLW, a most common waiting time MWT, a forecast error rate PRE, a forecast change rate RPC, a full time of traffic RBP, an average riding time ABT, a most common riding time MBT, a power consumption PWC, an adjacent car response rate RNR (a rate at which floor calls are handled by cars close to the floor), a light car response rate RLR (a rate at which floor calls registered by a light car will be assigned to it), and a car-specific response rate RSR (a rate at which floor calls registered to a car will be assigned to it).

Referring again to fig. 3, P at the upper right side is data representing the number of sets (perhaps referred to as good sets) recorded in the memory 11; EPS (1) to EPS (Pmax) are data representing sets from first to Pmax; PRE (1) to PRE (pmax) are data indicating the group control performance values corresponding to EPS (1) to EPS (pmax). The aggregation number P, the aggregation data EPS (1) to EPS (pmax), and the group control performance data PRE (1) to PRE (pmax) correspond to the signal 1a shown in fig. 1. As described below, Pmax is a number representing the maximum value of the number of sets that can be recorded.

Fig. 8 shows a specific structure of a parameter value set as an example. In fig. 8, the set includes 25 types of control parameters. That is, each of the set data EPS (1) to EPS (pmax) shown in fig. 3 is as shown in fig. 8. The group control performance data PRE (1) to PRE (pmax) shown in fig. 3 have almost the same structure as the group control performance data PRE (see fig. 7 for a specific structure thereof).

Pn shown between the top and middle on the right side of fig. 3 is data representing the number of newly formed collections; NPS (1) to NPS ((Nmax) is data indicating a new set of the number of sets from 1 st to Nmax the new set number Pn and the new set NPS (1) to NPS (Nmax) correspond to the signal 12a shown in fig. 1 as described below, Nmax is a number indicating the maximum value of the number of new sets that can be generated.

The SIM located at the upper left side of fig. 3 corresponds to the output data of the simulation condition signal 13a in fig. 1, and is composed of aggregate data for estimating NPSX, elevator car specification data ELSX, and traffic flow specification data TRSX. The aggregate data used for estimating NPSX is data representing the content of a new aggregate, and its group control performance is estimated by simulation and constitutes the EPS shown in fig. 8. When the simulation is completed and the ELS in fig. 5 and the TRS in fig. 6 are constructed, respectively, the elevator car specification data ELSX and the traffic flow specification data TRSX are data indicating the elevator car specification and the traffic flow specification, respectively.

RES located below the SIM at the upper left side of fig. 3 is data corresponding to the estimation result signal 13b in fig. 1, which includes the estimation times NE, the set NPSY for estimation, and the group control performance data PRFY. The estimation count NE is data indicating the estimation accumulation count. The set for estimation NPSY is data representing a new set after estimating the group control performance by simulation, which constitutes the EPS in fig. 8. The group control performance data PRFY is data representing a group control performance value given by simulation, and constitutes a PRF in fig. 7.

BX is data representing an additional reference value to determine whether the estimated new set is to be additionally recorded, and corresponds to the additional reference value signal 14a in fig. 1.

RAP is data corresponding to the additional recording signal 15a in fig. 1, and includes the additional recording number NR, the estimation set NPSZ, and the group control performance data PRFZ. The number of additional recordings NR is data indicating the number of times of determining additional recordings. The estimation set NPSZ is data representing a good new set to be recorded in the memory 11, and constitutes the EPS in fig. 8. The group control performance data PRFZ is data representing the group control performance after the group control simulation is performed with the estimation set NPSZ, and constitutes a PRF in fig. 7.

RP shown in the middle on the left side of fig. 3 is data indicating the number of sets whose records are deleted as defective sets with respect to P of the recorded sets EPS (1) to EPS (P), which corresponds to the deletion instruction signal 16a in fig. 1.

FLAG is data (search permission FLAG) for instructing whether to continue searching for the best set or to end the search, and corresponds to the search end signal 17a in fig. 1.

CBX is data to newly rewrite the additional reference value BX, which corresponds to the modification signal 18a in fig. 1.

STR is data used to instruct the restart of the seek to the best set, and corresponds to the re-seek command signal 19a in FIG. 1.

The BPD is output data corresponding to the best-fit signal 20a in fig. 1, and includes a best-fit BPS, elevator car specification data ELSY, traffic flow specification data TRSY, and search status data SS. The best set BPS is the set with the best performance value among the recorded sets, which constitutes the EPS in fig. 8. When the best set BPS is used for the group control simulation, the elevator car specification data ELSY and the traffic flow specification data TRSY are data indicating the specification of the elevator car and the specification of the traffic flow, respectively. Which constitute the ELS in fig. 5 and the TRS in fig. 6, respectively. The search state data SS is data indicating a search state when the optimum set is selected and set to a value indicating the number of estimations NE in the present embodiment.

The GPSO shown below the BPD on the left side of fig. 3 is data corresponding to the initialization signal 21a in fig. 1, and includes an initial set number PK, a plurality of initial sets IPS (1) to IPS (PK), and a plurality of group control performance data PRI (1) to PRI (PK). The initial set number PK is data indicating the set number at the start of search, and is initially set to the same value as the value Pe used to determine the end of the deletion process. The initial sets IPS (1) to IPS (PK) are used as clusters for the start of the search and constitute the EPS in FIG. 8. The group control performance data PRI (1) to PRI (pk) are data indicating the group control performance when the group control simulation is performed using the initial sets IPS (1) to IPS (pk), and constitute the PRF in fig. 7.

VPD (1) to VPD (pmax) shown on the right side of fig. 3 are performance estimation values for erasure decision; VPE (1) to VPE (pmax) are performance credit values used to set the additional reference value BX when modifying the additional reference value; VPS (1) to VPS (pmax) are performance estimates used to determine the best set when it is selected; the VPNs located below the GPSO on the left side of fig. 3 are performance estimation values used as additional decisions when determining whether the estimation set NPSY is to be additionally recorded. In this embodiment, the average latency AWT taken from the group control performance data is placed within each performance value without modification.

NP is data representing the number of sets whose group control performance is estimated between the new sets NPs (1) to NPs (nmax).

WVPE is data representing the worst performance estimate; BVPE is data representing the best performance estimate; RC is a lookup counter used for calculating the number of sets to be used when looking up the worst value WVPE and the best value BVPE; BP is data indicating the number of record sets having the best BVPE.

PS1 is a first parent set number representing the number of parent sets used to generate the new set; likewise, PS2 is the second mother set number; PX is data indicating the number of parameters (positions) for performing the crossover process or the variation process; CR is data representing the cross-selection probability (occurrence rate); MR is data representing the probability of variant selection (frequency).

In fig. 4, Pmax is data representing the maximum value of the number of recordable sets; nmax is data representing the maximum number of newly producible collections; in this embodiment, Pmax is set to 50 and Nmax is set to 20.

NEa is a search end decision value which is used to decide whether the best set search converges or not at a time together with the number of searches NE. In this example, NEa was placed 1,000 times.

Ps is a deletion start determination value for determining whether or not a damaged set is to be deleted at a time together with the number P of recorded sets; pe is a deletion end determination value that is used to determine once whether the deletion process for a damaged set has ended. In this first embodiment, Ps is set at 50 and Pe at 30.

AVPE is data indicating a correction value added to the worst value WVPE of the performance estimation value when the additional reference value CBX is set. In other words, the additional reference value is modified with the correction value AVPE added to the worst value WVPE. As the correction value, a value of zero second or more is usually set; in this embodiment, AVPE is set to 1 second. GPS1 through GPS4 are initialized clusters corresponding to regular (business hours), rush hour, top up and bottom down operations. Each initialization cluster GPS (1) to GPS (4) constitutes the initialization cluster GPS0 in fig. 3.

Description of the operation

Referring to fig. 9 to 20, the operation of the first embodiment will be described. Fig. 9 shows a main part of a control routine in the group control apparatus 1. The control program includes a group control algorithm, and the group control device 1 operates according to the control program. Group control algorithms are known per se.

In fig. 9, step 221 executes a floor call registration routine. In particular, floor calls that occur when a passenger operates a button are recorded in the memory. The call record is erased when any elevator car processes the call.

Step 222 performs an allocation routine, and in particular, calculates an allocation estimate for each elevator car using the allocation estimation function of equation [1 ]. The elevator car whose estimate is the lowest is assigned to handle the call. This step includes, in addition to the basic allocation calculation of the estimation function, the operation based on various functions of a travel time priority allocation function [2], a power saving allocation function [3], an adjacent car priority allocation function [4], a light-load car priority allocation function [5] and a specific car priority allocation function [6 ].

At step 223, the assignment change routine is executed. In particular, the system detects the degradation of the floor call service assigned as described above and performs an assignment restart it. This step comprises the processing [8] of long waiting calls according to an assignment modification operation and the assignment modification operation [10] of floor calls to pass when the car is full.

Step 224 executes a rush hour run program. In particular, the mode of operation is selected or cancelled, e.g., peak hour operation is selected, based on a selection and cancellation condition for peak hour operation [11], and the system controls operation based on peak hour operation [14 ].

Step 225 executes a top-up (up-peak) run. In particular, the mode of operation is selected or cancelled based on a select and cancel condition [12] for overhead operation, and if overhead operation is selected, the system controls operation based on overhead operation [15 ].

Step 226 executes a lower-floor (down-peak) running program. In particular, the operating mode is selected or cancelled based on a selection and cancellation condition [13] for the lower operation, and if the lower operation is selected, the system controls the operation based on the lower operation [16 ].

Step 227 executes a scatter-wait-to-run routine. Particularly, the distributed waiting operation is selected during the off-peak time operation, and the top operation or the bottom operation is selected. When the scatter-wait operation is selected, the system controls the operation according to the scatter-wait operation [17 ].

Step 228 executes a power saving run program. In particular, in order to take into account the power saving in the case of the operation service, the operation [18] is controlled by increasing or decreasing the number of elevator cars put into service in accordance with the power saving operation.

A final step 229 executes the output program. In particular, during a full load condition, a reference value DB [9] required for an automatic passing function to pass when the car is fully loaded is fed to the four controllers 1A to 1D connected to the group control device 1. Each controller 1A to 1D determines whether it is in a full load state based on the reference value DB of traffic when the car is fully loaded and the load. If in a fully loaded state, the control causes the elevator car to move itself through the floor or floors where the call has been made. Since the reference value DB that is passed when the car is fully loaded greatly affects the group control performance, it is treated as a control parameter as if it is finding a target.

Note that the entire group control program (including the control program shown in fig. 9 and the search instruction program shown in fig. 10) is executed periodically (for example, once every 100 milliseconds).

Fig. 10 shows a search instruction program provided in the group control device 1. The program is a command lookup program executed by the lookup apparatus 10.

In fig. 10, when the best set has been obtained with respect to any traffic flow, step 232 is executed to retrieve the best set signal 20a from the finding device 10 to store the best set BPS corresponding to its traffic flow TRS in the memory of the group control device 1. At the same time, the search state data SS contained in the optimal aggregate signal 20a is also stored.

Steps 232 and 233 determine the previous lookup from the lookup status data SS. If the search status data SS is equal to NEa (=1.000), since the search process has been completed, step 234 determines that the optimum set of traffic flows is found, and newly generates and transmits a search condition signal 1a, which includes specification data TRS of traffic flow, elevator car specification data ELS, and search instruction data SCM set to "1".

Conversely, if the search status data SS determined in step 233 is 1 or more, which indicates that the search process has started, then in step 235 the group control device 1 rewrites the value of the search command data SCM in the search condition signal 1a to "0", and outputs a new search condition signal 1a at its output. In order to find the optimum set of each traffic flow, the group control apparatus 1 sequentially selects four types of traffic flows, which correspond to regular operation (business hours), rush hour operation, top up operation, and bottom down operation.

Although the traffic flow specification data TRS exists based on the actual measurement result of the group control apparatus 1, the group control apparatus 1 may be connected to an existing traffic measuring apparatus that accumulates collected traffic condition data (e.g., the number of passengers entering/exiting the car, the number of calls, etc.) and transmits the data to the group control apparatus 1, and based on the data, the group control apparatus 1 may generate the traffic flow specification data TRS.

When the search means 10 adopts a mode in which the optimum set is fixedly provided even during the search, the group control apparatus 1 can determine how many of the optimum sets obtained so far are reliable based on the search status data SS stored at step 231. For example, if the search status data SS indicates a search start phase, the group control apparatus 1 can adopt not only the set provided by the search apparatus 10 but also a set having a satisfactory use effect since it is actually set to be used. This method can prevent the deterioration of the group control performance of the system. If the search status data SS indicates that a half or final phase is found, the best set from the search means 10 is judged to be very reliable, so that the system performance can be improved with the group control run set before the search process is completely finished.

Fig. 11 shows a search routine (main routine) stored in the search device 10. The program is stored in the ROM 10B.

In fig. 11, step 25 executes a re-search judging program having the function of the re-search unit 19 shown in fig. 1. Step 26 executes a seek start determination routine and the device 10 determines whether it is restarting the optimal set seek time. Referring now to FIG. 12, a re-lookup decision method is shown.

In fig. 12, in step 261, the lookup device 10 receives the lookup condition signal 1a from the group control device 1, and stores the elevator car specification data ELS, the traffic flow specification data TRS, and the lookup command data SCM in the RAM 10C. Then, in a next step 262, the finding means 10 detects a change of the finding instruction data SCM from "0" to "1", if a change is detected, and in a step 265, the finding means 10 sets the finding start flag STR to "1". On the contrary, when the lookup means 10 judges that the lookup command data SCM has not yet changed from "0" to "1", the means 10 judges at the detection step 263 whether or not the elevator car specification data ELS is different from the elevator car specification data ELSX which has been checked so far, and judges at the step 264 whether or not the traffic flow specification data TRS is different from the traffic flow specification data TRSX which has been checked so far. If ELS is different from ELSX or TRS is different from TRSX, the search start flag STR is set to "1" in step 265, otherwise, the search start flag STR is set to "0" in step 266.

The reason why the search is restarted is judged at steps 263 to 265 is that the best set needs to be searched again when the search situation changes, so that it is highly likely that the best set of the current record is no longer the best. For example, when the traffic flow in a building changes due to changes in occupants, or when a portion of the group control algorithm changes to improve system performance, a re-lookup is performed.

Referring to fig. 11, in step 27, the lookup apparatus 10 determines whether re-initialization is required according to the result of step 26. On the one hand, STR equal to "0" means that the search is in progress. On the other hand, STR not equal to "0" means that the search is stopped halfway and started from the beginning, or that the search is restarted after the current search is ended. Therefore, when the STR is equal to "0", the operation procedure proceeds to the generation program 29, but when the STR is not equal to "0", the initialization program is executed at step 28, and then, after initializing various data, the procedure proceeds to the generation program 29.

Referring to fig. 13, the operation of the initialization program 28 will be described below.

In fig. 13, at step 281, an initial data set suitable for a designated traffic flow is selected from among a plurality of initial data sets previously stored. Each initial data set includes an initial set number PK, an initial set of PK terms, and group control performance data for the PK terms.

For example, in a case where a regular time range is specified according to the traffic flow specification data TRS, an initial data group suitable for regular operation is selected from the plurality of initial data GSP1 to GSP4, and the selected initial data group is recorded as the data group of the initialized GPS0 shown in fig. 3. The data set of the initialization GPS0 includes an initial set number PK, a plurality of initial sets IPS (1) to IPS (PK), and group control performance data PRI (1) to PRI (PK).

The value of the initial set number PK to be set is the same as the deletion end determination value Pe (= 30).

At step 282, an initial set number PK is entered as a set number P; initial set IPS (1) to IPS (pk) as recorded set EPS (1) to EPS (p) inputs; the group control performance data PRI (1) to PRI (PK) are inputted to the group control performance data PRE (1) to PRE (P). That is, as shown in fig. 50, initialization a1 of the memory a2 is performed.

In step 283, as initialization, the estimated number of times NE is set to zero, the additional recording number of times NR is set to zero, the estimated number of sets NP is set to zero, the search permission flag is set to "1", the crossover probability CR is set to 1.0, and the mutation probability MR is set to 0.01. Then, the routine is ended.

Referring to fig. 11, step 29 executes a generation procedure corresponding to the generator 12 of fig. 1. First, step 30 determines whether the search is to continue. If the search permission FLAG is "0", the process returns to the search start judgment routine in step 26, while on the other hand, if the search permission FLAG is "1", the process enters the new set generation routine in step 31.

Referring to fig. 14, the new set generation procedure will be described below.

In fig. 14, first, in step 311, it is determined whether a new set that has not yet been estimated remains. If the number of estimated sets NP is less than the maximum Nmax, the process immediately exits from step 311 of the procedure to estimate a new set since there is a remaining set that has not yet been estimated. Conversely, if the number of sets of estimates NP is at the maximum value Nmax or greater, or if the estimates for all new sets have been completed, the process proceeds to step 312 and initializes, placing the number of sets Pn that have survived at zero.

In the next step 313, each average waiting time AWT (1) to AWT (P) is taken from the group control performance data PRE (1) to PRE (P) of the corresponding P record sets EPS (1) to EPS (P) and placed in the performance estimation values VPS (1) to VPS (P) for maximum value determination. The probability (occurrence rate) that each set is selected as the parent set is determined from the inverse of the performance estimates VPS (1) to VPS (p).

Then, a new set up to the maximum value Nmax is generated by repeating the process of steps 314 to 324 as described below.

First, the generated number of collections Pn is incremented by 1 at step 314. In step 315, a random number having a value between zero and [ CR + MR ] (the sum of the cross-over rate and the variance rate) is determined, optionally with a selection of generation. The selected mode is "crossover" if the random number is less than CR (=1.0), and "mutation" if the random number is CR (=1.0) or above 1.0, and the ratio between the crossover rate (probability of selective crossover) CR and the mutation rate (probability of selective mutation) MR can be modified according to certain references.

If a "cross" is selected at step 316, the process proceeds to step 317. The reciprocal value of the corresponding performance estimate VPS is given as a weight value for each set. The weighted value represents the probability that the set will be selected. Then, two random numbers are generated in the range from "0" (as a lower limit) to "the sum of reciprocals of performance estimation values VPS (1) to VPS (p) (as an upper limit). Then, two sets are selected in conjunction with the generated random numbers. Now, assume that the two parent sets (cross-set pairs) are a set EPS (PS1) of number PS1 and a set EPS (PS2) of number PS 2.

In a next step 318, a random number having a value between 0 and 25 is generated and a parameter PX that is specified by the random number is selected. The parameter is shared between the two parent sets and determines the location of the parameter where the value is exchanged.

In step 319, the PX-th values in the parent set EPS (PS1) and the PX-th values in the parent set EPS (PS2) are interchanged. This operation generates two new sets. The two new sets are set to a new set NPS (Pn) of number [ Pn ] and a new set NPS (Pn +1) of number [ Pn +1 ].

Finally, the number of new collections Pn that have been generated is incremented by 1 in step 320.

On the other hand, if "mutation" is selected at step 316, the process proceeds to step 321. The reciprocal value of the corresponding performance estimate VPS is given as a weight value for each set. The weighted value represents the probability that the set will be selected. Then, a random number is generated in the range from "0" (as a lower limit) to "the sum of all the reciprocals of the performance estimation values VRS (1) to vps (p)" (as an upper limit). Then, a set is selected in conjunction with the values of the generated random numbers.

Like step 318, a random number is generated in step 322 to select the parameter PX in which the variation is generated.

In step 323, another random number survives between [ min ] and [ max ] (parameters may be taken that are specified by the number PX). The random number is substituted for the number PX in the good set EPS (PS1) of the number PS 1. In this operation, the generated set is the new set NPS (Pn) of numbers Pn.

At next step 324, a determination is made as to whether a new set with the required number is generated. Since two new sets are generated in one "crossover", if Pn +2 > Nmax, the generation of the new sets can be completed. Before this is done, a new set is then generated by repeating steps 314 to 324 up to a number Nmax. When the generation of the new set is completed, the first estimated set number of the new set or the estimated set number NP is set to 1 in step 325.

"crossover" is a method of finding a convergence solution (solutions). In contrast, "mutation" is a search method with a changing solution (solved solutions). That is, if the process is performed only by interleaving, the search direction is defined, thereby increasing the probability of missing the optimal solution. However, in addition to interleaving, the system may avoid a partial solution if mutation is employed as appropriate. In this case, the two methods are complementary. However, the risk of using variants is that the best solution, even if eventually found, may be corrupted. In this case, the two methods are again in a competitive relationship.

Therefore, in order to avoid the risk that both will form a competitive relationship while utilizing their complementary relationship, in the first embodiment, the variation rate MR is set to an extremely small value compared to the crossover rate CR.

In the present embodiment, after twelve (= Nmax) new sets are generated, the group control performance is estimated with respect to each new set. However, other methods may be used.

For example, by setting the maximum value Nmax to "1", the group control performance with respect to the new set can be estimated for each "crossover" or "mutation" and can be repeated.

In this connection, when the maximum value Nmax becomes larger, the operation time can be shortened because a new set can be generated immediately. In this case, however, the system needs to secure the RAM10C with a large capacity. Preferably, the maximum value Nmax is determined in relation to the running time and the memory capacity.

Referring to fig. 11, the estimation procedure of step 33 will be described below. This estimation procedure corresponds to the simulator 13 in fig. 1, which executes the group control algorithm by supplying the new set to the simulator, thereby obtaining the execution result.

Referring to fig. 15, the estimation procedure will be described in detail below.

In fig. 15, step 331 generates simulation condition data including sets NPSX for estimation, elevator car specification data ELSX, and traffic flow specification data TRSX. That is, the system sets the new set nps (np) to the set NPSX for estimation, and sets the elevator car specification data ELS and the traffic flow specification data TRS included in the search condition signal 1a to the elevator car specification data ELSX and the traffic flow specification data TRSX, respectively. In a next step 332, the finding device 10 is configured to output the simulated condition data SIM as the simulated condition signal 13a to the simulator 2, so that the simulator 2 performs the virtual group control operation. At step 333, the process waits for the simulation to complete.

The simulator 2 simulates according to the simulation condition signal 13a and sends the group control performance value signal 2a to the look-up device 10 when the simulation is completed. Upon receipt of the group control performance value signal 2a, it is determined that the simulation is complete in step 333, and the group control performance data PRF contained in the group control performance value signal 2a is stored in the RAM10C in step 334. The process then proceeds to the next step 335.

In step 335, the number of estimation times NE is added by 1, and estimation result data RES including the number of estimation times NE, the set for estimation NPSY (= NPSX), and the group control performance data PRFY (= PRF) is generated. Then, in step 336, 1 is added to the value of the estimated number of sets for the new set.

Referring to fig. 11, the addition procedure of step 34 is relative to the addition unit 15 of fig. 1, which determines whether a new set (number of units NP) is recorded. This addition procedure is described with reference to fig. 16. In fig. 16, in step 341, the average waiting time AWT is taken out from the group control performance data PRFY and set as the performance estimation value VPN for addition recording judgment. In step 342, the performance estimation value VPN and the additional reference value BX are compared to determine whether they are to be recorded in the memory. If VPN is equal to or greater than BX, the system does not allow the record and terminates program 34 immediately.

On the contrary, if the VPN is smaller than BX, the process proceeds to step 343, and the number of addition recordings NR is added by 1, thereby generating data RAP for addition recording, which includes the number of addition recordings NR, the estimated set NPSZ (= NPSY), and the group control performance data PRFZ (= PRFY). In step 344, the new set is additionally recorded as a set of numbers [ P +1], while the value of the already recorded set P is incremented by 1.

Referring to fig. 11, the deletion process of step 35 corresponds to the deletion unit 16 in fig. 1, which deletes the set of performance estimation value differences.

Referring to fig. 17, the deletion procedure is now described. In fig. 17, the record set number P and the deletion start determination value Ps are compared at step 351 to determine whether or not they are the time to delete a record. If P is less than Ps, it is determined that it is not the correct time to delete the corrupt set and the process immediately exits routine 35. If P is equal to or greater than Ps, it is judged as the correct time to delete the damaged set, so the damaged set is deleted by repeating steps 352 to 359 until the number of sets P becomes the deletion end judgment value Pe.

In step 352, the average waiting times AWT (1) to AWT (p) are respectively extracted from the group control performance data PRE (1) to PRE (p) and set to the performance estimation values VPD (1) to VPD (p), respectively. Initialization is then performed at step 353 to detect a corrupt set to be deleted. That is, the counter RC for the lookup is set to 1, the worst value WVPE of the performance estimation values is set to zero, and the deletion set number RP is set to zero.

The set (the number of sets RP) having the worst performance estimation value is determined by repeating the process of steps 354 to 357. That is, when any set of performance estimates vpd (rc) that are inferior to the existing worst WVPE is determined at step 354, the performance estimates vpd (rc) are newly set to the worst WVPE at step 355. The counter value RC for the lookup is set to the number of deletion sets RP. The counter RC is incremented by 1 at step 356 and a determination is made at step 357 as to whether the lookup has been completed for all sets.

At step 358, the record for the set with the worst value WVPE (whose number of deleted sets is RP) is deleted, and the group control capability data Record Pre (RP) is also deleted. The already recorded set value P is also decremented by 1. Thus, the retained sets are sequentially renumbered from 1 so that the sets are restored and the process ends at step 358.

In step 359, it is judged whether the number of sets P after deletion is equal to or smaller than the deletion end determination value Pe. If not, the system repeats the process of steps 352 through 358 as described above. When P becomes equal to or smaller than Pe, the deletion program ends.

In the present embodiment, although step 351 sets the deletion start determination value Ps and the deletion end determination value Pe to 50 and 30, respectively, they may be set to other numbers.

The deletion start judgment value PS is set to a range in which the value does not exceed the maximum value Pmax of the set that can be stored in the RAM 10C. If the deletion start judgment value is set to Ps = Pe +1, a replacement set is deleted whenever a set is additionally recorded. This method is convenient when the RAM10C does not have sufficient storage capacity.

The deletion end judgment value Pe of step 359 means that the reserved sets become parent sets. If the deletion end determination value Pe is small, the probability of generating a good new set will decrease because it is difficult to retain the change of the generated new set. On the contrary, if the deletion end determination value Pe is large, the generated new set will be surely changed, and as a result, the probability of generating a good new set will be increased. However, as an effective lookup process, it is not desirable to provide a larger number for the deletion end determination value Pe because this would increase the generation operation.

Therefore, it is desirable to determine the deletion end judgment value Pe according to the combination of the two sets performing the intersection, the kind and number of control parameters, and the like, or sometimes by trial and error. Since Pe =30, the present embodiment ensures the combination (=30 × 29 ÷ 2).

Referring to fig. 11, the additional reference value modification routine of step 36 is directed to the additional reference value modifier 18 of fig. 1, which modifies the additional reference value BX based on the record status of the collection in the memory 11.

Fig. 18 describes an additional reference value modification procedure. In fig. 18, in step 361, each average waiting time AWT (1) to AWT (p) is taken from the group control performance data PRE (1) to PRE (p) and placed in the performance estimation values VPE (1) to VPE (p) for setting the reference values. In step 362, the system operates to determine a worst value WVPE among the performance estimates VPE (1) through VPE (p) for setting the reference value. This operation is the same as the procedure of steps 353 to 357. At step 363 the modified value CBX is obtained by calculating the worst value WVPE-correct value AVPE, which is modified by placing it at the additional reference value BX at step 364.

In the present embodiment, the correction value AVPE is fixedly placed one second between the start and end of the seek. That is, the additional reference value BX (which refers to the average waiting time) is set to be less than one second per cycle. However, other values may be used by the system.

If the correction value AVPE is set to a large value, the condition for additional recording becomes increasingly strict, so it will not be a very large value in order to obtain as many good sets as possible within a limited evaluation time. Conversely, if the correction value AVPE is set to zero seconds, the probability that additional records have many sets with similar features and a small number of performance distinctions can be increased. Therefore, the value must be appropriately determined according to the search condition.

Referring to fig. 11, the search end judgment routine of step 37 corresponds to the search end judgment unit 17, which judges whether or not the search for the optimum set has ended. The following description is made with fig. 19.

In fig. 19, step 371 judges whether the search process has ended based on the estimated number NE and the search end judgment value NEa. If NE < NEa. It is determined that the search process is not completely completed and the search permission FLAG is set to "1" in step 372, and the search process is continued. If NE ≧ NEa, judge that the search process has all been carried out, and set up the permission FLAG of the search to "0", finish the search process.

In the present embodiment, although the search end determination value NEa is set to 1,000, the determination value NEa may be other values.

It is often difficult to determine how many times to estimate is sufficient. This is because convergence of the search depends substantially on various search conditions such as the kind and number of control parameters, the content of the initial collection, the method of generating the new collection, and the conditions of the additional recording.

In order to obtain many good sets representing good group control performance, the end-of-search decision value NEa should be placed at a value as large as possible. However, if the accumulated value NE of the number of seeks becomes too large, it will take considerable time to end the seeking process, thereby resulting in a less effective seeking process. Therefore, in order to obtain many good sets by a few, it is necessary to determine the search end judgment value NEa according to the search condition.

Referring to fig. 11, the best set extraction routine of step 38 corresponds to the extractor 20 of fig. 1 for extracting a best set from each set. The following will be described with reference to fig. 20. In fig. 20, step 381 extracts each average waiting time AWT (1) to AWT (p) from the group control performance data PRE (1) to PRE (p), and places the performance estimation values VPS (1) to VPS (p) for optimum value judgment. The system then initializes to detect the best set at step 382. That is, the counter RC of the lookup is set to 1; the best value for the performance estimate, BVPE, was set to 9,999; the number of sets BP is set to zero.

Next, the optimal set (set number BP) having the best performance estimation value is determined by repeating the process of steps 383 to 386. That is, in step 383, the previously obtained performance estimation value vps (rc) and the best value BVPE are compared. If the performance estimate VPS (RC) is detected to be better than the best value BVPE, the performance estimate VPS (RC) is set to the best value BVPE and the value of the counter RC looked up is set to the collection number BP at step 384. The counter RC of the lookup is incremented by 1 in step 385. At step 386, a determination is made as to whether all sets of lookups have been completed. At step 387 best set data BPD is generated which includes best set BPS, elevator car specification data ELSY, traffic flow specification data TRSY and lookup status data SS. Specifically, the contents of the set having the best value BVPE are set as the best set BPS, and the same contents as the elevator car specification data ELSX and the traffic flow specification data TRSX of the simulation condition data SIM are set as the elevator car specification data ELSY and the traffic flow specification data TRSY, respectively. The estimated value NE up to this point is set as the search state data SS.

Finally, the best aggregate signal 20 containing the best aggregate data BPD is sent to the group control device 1 in step 388.

Referring to fig. 11, as described above, if the search for the optimal set is completed, the process returns to step 26, and steps 26, 27, 30 to 38 are repeatedly performed until the completion of the search is detected, and the search permission FLAG is reset to "0" in the search end judgment routine. In the searching process, if the contents of the elevator car specification data and the traffic flow specification data are changed, re-searching is performed. That is, the search permission FLAG is changed to "1", and the steps from step 31 onward are executed.

Advantages of the first embodiment

As described above, according to the first embodiment, a good set can be efficiently generated, whereby an optimum set can be efficiently found. Since the simulator 2 for group control is a device independent from the group control device 1, finding the optimal set does not interfere with the original group control operation.

Since the new set is generated in two ways, the first embodiment can effectively accept both the "cross" and "variant" properties. In other words, the new set can be prepared with the appropriate changes and convergence, so the best set can be found earlier by combining extensive and local search.

In a first embodiment, the parent sets are selected after each parent set is weighted with a selection probability based on performance estimates, so the system can increase the probability of selecting good parent sets, in other words, the system can increase the probability of generating good new sets that inherit the superior properties of the parent sets.

With the first embodiment, since the additional reference value is modified based on the worst value of the plurality of performance estimation values, the system can avoid a useless process that needs to be deleted immediately after the additional recording, thereby becoming progressively tighter. If the search end judgment, the mother set condition modification and the parameter selection are completed according to the additional record number, the system can execute appropriate processing according to the search progress condition.

It can also be seen in the first embodiment that reasonable set recording is achieved in view of the capacity of the memory, since the recorded set can be maintained smaller than a certain number resulting from the deletion process. As a result, more good sets can be selected from as many sets as possible.

With the first embodiment, since the system constantly deletes sets whose performance estimation values are poor, the system can keep the performance estimation values as good sets, thereby always providing good sets like the mother set when generating new set pairs.

In a first embodiment, the extractor finds the best set even during the search and provides the best set at its output. Therefore, even in the search process and during use, the group control apparatus 1 can obtain the optimal set without waiting for the completion of the search.

Since the cluster control apparatus 1 can obtain the search state data (the estimated number NE) from the optimum set in the search process, it can correctly judge a value for using the provided optimum set in the search process.

With the first embodiment, successive lookups can be performed until the lookup time reaches a certain number, preventing the lookup process from ending before enough lookups are performed.

Furthermore, in the first embodiment, the system also has a re-search function, so that when any of the elevator car specification data and the traffic flow specification data changes, the system can automatically restart searching for a good set even after the search is finished. Therefore, even if the instruction to start the seek from the group control apparatus is delayed for some reason, the system can start the seek quickly. In an early stage, the system may thus obtain an optimal set corresponding to the latest group control conditions. When the elevator car specification data or traffic flow specification data changes, the system can automatically re-search from the beginning under new group control conditions using a re-search function even during the search.

In a first embodiment, an initial cluster corresponding to each traffic flow specification is prepared in advance. When starting the lookup, the system may select the most appropriate initial cluster for traffic flow at its own initialization. Thus, the system can provide a set of degrees of optimization from the beginning as the parent set, thereby enabling rapid lookups. When the set provided as the best set is used in the search process, the system can still share good group control performance on a certain program even at an early stage of the search process.

In the first embodiment, the group control performance data PRF obtained by the simulator 2 is stored in the memory 11. The group control performance data PRF includes a plurality of data shown in fig. 7. Some data included in the group control performance data PRF are replaced by performance estimation values VPN for additional recording, performance estimation values VPE (1) to VPE (p) for setting reference values, performance estimation values VPD (1) to VPD (p) for deletion judgment, and performance estimation values VPS (1) to VPS (p) for optimum value judgment. Thus, no simulation is required at any time when each performance estimate needs to be obtained. When the performance estimation value includes common data (such as average latency), it is natural that only the common data is stored as the group control performance data PRF.

[ production method and selection method outline ]

The generation method of generating new (n +1) th generation individuals (subsets) from the nth generation individual group (parent set) is divided into two types. The first method is to use the newly generated individuals (subset) as the parent set to generate the next new individuals (subset) belonging to the same generation as the parent set, i.e., (n +1) th generation. On the other hand, the second method is not the above method, and in particular, wherein:

gn (mn) denotes a population of nth generation individuals whose compiled length is Mm;

gn x (j) represents a new population of individuals, whose compilation has a length j, including the new individuals and generated on the basis of at least the population of individuals Gn (mn);

gn (i) denotes an individual having an nth algebraic i; and

gn (j) represents a new individual having a new individual group Gn (j) number j, and there are two and two methods for newly generating a new individual Gn (j +1) having a number (j +1) to be added, and generating a new individual group Gn (j + 1). [ production method A ]: in the generation method a, only the contemporary individual group gn (mn) is used as the mother set to generate a new individual group gn (j +1) as a subset from crossover or mutation. [ production method B ]: in the generation method B, all or part of the current generation population of individuals gn (mn) and the new population of individuals gn (j) are used as a mother set to generate a new population of individuals gn (j +1) as a subset from the crossover or mutation, wherein:

Gn(Mn)={gn(1),gn(2),…,gn(Mn)}；

Gn*(j)={gn*(1),gn*(2),…,gn*(j)}；

gn (j +1) = { Gn (1), Gn (2), …, Gn (j), Gn (j +1) } note that there is another method (hereinafter referred to as "generation method Ba") as a modification to generation method B in which only individuals suitable for the mother set are employed, which makes a part of the new individual group Gn (j) the mother set.

On the other hand, the selection method of the next-generation individual group may be classified according to whether or not the current-generation individual remains as the next-generation individual. Namely, wherein Gn (Mn) represents a new population of individuals, whose compiled length is Mn; gn +1(Mn +1) represents the next generation population of individuals, whose compilation length is Mn +1, and two methods a and B are described below. [ selection method A ]: selection method a is a method in which Mn +1 individuals (new individuals Gn +1(i) (i =1, …, Mn +1) selected under a certain threshold condition deviating from a new individual group Gn (Mn) are used as a next-generation individual group Gn +1(Mn + 1). with this method, a present-generation individual group Gn (Mn) never remains as a next-generation individual [ selection method B ]: selection method B is another method in which, a new individual Gn +1(i) (i =1, …, Mn +1) of Mn +1 individuals selected under a certain threshold condition deviating from all or part of the present-generation individual group Gn (Mn) and the new individual group Gn (Mn) is used as a next-generation individual group Gn +1(Mn +1), provided that:

Gn*(Mn*)={gn*(1),gn*(2),…,gn*(Mn*)}；

gn +1(Mn +1) = { Gn +1(1), Gn +1(2), …, Gn +1(Mn +1) }. Note that there is another method (hereinafter referred to as [ selection Ba ]) as a modification to [ selection B ], in which an individual that cannot be a parent set of the new individual group Gn (Mn) among the current individual group Gn (Mn) cannot be retained as a next-generation individual. Furthermore, there is another method (hereinafter referred to as "selection method Bb") as a modification to the selection method B, in which new individuals that are not suitable as a mother set of the new individual group Gn (Mn) cannot be retained as next-generation individuals. [ combined example of generation method and selection method ] (A) the optimal set search method is performed in combination of the generation method B and the selection method B, and the method is as follows:

the memory 11 is divided into two, i.e., a current generation aggregate group area and an additional recording aggregate group area. In this way, the generator 12 generates a new cluster using the current generation cluster and the additional record cluster. By means of the adding unit 15, a new set is selected among the new set group with a certain reference (including the case where all new sets are unconditionally selected) and a record is appended. On the other hand, as soon as the number of additional records reaches a predetermined value (for example, Ps-Pe +1), the deletion unit 16 selects a fixed number of sets among the current-generation cluster and the cluster group of additional records with a certain reference, and newly sets the cluster group as the current-generation cluster. These steps are then repeated.

In this combination, the function of [ generating method B ] is assigned to the generator 12 and the adding unit 15, and the function of [ selecting method B ] is assigned to the adding unit 15 and the deleting unit 16.

The first embodiment uses the generation method Ba as a modification to the [ generation method B ] and the selection method Bb as a modification to the [ selection method B ]. From the viewpoint that the adding unit 15 of the first embodiment additionally records only sets suitable as the parent sets in the new set and uses them as one of the parent sets for the next new set, it can be considered that the unit 15 and the generator 12 are responsible for the function of the [ generation method Ba ]. Meanwhile, the addition unit 15 and the deletion unit 16 take on the function of [ selection method Bb ] from the viewpoint that only sets suitable as mother sets are additionally recorded by the addition unit 15 of the first embodiment, and they are used as next-generation possible mother sets. (B) When [ generation method A ] and [ selection method A ] are combined, the method is as follows. The memory 11 is divided into two, i.e., a current generation cluster area and an additional recording cluster area. The generator 12 then generates a new cluster from the current generation of good clusters. The adding unit 15 selects and additionally records a plurality of sets using a reference of the new cluster. On the other hand, the deletion unit 16 deletes all the current generation cluster as soon as the number of additional records reaches a predetermined value (e.g., Pe), and the cluster group of additional records is shifted to update the generation. These steps are then repeated.

In this combination, the function of generating method a is assigned to the generator 12 in the first embodiment, and the function of [ selection method a ] is assigned to the adding unit 15 and the deleting unit 16. (C) When [ production method A ] and [ selection method B (or Bb) ] are combined, the method is as follows:

the memory 11 is divided into two, i.e., a current generation aggregate area and an additional recording aggregate area, and then the generator 12 generates a new aggregate group from the current generation aggregate group. The adding unit 15 newly selects a cluster using a reference of the new cluster and makes an additional record. On the other hand, as soon as the number of additional records reaches a predetermined value (for example, Ps-Pe +1), the deletion unit 16 selects one cluster from the current-generation good cluster and the good cluster of the additional records, and sets the cluster as the current-generation good cluster. These steps are then repeated.

Therefore, in such a combination, the function of [ generating method a ] is assigned to the generator 12 in the first embodiment, and the function of [ selecting method B (or Bb) ] is assigned to the adding unit 15 and the deleting unit 16. (D) When [ production method B (particularly Ba ] and [ selection method A ] are combined, the method is as follows:

the memory 11 is divided into two; i.e., into a current generation aggregate group area and an additional recording aggregate group area. The generator 12 then generates a new cluster from the current generation of good clusters and the good clusters of the additional records. The adding unit 15 selects a new good cluster based on a reference of the new cluster and makes an additional record. On the other hand, as soon as the number of additional records reaches a predetermined value (e.g., Pe), the deletion unit 16 deletes all the current-generation good cluster groups while moving the cluster group of additional records as it is to update the generation. These steps are then repeated.

With such a combination, the function of [ generation method B (particularly Ba) ] is assigned to the generator 12 and the addition unit 15 of the first embodiment, and the function of [ selection method a ] is assigned to the addition unit 15 and the deletion unit 16. [ Performance evaluation value ]

For reference, each performance estimation value used in the present embodiment is summarized as follows according to the realized apparatus and use: [19] performance estimation for best set judgment: VPS (1) to VPS (P)

The device comprises the following steps: an extractor 20.

The application is as follows: the best set is selected. [20] Additionally recording the judged performance estimation value; VPN

The device comprises the following steps: an adding unit 15.

The application is as follows: a new set of additional records is determined. [21] Deletion of the judged performance estimation value: VPD (1) to VPD (P)

The device comprises the following steps: and a deleting unit 16.

The application is as follows: a deletion for the set of records is determined. [22] Setting performance estimation values of additional reference values: VPE (1) to VPE (P).

The device comprises the following steps: a reference value modifier 18.

The application is as follows: reference is made when modifying the additional reference value BX. [23] Setting a performance estimation value of the mother set selection probability: VPS (1) to VPS (P).

The device comprises the following steps: a generator 12.

The application is as follows: a parent set is selected.

Wherein for each set its good set depends on various aspects. For example, it depends on how much the control objectives instructed by the group control device are met. At this time, a group control performance value directly related to the command control target (see fig. 7) may be used as the performance estimation value.

Regarding interleaving, the goodness like the parent set of each set depends on the variation in memory. This is because intersections composed of sets having different properties from each other can increase the probability of generating a better set.

An "assignment index" may be used as a performance estimation value to indicate its variation. With respect to each set at the center, the allocation index may be defined, for example, as the number of other sets such that the distance between the sets is equal to or less than a certain value. Wherein the distance between the sets is defined by a multi-dimensional space defined by the plurality of set forming elements. The assigned index represents the similarity between sets, and the lower the index value, the better the performance as a parent set.

The allocation index may be defined as the total distance from other sets. In this case, the larger the index value is, the better the performance as the parent set is. The allocation index may be defined as the number of other sets that are more than a predetermined value away from the set. In this case, the larger the index value is, the better the performance as the parent set is.

Next, a method of finding a performance estimation value is described in detail. In a first embodiment, any performance estimate includes an average latency AWT. However, it is possible to change the content of each performance evaluation value according to its use. For example, each of the performance estimation values E may be calculated using mutually different performance estimation functions.

In summary, the performance estimation function related to group control performance is represented by the following equation, where F (X) is a function of X.

E=F(X1,X2,...,Xi,...,Xn,T1,T2,...,Ti,...,Tn)

Wherein,

n: the number of estimated terms of the group control performance,

xi: estimating a performance value of the term i (i =1, 2...., n),

ti: a performance reference value for term i (i =1, 2...., n) is estimated.

The performance reference value Ti represents a [ target value ] or a [ limit value ] that must be met as the group control performance will eventually reach. [ limiting value ] includes [ upper limit value ] and [ lower limit value ]. Regardless of whether the performance reference value is given as [ target value ] or [ upper limit value ] or [ lower limit value ], different determinations are made depending on the purpose of group control to be set.

Note that, when the performance reference value Ti means a "target value" relating to the performance estimation function [24], the smaller the | Xi-Ti | becomes, the better the performance will become. When the performance reference value Ti refers to [ upper limit value ], (Ti-Xi) becomes larger, the performance will become better. When the performance reference value Ti refers to [ lower limit value ], (Xi-Ti) becomes larger, the performance will become better.

In summary, the performance estimation values shown by the functions [19] to [23] depend on the performance estimation function, the estimation terms therein, and the performance reference value.

The following are specific examples to illustrate the method of performance estimation. [25] First example performance estimation function (in the first embodiment): a control target; [ average waiting time is reduced as much as possible ]. Performance reference value: t1=0 seconds. (T1: target value of average waiting time). Performance estimation function: e = | AWT-T1 | AWT (AWT: average waiting time). Additional recording judgment: e < BX (BX: additional reference value, e.g., BX =15 seconds). [26] Second example Performance estimation function (in the case of the second example below): the control target: [ average latency as close as possible to a predetermined value ]. performance reference value: t1=20 seconds. (T1: [ target value of average waiting time ]). Performance estimation function: e = | AWT-T1 | (AWT: average latency). Additional recording judgment: e < BX (BX: additional reference value, e.g., BX =3 seconds). [27] Third example performance estimation function: the control target: [ average waiting time AWT, long waiting rate RLW, and prediction error rate RPE are made as close to respective target values as possible ]. Performance reference value: t1=15 seconds, T2= 2%, T3= 3% (where T1: the [ target value ] of the average waiting time, T2: the [ target value ] of the long waiting rate, T3: the [ target value ] of the predicted error rate, [ target value ] performance estimation function E = Apx | AWT-1 | Ag x | RLW-T2 | Ar x | PRE-T3 | (where Ap, Ag and Ar are weighting coefficients), additional recorded judgment E < BY (BY: total estimated reference value, e.g., BY =10) & [28] fourth example performance estimation function: [ usually make the average waiting time AWT, long waiting rate RLW and predicted RPE possibly small ] & performance reference value T1=0 seconds, T2= 0%, T3= 0% (where T3: the [ target value ] of the average waiting time = 2] and T5842 × AP × T × ATR ] of the predicted error rate [ AP + AP × AW ] target value of the predicted error rate [ APW ] & AP × 73742 ] & AP + AP × AW [ target value ] of the predicted error rate ], where T6342 [ APE + AR ] is estimated target value, Ag and Ar are weighting coefficients). Additional recording judgment: e < BY (BY: total estimated reference, e.g., BY = 1000). [29] Fifth example performance estimation function: the control target: the number of estimation terms is such that the average wait time AWT, the long wait rate RLW, and the prediction error rate RPE remain as much as possible within their respective allowable increase ranges ]. performance reference values: t1a =15 seconds, T1b =3 seconds, T2= 2%, T3= 3% (where T1 a: the [ target value of the average waiting time ], T1 b: the allowable deviation range of the average waiting time, T2: [ upper limit value of the long-term waiting rate ], T3: the [ upper limit value of the predicted error rate ]. performance estimation function E = f (| AWT-1la | -T1b) + f (RLW-T2) + f (RPE-T3) (where f (X) represents a function, i.e. f (X) =1, X ≧ 0 and f (X) =0. X < 0.) is additionally recorded judging E < BY (BY: the total estimated reference value, on the side e.g. BY =1) [30] sixth example performance estimation function. control the target [ keeping the Average Waiting Time (AWT) from its minimum value (AWT =3 seconds) and reducing the allowable long-term performance estimation reference value (BV 1 a) as much as possible, t1b =2 seconds, T2= 2% (where T1 a: the [ target value ] of the average waiting time, T1 b: the allowable deviation range of the average waiting time, T2: the [ target value ] of the long-term waiting rate). Performance estimation function: e = (100-RLW) × f (T1b- | AWT-BVPE |) (where f (X) represents a function, i.e., (X) =1, X ≧ 0 and f (X) =0, X < 0). Optimum value determination: max { E }.

The first and second example performance estimation functions [25], [26] are applicable to a case where the estimation term is only one average latency and the control target thereof is also simple, and therefore, the estimation functions can be generated relatively easily. It is very easy to replace the average waiting time with the estimation term shown in fig. 7 (of course, even if the estimation term exceeds that shown in fig. 7).

On the other hand, for a plurality of estimation terms, the performance estimation function becomes complicated as the control target becomes variable. When the third and fourth performance estimation functions [27] and [28] are employed, the performance estimation functions are weighted in combination with the deviation between the estimate and the target value. Among them, a method of estimating the target achievement degree in general in relation to the priorities of the estimation items is generally known. This is a very convenient method, especially when the control targets of the estimation terms are opposite to each other.

As shown in the fifth performance estimation function [29], the number of estimation items in which the group control performance value falls within the judgment permission range (for example, equal to or lower than the upper limit value, equal to or higher than the lower limit value, or the deviation from the target value remains within a certain value) is looked up according to each estimation item, and the group control performance can be estimated according to the number.

Further, each of the estimated values is calculated based on two or more different performance estimation functions, respectively, and additional recording judgment, deletion judgment, optimal set judgment, and the like are performed based on a combination of the plurality of estimated values.

As shown in the sixth performance estimation function [30], where there are two conditions that the average latency is maintained within a predetermined allowed range from a minimum value (E1= | AWT-BVPE |, E1 ≧ T1b) where the long-term latency is at a minimum (E2= RLW, Min { E2}), the best value of the candidate can be selected first by the performance estimation function E1 and then the best value can be selected finally by the performance estimation function E2. As shown in the sixth performance estimation function [30], two sets of performance estimation functions and two conditions may be combined, overwriting one performance estimation function and one judgment condition, to select the best set using the performance estimation function and the judgment condition.

In the first embodiment, although the selection probability of the parent set depends on the performance estimation value, the selection probability may be halved. In the case of the first embodiment, new sets having similar properties tend to be generated, and therefore, the change of the sets stored in the memory 11 disappears. When this change disappears, there may be a convergence problem of the local solution even in the initial stage of the search. Therefore, in the case where the problem of initial convergence must be avoided or the amount of calculation must be reduced, the selection probability of the supplied parent set can be halved. [ second embodiment ]

Referring to fig. 21 to 25, a second embodiment will be described below. In describing the second embodiment, the portions thereof different from the first embodiment are mainly described.

Fig. 21 is a view corresponding to fig. 1 showing a second embodiment. In fig. 21, the performance reference value setting means 3 includes a personal computer which provides a reference value signal 3a to the group control apparatus 1. The reference value signal 3a comprises a [ performance reference value ] for the performance of the group control and a [ control reference value ] for the control of the search means. In the present embodiment, the performance reference value is [ target value ] of the average waiting time; the control reference value is the [ specified value ] supplied to the additional reference value BX. The reference value signal 3a can be fed directly to the look-up means 10 (estimator 13, reference value updating unit 18, re-look-up unit 19 and initialization unit 21).

Fig. 22 corresponds to fig. 3 and shows the memory contents of the RAM 10A. In fig. 22, TGT is one of the data items forming the content of the seek condition signal 1a, and it includes data TAW (wait time target value) indicating [ target value ] of the average wait time AWT, and data TCB (additional reference specified value) indicating a specified value of the additional reference value BX. For example, the waiting time target value TAW is set at 5 seconds, and the additional reference specification value TCB is set at 3 seconds. TAWX is data transcribed from the latency target value TAW to calculate a performance estimate.

When the reference value signal 3a is input, the group control device 1 takes out the performance reference value (the waiting time target value TAW) and the control reference value (the additional reference designation value TCB) contained in the reference value signal 3 a. As with the operation of step 234 in fig. 10, the group control device 1 feeds a search condition signal 1a to the search device 10, which includes elevator car specification data ELS, traffic flow specification data TRS, search instruction data SCM, a waiting time target value TAW, and an additional reference designation value TCB. If the search process starts (see fig. 11), the generation, estimation, addition, deletion, and additional reference value modification are sequentially completed in the search apparatus 10, which is similar to the first embodiment. When the lookup condition signal 1a is input to the lookup device 10 according to the lookup start judgment program 26 (see fig. 25), the RAM10C stores the elevator car specification data ELS, the traffic flow specification data TRS, the lookup instruction data SCM, the waiting time target value TAW, and the additional reference designation value TCB, as shown in fig. 22.

In the second embodiment, the addition program 34 and the deletion program 35 have special features. In the second embodiment, the performance estimation value is calculated based on the performance estimation function (E = | AWT-T1 |) shown by the second performance estimation function [26], and the performance estimation value E and the additional reference value BX. are compared with each other, if the performance estimation value E is smaller than the additional reference value BX, to determine an additional record.

At step 352 (see fig. 17) of the deletion program 35, calculation [ VPD (p) < | awt (p) -TAWX | is performed, thereby setting performance estimation values VPD (1) to VPD (p) for the deletion judgment.

Fig. 23 shows the contents of an additional reference value modification program 36 of the second embodiment.

In fig. 23, the additional reference designation value TCB read out at step 40 is substituted into the modification value CBX. In a next step 402, the additional reference value BX is overwritten with the modified value CBX. In step 403, the latency target value TAW is read out, and the performance reference value (latency target value TAWX) currently in use is overwritten with this value.

When the modification of the additional reference value is completed, the judgment of the end of the search process and the extraction of the optimum set (see fig. 11) are performed as in the first embodiment, assuming that the calculation method of the performance estimation value is made different from that in the first embodiment in the extraction program 38 of the optimum program. That is, at step 381 (see fig. 20) of the optimal set extraction program 38, the performance estimation values VPS (1) to VPS (p) [ (-) AWT (1) -TAWX | are calculated for the optimal set judgment in such a manner that [ VPS (1) < - ] | AWT (1) -TAWX | and VPS (p) < - ] | AWT (p) -TAWX | are calculated.

After the best set extractor 38 is processed, the process proceeds to the step of the re-lookup decision routine 25. Referring to fig. 24, the operation steps of the search start judgment program 26 in the re-search judgment program 25 will be described below. Note that fig. 24 corresponds to fig. 12 of the first embodiment.

In fig. 24, in step 261, the group control device 1 provides the lookup condition signal 1A, and then the RAM10C stores the elevator car specification data ELS, the traffic flow specification data TRS, the lookup command data SCM, the waiting time target value TAW, and the additional reference designation value TCB. Then, similarly to the search start judgment program 26 of the first embodiment (see fig. 12), in steps 262 to 264, the search instruction data SCM is changed from [0] to [1], and the elevator car specification data ELS or the traffic flow specification data TRS is detected. In step 265, when at least one change is detected, the search start flag STR is set to [1] and instructs the start of the search process after being initialized in the first way described below.

On the other hand, if any of the search command data SCM, the elevator car specification data ELS, and the traffic flow specification data TRS does not indicate a change, it is determined in step 267 whether or not the waiting time target value TAW has changed. That is, the currently set waiting time target value TAWX and the waiting time target value TAW are compared with each other to judge any change. If TAW is not equal to TAWX, then noted as [ changed ], and then the search start flag STR is placed in [1] at step 265. Otherwise, if TAW is equal to TAWX, then it is determined as [ unchanged ], and the operation of step 269 is performed.

At step 269, it is determined whether the additional reference specification value TCB has changed. That is, the additional reference designation value TCB and the currently set additional reference value BX are compared with each other to determine that there is any change. If BX is equal to TCB, it is determined as unchanged, and then the search start flag STR is set to "0" in step 266, whereby the continuation state is maintained if the best set is being searched, and the completion state is maintained if the search process has been completed. If, on the other hand, BX is not equal to TCB, then it is determined as changed, and in step 268 the finding means 10 sets the finding start flag STR to 2, and then instructs the start of the finding process after being initialized by the second way. The initialization in the first manner includes initialization of the memory and total initialization (initialization of the estimation number NE, the additional recording number NR, and the like). Initialization in the second manner refers to the overall initialization process except for the memory.

The reason why the initialization by the second way is performed when the additional reference value BX is changed at step 269 is that the search process is continued within the scope of the previous search process without any problem even if the additional reference value BX is changed to a more restricted (i.e. smaller) value. From a convergence point of view, a better search efficiency can be obtained when the search process is restarted with the generation and selection set used as the initial set. In particular, the initialization of the second approach is more appropriate when the performance estimation function contains only a single estimation term and only the additional reference value BX is changed.

Now, assume that the search start flag STR is set to [0], so that the determination result becomes whether to leave the best set in the current state (continuation state or end state) or to restart from the beginning. In this case, the process proceeds from step 27 in fig. 11 to step 30, and determines whether the search process is in progress or ends based on the value of the search permission FLAG. If the search process is in progress, FLAG is equal to [1], the process proceeds to step 31, whereupon at search end decision routine 37, the search enable FLAG is reset to [0], and the operations of steps 31 through 38,26 and 27 are repeated until it is determined at step 27 that the search has ended.

If the search is completed, the search end decision routine 37 sets the search permission FLAG to [1], thus causing the process to wait for a re-search by repeating steps 26, 27, 30, 26 in sequence.

If the elevator car specification data ELS, the traffic flow specification data TRS, the performance reference value TAW, or the control reference value TCB is changed during or after the search, the search start flag STR is set to [1] or [2] by the search start judgment routine 26, and the process proceeds from step 27 to the initialization routine 28.

Fig. 25 shows the initialization program 28.

Fig. 25 corresponds to fig. 13 of the first embodiment. Referring to fig. 25, it is judged in step 284 whether initialization of the memory is necessary or not based on the value of the search start flag STR. If the initialization in the first manner is specified, i.e., if the search start flag STR is set to [1], the initial cluster and group control performance data corresponding to the traffic flow specification data TRS are read out at step 281, similarly to the initialization program 28 (see fig. 13) of the first embodiment. Initialization is then performed using the initial cluster and cluster control performance data at step 282.

In step 285, the additional reference value modification procedure is initiated. The modification procedure of the additional reference value of step 285 has the same contents as those of procedure 36 shown in fig. 23. A new additional reference value (BX) and a new performance reference value (latency target value TAWX) are set at this step 285. Then, the total initialization is also performed in step 283 like the initialization program 28 (see fig. 13) of the first embodiment. This is the initialization in the first way.

On the other hand, when it is judged in step 284 that the search start flag STR is "2", that is, when the initialization by the second manner is specified, only the overall initialization of step 283 is performed. That is, the cluster group recorded in the memory before the search is started is used as the initial cluster group without performing the operations of steps 281 and 282. In particular, when the last search is finished, the good sets EPS (1) to EPS (p) are reserved in the memory 11 and their group control capability data PRE (1) to PRE (p) are used. This is the initialization in the second way.

As described above, in the second embodiment, with the performance reference value setting means 3, the target value TAWX of the average waiting time and the specified value BX of the additional reference value can be supplied from the outside. Thus, an optimal set that meets the required policy can be found for group control or lookup.

In the second embodiment, a change in the reference value TGT (the latency target value TAW and the additional reference value BX) is detected during or after the end of the seek, whereby a re-seek can be performed. Therefore, when the control strategy of the group control is changed manually or the reference value TGT is changed manually, the re-search is automatically performed, so that the optimal set can be obtained quickly.

In a second embodiment, the initialization of the first or second mode may be selected based on the circumstances at the beginning of the search, so that appropriate initialization may be performed. For example, when the additional reference value BX is slightly modified, the second way of initialization can be used to quickly perform a re-lookup. When the additional reference value BX changes significantly, the initialization in the first way is naturally required. In contrast, in the second embodiment, although the initialization in the first manner is employed when a change in the performance estimation function (estimation term, performance reference value, structure) is detected, the initialization in the second manner may be employed when the change is small.

In short, the choice of whether to initialize in the first or second manner creates the problem of which cluster, i.e., the one obtained at the time and the one initialized by the GPS0, can be used to perform an efficient search under the new search conditions. However, such a determination requires considerable computation time and is therefore considered impractical. Therefore, it is necessary to select a mode according to the change item and the change amount.

The above performance estimation function is an example, and other functions such as the performance estimation function [24] or the first to sixth performance estimation functions ([25] - [30]) and the like may be used.

The performance estimation value is not only a performance estimation value for additional recording judgment, but also a performance estimation value [19] for optimal set judgment, a performance estimation value [21] for deletion judgment, a performance estimation value [22] for reference value judgment, or the like. Third embodiment (another embodiment of the deletion unit)

Another embodiment of the deletion unit 16 will be described below with reference to fig. 26, and the description will be mainly directed to portions thereof different 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, a formula NRH = NR-NRX is calculated to calculate a new number of records NRH, which is an additional number of records for a new set that survives the last deletion process. The number of records NRX at the time of the last judgment is a value represented by the additional number of records NR when the last deletion process is performed.

Next, in step 412, the new record count NRH and the deletion start judgment value NRa are compared to judge whether or not it is time to delete. The deletion start determination value NRa is set to 10 times, for example.

If NRH < NRa, it is determined not to be the time to delete the collection and the process exits the routine 35 directly. On the other hand, if NRH ≧ NRa, the deletion is judged, and then the additional record number NR at that time is updated by being added to the number NRX at the time of the final judgment.

Then, the set is deleted by repeating steps 414 to 422 until the number of units P reaches the deletion end judgment value Pe. At step 283 (see FIG. 13) of the initialization program 28, the initial value of NRX is initialized to [0] (not shown).

In step 414, the distance DST (i, j) between the sets is calculated (where i, j =1, 2. The relationship of the distance DST (i, j) with respect to the two sets (i, j) is as follows.

DST (i, j) = | epu (i) -epu (j) | 31 (where i, j =1, 2. ·, P; i ≠ j).

Provided that the parameter values eps (i) of each set, (i =1, 2., P) have been normalized with respect to each parameter. I.e. each parameter value is expressed as a ratio to the possible maximum value derived from each parameter, which value is between 0 and 100.

For example, if the possible maximum value taken by the full-state estimation coefficient Ca as one of the parameters is assumed to be 50,000, the normalized parameter value regarding the full-state estimation coefficient Ca is calculated as 20({ = (10,000 ÷ 50,000) × 100} in the set EPS (1) shown in fig. 8, and likewise, when the parameter value Cb (maximum value: 1,600) of the prediction error coefficient is normalized, it is calculated as (400 ÷ 1,600) × 100= 25.

Since there are 25 control parameters in total, DST (i, j) will be 0 ≦ DST (i, j) ≦ 500. When the best set data BPD is generated at step 387 (see fig. 20) of the best set extractor 38, each normalized parameter value is reconverted to the original value. (for example, in the case of the full state estimation coefficient Ca, it is converted again to 20 × 50,000 ÷ 100=10,000).

In step 415, the pair of sets Pd1, Pd2 forming the shortest distance DST (i, j) is selected. Then, in step 416, it is determined whether the characteristics of the Pd1, Pd2 in the set pair are the same. That is, the distance DST (Pd1, Pd2) and the determination value DSTa are compared, and the identity is determined based on the comparison result. The determination value DSTa is set to 25, for example.

If DST (Pd1, Pd2) ≦ DSTa, the process proceeds to step 417, where the average wait time AWT (Pd1), AWT (Pd2) is retrieved from the group control performance data PRE (Pd1), PRE (Pd2) of the two sets Pd1, Pd2, respectively, and then set as the deleted performance estimates VPD1, VPD 2.

In step 418, the performance estimates VPD1 and VPD2 are compared to determine the set to be deleted. If VPD1 < VPD2, the number of deleted sets is determined to be Pd2 sets, and the process proceeds to step 419. Then, the records of the set EPS (Pd2) and the group control performance data PRE (Pd2) are deleted. Then, the value of the number of recorded units P is decremented by 1. The number of sets is reassigned to the remaining sets and the process ends at step 419.

If VPD1 ≧ VPD2 at step 418, a determination is made that one of the sets Pd1 was deleted, the process proceeds to step 420. Then, the records of the set EPS (Pd1) and the group control performance data PRE (Pd1) are deleted. Then, the value of the number of recorded units P is decremented by 1. The remaining sets are renumbered and the process ends at step 420.

In step 416, if DST (Pd1, Pd2) > DSTa, it is judged that their characteristics are not the same, and then the procedure proceeds to step 421, where performance estimation values VPD (1) to VPD (P) for deletion judgment are found with respect to all the unit numbers 1 to P, from which a set having the worst value is specified, and then the set number thereof is set to Pd 1. This step 421 is similar to steps 352 to 357 of the deletion program 35 (see fig. 17) in the first embodiment, and thus a description thereof will be omitted. The number of deleted sets RP in fig. 17 corresponds to the number of sets Pd 1.

In step 422, it is determined whether the number of sets P after deletion becomes the deletion end determination value Pe or smaller than Pe. If not, the process repeats steps 414 to 422, and ends the execution of the deletion program 35 when P becomes equal to or smaller than Pe.

As described above, in the third embodiment, a pair of sets having the same characteristics is specified in accordance with the distance DST between the sets, and one set of the pair of sets is deleted, so that a plurality of sets having mutually different characteristics can be retained in the memory 11, thereby ensuring the change of the memory 11. When a pair of sets is selected, a pair of sets having the best similarity (shortest distance) has a priority, so that a plurality of sets having mutually different characteristics can be retained.

In the third embodiment, when one of the sets in the set pair is deleted, the set of better performance estimation values is retained, and the set of worse performance estimation values is deleted, so that the group control performance of the sets stored in the memory 11 can be maintained all at a high level. It should be noted that in the third embodiment, when there is no pruning of the same pair sets, the set of worst performance evaluation values will be pruned out in order. Therefore, the unnecessary sets continue to be deleted until the deletion end determination value is Pe. Fourth embodiment (another embodiment of the search end judgment)

Referring to fig. 19, another embodiment of the seek end determination unit will be described below. In this fourth embodiment, the portions different from the first embodiment are mainly described.

In step 371 of fig. 19, the estimated number NE is replaced with the additional recording number NR and the seek end judgment value NEa is replaced with the seek end judgment value NRb. In particular, the number of additional recording times NR is set to, for example, 200.

That is, if NR < NRb, the search permission FLAG is set to [1] in step 372, and the search process is continued, and if NR ≧ NRb, the search permission FLAG is set to [0] in step 373, and the search is ended.

In general, when the number of times of search NE becomes a large value and when the search converges to some extent, the ratio of the number of additional recordings to the number of times of search tends to decrease. Therefore, in order to complete the lookup efficiently, the end of the lookup must be judged in consideration of the result of the lookup run, i.e., the number of sets of additional records. Therefore, in this fourth embodiment, the search process is continued until the number of additional recording times NR (indicating the number of additional recording times) reaches the search end judgment value NRb.

As described above, according to the fourth embodiment, the system can avoid terminating its search process before it is effectively completed. Fifth embodiment (another embodiment for judgment of end of search)

Referring to fig. 27, another embodiment of the seek end determination unit will be described below. In this fifth embodiment, portions different from the first embodiment will be mainly described.

Fig. 27 shows a seek end judgment routine 37 of the fifth embodiment, which corresponds to fig. 19 of the first embodiment.

At step 431, the number of elapsed estimations NEH is calculated according to formula NEH = NE + NEX. The elapsed number of estimations NEH is the new number of estimations after the last termination judgment. The estimated number of times NEX at the time of the last end judgment represents the estimated number of times NE at the time of the judgment to the end in the last cycle. Step 432 determines whether the estimated number of times after the last time is equal to or greater than a certain number of times. If the estimated number of times NEH has been experienced is less than a certain value NEb, the search permission FLAG is set to "1" in step 433, and the search is continued. Specifically, the estimated number of times NEH elapsed is set to 20, for example.

If step 432 determines that the estimated number of passes NEH is equal to or greater than a certain value NEb, the process proceeds to step 434 and a success index RSC is calculated. First, the new recording number NRH is calculated according to the formula NRH = NR-NRX. The new recording number NRH indicates the number of times a new collection is additionally recorded after the last end judgment is completed. The success index is then calculated according to the formula RSC = NRH ÷ NEH. The number of times of recording NRX at the last judgment represents a value of the additional number of times of recording NR calculated when the end judgment has been completed in the last cycle.

In the next step 435, the estimated number NEX and the number of times of recording NRX at the time of the last judgment are updated based on the estimated number NE and the additional number of times of recording NR at that time. In step 436, it is determined whether to end the search according to the estimated number NE, the search final stage judgment value NEC, the success index RSC, and the search end judgment value RSCa. Wherein the find final stage decision value NEC is, for example, set at 600. The search end determination value RSCa is set to 0.05, for example.

If NE < NEC or RSC is greater than or equal to RSCa, then the search is not completed effectively, the search permission FLAG is set to [1], and the search is continued in step 433. If NE ≧ NEC and RSC < RSCA, it is judged that the search has been completed efficiently, and in step 437 the search permission FLAG is set to [0], the search is stopped.

It should be noted that step 436 contains a reason for the condition on the number of estimations NE, in order to prevent the search process from ending up without enough estimations due to the reduced success index RSC, which may be due to the size GP-SO of the initial set, the crossing rate CR and the variation rate MR in the initial phase of the search. If such a problem does not occur, the condition on the number of estimations NE will no longer be necessary as a judgment condition on the end of the search, and thus it is sufficient to use even the condition on the successful indexing RSC.

As described above, in the fifth embodiment, the end of the search is judged based on the success index RSC obtained by the number of estimations and the number of additional recordings, whereby it is possible to judge with high accuracy whether the search is sufficiently converged. Thus, the best set can be found efficiently without meaningless repeated lookups.

Further, in the fifth embodiment, the initial stage of the search during which the search process is not ended even if the successful index RSC becomes smaller than the search end judgment value RSCa each is detected by the estimated number of times NE, so that the search process will not be ended without a sufficient number of estimations. Sixth embodiment (another embodiment of search end judgment)

Referring to fig. 28, another embodiment of the seek junction determining unit 17 will be described below. In the sixth embodiment, portions different from the first embodiment are mainly described.

Fig. 28 shows a seek end judgment routine 37 of the sixth embodiment, which corresponds to fig. 19 of the first embodiment.

Step 451 calculates the distance DST (i, j) between the sets. The distance DST (i, j) is calculated according to the above equation [31 ]. Note that this calculation is almost the same as the step 414 of the deletion program 35 in the third embodiment.

In step 452, a similar number of sets NDST, i.e. the number of sets whose DST (i, j) ≦ DSTa, is calculated based on the distance DST (i, j) calculated above. DSTa is a judgment value of whether or not a judgment set is the same, and in the sixth embodiment, it is set to 25 as in the third embodiment. In the next step 453, the search end determination value NDSTa is calculated. In summary, when the search is converging, many sets have a tendency to converge around the optimal set. The same number of sets NDST is used as an index to detect such a trend, and convergence of the search depends on the ratio of the same number of sets NDST to the total number of sets clustered. If the determination threshold value is set at 80% with respect to the ratio, the search end determination value NDSTa is calculated from NDSTa = { PX (P-1) ÷ 2} × 0.8 because the entire combination number becomes { PX (P-1) ÷ 2} when the number of sets that have been recorded is P.

At this time, if the number of sets to which records have been added reaches the maximum value Pmax, although unnecessary sets will be deleted at a certain reference value, the same number of sets NDST will change depending on how unnecessary sets are deleted. Also, the same number of sets NDST may be changed according to the method of generating the sets or the additional records. Therefore, the judgment threshold is not limited to 80%, and must be appropriately modified depending on other factors.

In step 454, it is determined whether the search is finished according to the same set number NDST and the search end determination value NDSTa. If NDST is less than NDSTa, judging that the search is still not fully completed; setting a search permission FLAG to [1] in step 455, and continuing the search; and terminates the execution of the seek end determination program 37. If the NDST is more than or equal to NDSTa, judging that the search is fully executed; setting a search permission FLAG to [0] in step 456, and stopping searching; and terminates the execution of the seek end determination program 37.

As described above, in the sixth embodiment, the end of the search is judged based on the distance DST between sets, and thus, the convergence of the search can be detected with high accuracy. Therefore, the search does not need to be repeated meaningfully, whereby the search process can be performed efficiently. The search end determination condition is not converged by those conditions described above, and other conditions may be employed. (seventh embodiment (another embodiment of optimal set extraction))

Referring to fig. 29, another embodiment of the extractor 20 will be described below. In the seventh embodiment, the portions different from the first embodiment are mainly described.

Fig. 29 shows a best set extractor 38 of the seventh embodiment, which corresponds to fig. 20 of the first embodiment.

In step 471, the average waiting times AWT (1) to AWT (P) are taken from the group control performance data PRE (1) to PRE (P), respectively, and then set as the first performance estimation values VPS1(1) to VPS1 (P). In step 472, the long latency times RLW (1) to RLW (P) are retrieved from the group control performance data PRE (1) to PRE (P), respectively, and then set as the second performance estimation values VPS2(1) to VPS2 (P). In step 473, the minimum values of the first performance estimation values VPS1(1) to VPS1(P) are found and set as the best values BVPE.

At step 474, the best set BP is selected based on the performance estimates. I.e. to find sets i from memory 11 whose values (VPS1(i) -BVPE) are equal to or less than BZ. Then, the set having the smallest second reference estimate VPS2(i) therein is selected as the optimal set BP. I.e. the two-step selection is completed. BZ is a reference value indicating an allowable range of deviation from the best value BVPE, and is set to two seconds in the present embodiment.

Then, in a similar manner to the first embodiment, the best set data BPD is generated and sent to the group control device 1 in the next step 388.

As described above, in the seventh embodiment, two-step selection is employed to extract the optimal set. In the case where some priority is judged by two estimation items, since two-step selection is applied to extraction, the best set can be extracted based on such priority. Of course, three or more step selections may be used. The seventh embodiment is equivalent in meaning to the contents of the sixth performance estimation function [30] described above. Eighth embodiment (another calculation method of group control performance value)

As described in Japanese patent laid-open publication (KOKAI) No. 57-57,168, a new set of group control performance can be obtained by replacing the simulator 2 of the first embodiment with an actual group control device 1. This is described below with reference to fig. 30 and 31.

Fig. 30 shows a system of the eighth embodiment, which corresponds to fig. 1 of the first embodiment. Fig. 31 shows the operation of the group control device 1, which corresponds to fig. 9 of the first embodiment. An eighth embodiment will be described below, in which portions different from the first embodiment are mainly described.

Referring to fig. 30, when a seek is commanded according to the seek condition signal 1a input to the seek means 10 by the group control means 1, the seek means 10 executes an estimation procedure (see fig. 15) at step 33 of the operation procedure, generates simulated condition data similar to the first embodiment and outputs the simulated condition signal 13 a.

However, as shown in fig. 30, the signal 13a is fed to the group control device 1 in the present embodiment. Upon receiving the signal 13a, the group control device 1 enters the commissioning mode. This operation will be described in detail below using the flowchart shown in fig. 31.

In fig. 31, step 491 judges whether or not it is a test run, and step 492 judges whether or not the test run is started. When the commissioning flag FLG is "0" and when the signal 13a does not contain a designation to start commissioning, the normal group control operation is performed according to steps 221 to 229.

On the other hand, when the start commissioning mode is detected from the content of the signal 13a at step 492, the commissioning flag FLG is set to [1] at step 493, and the currently used temporary escape code is set by the parameter value at step 494. These escape codes are replaced by writing the estimated set (new set) contained in the signal 13 a.

Then, in the conventional group control operation, the group control operation is performed in steps 221 to 229. During the commissioning, with the help of steps 491 to 495 after the commissioning flag FLG is [1], the group control operation is performed with steps 221 to 229.

When the group control operation has been performed for a while (for example, one hour), the group control device 1 detects the end of the trial operation at step 495, resets the trial operation flag FLG to [0] at step 496, and returns the parameter set that has been escape at step 497, and at the same time, calculates the group control performance data PRF (such as average waiting time, long waiting time) related to the trial operation. In step 498 the group control performance data PRF are fed to the look-up means 10 as a group control performance value signal 2 a. Then, the group control apparatus 1 returns to the normal state, and completes the group control operation through steps 221 to 229.

As described above, when the group control performance value signal 2a is obtained by the trial operation, the search device 10 shown in fig. 30 obtains the performance estimation value VPN from the group control performance value signal 2a, and compares the performance estimation value VPN with the estimation reference value BX to determine whether to additionally record a new set.

As described above, in the eighth embodiment, the new set is estimated in the actual device, so that since the time to obtain the optimum set tends to become longer, although not appropriate, the simulator 2 is not necessary, thereby reducing the system cost. Ninth embodiment (another embodiment of generating a new set)

Although in the first embodiment, the crossover rate CR and the variation rate MR are fixed, the ninth embodiment is characterized in that the crossover rate CR and the variation rate MR can be modified according to the search case.

Referring to fig. 32 to 34, a ninth embodiment will be described below. The portions different from the first or second embodiment will be mainly described.

Fig. 32 shows the overall structure of the ninth embodiment, which corresponds to fig. 1 of the first embodiment. In fig. 32, the occurrence modifier 4 modifies the crossover rate CR and the variation rate MR (selection rates of various generation modes) according to the search situation.

Fig. 33 shows an operation procedure of the ninth embodiment, which corresponds to fig. 11 of the first embodiment. Note that it is basically the same as the operation program 100 shown in fig. 11, except that an appearance rate modification program corresponding to the function of the appearance rate modifier 4 shown in fig. 32 is added to step 50.

Fig. 34 shows the occurrence rate modification routine. At step 501, the number of elapsed estimations NEN, which represents the number of estimations to be newly performed after the last end judgment, is calculated according to the formula NEH = NE-NEX. The same value of the estimated number NE at the time of the last end judgment performed in the previous cycle is set as the estimated number NEX at the last judgment. Step 502 determines whether the estimation has been performed a certain number of times or more since the previous cycle. If the estimated number of passes NEH is less than a certain value NEb (e.g., 20), the occurrence modification procedure ends.

On the other hand, if it is judged in step 502 that the estimated number of times NEH has elapsed is equal to or greater than a certain value NEb, the routine proceeds to step 503, where the success index RSC is calculated.

First, the number of times NRH recording representing the number of times of new sets has been additionally recorded since the last end judgment is calculated according to the formula NRH = NR-NRX. The success index RSC is calculated according to the formula RSC = NRH ÷ NEN. The same number of additional recordings NR at the time of the last end judgment of the previous loop is set as the number of recordings NRX at the last judgment. In the next step 504, the estimated number of times NEX at the last judgment and the number of times NRX at the last judgment are updated based on the estimated number of times NE and the additional number of times NR at that time.

In steps 505 to 510, the crossover rate CR and the variation rate MR are modified according to the estimated number NE, the first judgment value NEd1, the second judgment value NEd2, the success index RSC, the success rate judgment value RSCb, and the success rate judgment value RSCc. For example, NEd1= 500; NEd2= 800; RSCb = 0.10; RSCc = 0.05.

If NE ≦ NEd1 and RSC ≦ RSCB, or if the success index was in a lower state than expected during the first half of the search, then it is determined that the currently set crossover rate CR and the variance rate MR are not appropriate, and the process proceeds to steps 505, 507, and 509, whereby the crossover rate CR is slightly decreased from the current value (e.g., decreased by 0.001) and the variance rate MR is slightly increased from the current value (e.g., increased by 0.001). Furthermore, in step 509, the cross-over rate CR is set to be slightly greater than the current value (e.g., 0.001) and the variability rate MR is set to be slightly less than the current value (e.g., 0.001) in the opposite manner to the above modification.

In fact, another method may be used to clear such a reduced success index. That is, if the crossover rate is too large to be used, the crossover rate CR is made small and the variation rate MR is made large. Conversely, if the crossing rate is too small to be used, the crossing rate CR is made larger, and the variation rate MR is made smaller.

In summary, if the search is in a reduced state, the ratio of the selection probabilities for each generation method is changed to clear such a reduced state.

If NE ≧ NEd2 and RSC ≦ RSCC, or if the successful index is in a lower state than expected at the post-lookup stage, it is judged that the good set lookup tends to converge, and the process proceeds to steps 505, 506, 508, and 510 in order, thereby causing the crossover rate CR to increase slightly (e.g., by 0.001) based on the current value and the mutation rate MR to decrease slightly (e.g., by 0.001) based on the current value.

If the above conditions are not met, the currently set crossover rate CR and the variance rate MR are both judged to be appropriate, and the occurrence rate modification procedure is terminated without modifying the rates.

As described above, in the ninth embodiment, the progress of the search is judged based on the estimated number NE and the success index RSC, and the crossover rate CR and the variation rate MR can be modified based on the judgment. Therefore, good sets can be found earlier than systems with fixed crossover rate CR and fixed variance rate MR, and the search time can be shortened. As a result, the system may improve lookup efficiency.

In particular, in the ninth embodiment, if the success index becomes lower than the expected value in the initial stage (or first half) of the search, the process sets the crossover rate CR to a lower value than the current value and sets the mutation rate MR to a higher value than the current value, so that the system can increase the probability of generating a set of more optimal group control performance when the mutation performs weighting in a wider search process. The system may also clear the reduced lookup state.

Further, in the ninth embodiment, if the success index becomes lower than the expected value in the latter stage (or latter half) of the search, the process sets the crossover rate CR to a value higher than the current value and the mutation rate MR to a value lower than the current value thereof, so that the system can make the search converge early when weighting the partial search. The system may also clear the reduced lookup state. Tenth embodiment (another embodiment of the appearance ratio modification)

Referring to fig. 35, another embodiment of the occurrence modifier 4 is described below. Fig. 35 shows an appearance ratio modification program, which is a program partially modified by the appearance ratio modification program (see fig. 34) of the ninth embodiment.

In FIG. 35, step 502 determines whether a certain estimated number of times NEb (e.g., 40 times) has been reached. Step 506 determines whether it is in the last lookup phase (or whether NE ≧ NEd 2). Finally, in step 510, the crossover rate CR is set slightly greater than the current value (e.g., slightly greater than 0.001), while the variance rate MR is set slightly less than the current value (e.g., less than 0.001). Therefore, in the final stage of the search, the crossing rate CR gradually becomes larger as the search progresses, and conversely, the variation rate MR gradually becomes smaller. The operation is identical to that of the ninth embodiment except for the above-described portion.

As described above, in the tenth embodiment, since the system shifts from generating weighted variations to generating mainly by interleaving as the search progresses, the system can maintain the change in the group control performance in the initial stage (the first half) of the search and can converge the search early in the final stage (the second half). As a result, a set having a more excellent group control performance can be efficiently found. Eleventh embodiment (another embodiment of cross-pair selection) referring to fig. 36 to 39, the eleventh embodiment will be described below, mainly explaining portions different from the first or second embodiment.

Fig. 36 shows a system of the eleventh embodiment. In this embodiment, the distance between sets is used as a condition for selecting a set pair (cross pair) for a cross destination. Further, in the present embodiment, the system is constructed so that the condition (hereinafter referred to as "distance condition") can be modified according to the distance between the sets. A mother set shown in fig. 36 selects the condition modifier 5 for modifying the distance condition according to the search case.

Fig. 37 shows an operation procedure of the eleventh embodiment, which corresponds to fig. 11 of the first embodiment. In FIG. 37, step 31 generates a new set. Although the operation of this generation is as described in fig. 14 of the first embodiment, this embodiment is different from the first embodiment in the process of selecting the cross-pair PS1, PS2 (step 317). The distance condition described above can be modified by adding a parent set selection condition modification program corresponding to the parent set selection condition modifier 5 (see fig. 36) to step 52. The other structure of this embodiment is constituted in the same manner as the first embodiment.

Fig. 38 shows the operation of step 317 included in the new set generation program 31 (see fig. 14). In step 317a of fig. 38, the value of the counter RC is initialized to [0 ]. In the present embodiment, the counter RC counts the number of selection times of the set pair by using the distance condition between the sets. The inter-set distance is calculated according to equation [31 ].

In the next step 317b, the system weights the selection probability according to the magnitude of the performance estimate, randomly selecting the two sets PS1, PS2 that form a pair. The next step 317C determines whether the selection using the set of distance conditions is completed a certain number of times or more (e.g., 10 times).

If the two sets PS1, PS2 satisfying the distance condition still cannot be selected by repeating the steps 317d to 317h a certain number of times or more, the two sets PS1, PS2 selected in the step 317b are determined as crossing set pairs.

On the other hand, when the number of times of set selection from the use of the condition is less than a certain number of times (RC < 10), the process proceeds to step 317c, then to step 317d, and the value of the counter RC is incremented by 1. In step 317e, the distance DST between the two sets PS1, PS2 selected in step 317e is calculated. I.e., as calculated by equation [31]

DST = | EST (PS1) -EPS (PS2) |. As described above, each parameter value is normalized to a value between 0 and 100. In step 317e, if the distance DST is calculated, it is judged in steps 317f to 317h whether the distance DST satisfies a distance condition that is one of the cross pair selection conditions.

If the condition selection flag SELS is set to [1] and if the first selection condition is specified as the distance condition, the process proceeds to steps 317f,317g where it is determined whether the distance DST is equal to or greater than the first selection reference value DSTb 1. If DST < DSTb1, then the two sets PS1, PS2 that satisfy the distance condition are discarded and the process returns again to step 317b, repeating the same process from the beginning.

On the other hand, if the condition selection flag SELS is set to [2], and if the second selection condition is specified as the distance condition, it is determined in step 317h whether the distance DST is equal to or less than the second selection reference value DSTb2, as described above. If DST > DSTb2, then the selection condition is determined not to be satisfied and the process proceeds to step 317b to restart from the beginning.

The execution of steps 317b to 317h is repeated until the distance condition reaches a certain number of times (10 times), or until two sets satisfying the distance condition are found. If two sets satisfying DST ≧ DSTb1 or DST ≦ DSTb2 are generated, then the two sets are combined into a regular cross-pair PS1, PS2, and the process proceeds to step 318. Those steps after step 318 are the same as those in the first embodiment.

As indicated above, the system may select a cross-pair using a distance condition. The distance condition may be modified by the selection condition modification program 52 according to the search situation.

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

In steps 521 to 524, the success index RSC is run as in the operation of steps 501 to 504 in the occurrence rate modification program 50 (see fig. 34) in the ninth embodiment. The success index is typically defined as the number of additional recordings divided by the estimated number.

The selection reference values DSTb1, DSTb2 are modified by the estimated number of times NE, the first judgment value NEd1, the second judgment value NEd2, the success index RSC, and the success rate judgment values RSCd, RSCe, RSCf in steps 525 to 536. For example, NEd1= 500; NEd2= 800; RSCd = 0.10; RSCe = 0.05; RSCf = 0.05. For example, at step 283 of the initialization program 28 (see fig. 13), SELS = 1; DSTb1= 250; DSTb2= 250.

Step 525 judges NE ≦ NEd1 no, i.e., the lookup is still in (first half), sets SELS to [1] in step 527, and designates the first selection condition as the cross-pair selection condition. Then, in step 529, the success index RSC and the success rate judgment value RSCd are compared. If the success index RSC is less than the success rate judgment value RSCD (RSC is less than or equal to RSCD), it is judged that there are too few sets that meet the condition because the value of the currently set first selection reference value DSTb1 is too limited, thereby preventing the success index RSC from being high. Therefore, in step 531, the first selection reference value DSTb1 is set slightly smaller (e.g., 5% smaller).

On the other hand, if NE ≧ NEd2, or, namely, the search is in [ last stage ], the procedure proceeds to steps 525, 526, 528, setting SELS to [2 ]; and the second selection condition is designated as a cross-pair selection condition. Then, in step 529, the success index RSC and the success rate judgment value RSCf are compared. If the success index RSC is less than the success rate determination value RSCf (RSC ≦ RSCf), it is determined that there are too few sets that meet the condition because the value of the second selection reference value DSTb2 that is currently set is too limited, thereby preventing the success index RSC from being high. Therefore, in step 532, the value of the second selection reference value DSTb2 is set slightly larger (e.g., 5% larger).

If the success index RSC is greater than the success rate judgment value RSCf, it is judged that the selection reference value of the currently selected cross pair selection condition will be increased on the current value without any problem, and the operation of the selection condition modification program 52 is ended.

When NEd1 < NE < NEd2 is judged as [ midway ] in steps 525, 526, the success index RSC and the success rate judgment value RSCE are compared in step 533. If the success index RSC is less than the success rate judgment value RSCE (RSC is less than or equal to RSCE), the current selection condition is considered to be not appropriate, and therefore the selection condition needs to be changed. Therefore, after determining whether the current selection condition is the first or second selection condition in step 534, the selection condition is changed to the second selection condition (SELS =2) in step 535, or to the first selection condition (SELS =1) in step 536.

If the success index RSC is greater than the success rate judgment value RSCe, it is judged that the currently selected selection condition is not problematic, and the operation of the selection condition modification program 52 is ended.

As described above, in the eleventh embodiment, since the inter-set distance representing the similarity between two sets is used as the selection criterion of the cross-pair, a new set can be generated with an extensive and local search process.

That is, if the pair of sets whose inter-set distance is equal to or greater than the first selection reference value DSTb1 takes selection priority, and if the intersection is matched with two sets having features as different as possible from each other, the probability of generating a set with better group control performance can be increased, although the process may suffer from frustration or loss of performance and poor convergence. On the contrary, if the pair of sets whose inter-set distance is equal to or less than the second selection reference value DSTb2 takes the selection priority, and if the intersection is matched with two sets having the same characteristics as each other as much as possible, the probability of generating a new set having excellent group control performance can be increased although the probability of generating a set having excellent group control performance is reduced.

In an eleventh embodiment, the system determines, in particular on the basis of the estimated number, the initial or final phase of the search in which the selection priority is given to the pair of sets whose inter-set distance is equal to or greater than the first selection reference value DSTb1, weighting the extensive search process. In the final stage of search, the system can improve the search efficiency by providing a selection priority for the pair of sets whose inter-set distance is equal to or less than the second selection reference value DSTb2 and weighting the search convergence.

In particular, in the eleventh embodiment, the selection condition is searched for using an intersection in which the selection priority is provided for the pair of sets whose inter-set distance is equal to or greater than the first selection reference value DSTb1, the first selection reference value DSTb1 is set to a value smaller than its current value, so that the intersection condition becomes more gradual if the success index RSC becomes smaller than the desired value at the initial stage of the search. As a result, since the assigned first selection reference value DSTb1 is not appropriate, the search is in a low-falling state, and the appropriate first selection reference value DSTb1 will be automatically set to clear the low-falling state.

In particular, in the eleventh embodiment, the selection condition is searched by using the intersection in which priority is given to a pair of sets whose inter-set distance is equal to or smaller than the second selection reference value DSTb2, the second selection reference value DSTb2 is set to a value larger than its current value, so that the intersection condition becomes more gentle if the success index RSC becomes smaller than the desired value at the last stage of the search. As a result, since the assigned second selection reference value DSTb2 is not appropriate, the search is in the low-falling state, and an appropriate second selection reference value DSTb2 is automatically set to clear the low-falling state.

In particular, in the eleventh embodiment, during the search, the search is performed midway through the first cross-pair selection condition, which is changed to the second cross-pair selection condition if the success index RSC becomes less than the expected value, so the system can automatically convert it to the appropriate selection condition when the selection condition does not make it suitable for the current situation and the search is brought into the low-falling state.

Furthermore, particularly in the eleventh embodiment, when a search is performed under the second cross-pair selection condition in the last stage of the search, if the success index RSC becomes less than the expected value, the condition is changed to the first cross-pair selection condition, so that when the selection condition does not make it suitable for the current situation, the search is brought into a low state, which the system can automatically convert to a suitable selection condition. Twelfth embodiment (another embodiment of cross-pair selection) ]

Referring to fig. 40, another embodiment of the selection condition modifier 5 is described. Fig. 40 shows a selection condition modification program which is a modified form of a part of the selection condition modification program 52 (see fig. 39) of the eleventh embodiment.

In fig. 40, step 522 determines whether the estimated number of times is a fixed number NEb (e.g., 50 times). Step 526 determines whether it is a second time period (post stage), and if it is not the second time period (NE < NEd2), the first condition is designated as a cross-pair selection condition in step 527; in step 531, the first selection reference value DSTb1 is set slightly smaller (e.g., 2% smaller). Otherwise, if it is determined in step 526 to be the second period (NE ≧ NE2d), then the second condition is designated as the cross-pair selection condition in step 528; in step 532, the second selection reference value DSTb2 is set slightly smaller (e.g., 2% smaller).

As described above, in the twelfth embodiment, during the period of the cross-pair selection condition used with the first selection reference value DSTb1, the cross-pair selection condition may be switched according to the progress of the search. That is, the value of the first selection reference value DSTb1 at the initial stage of the period is set to be larger than that during the later stage of the period, so that the variation of the group control performance is weighted at the initial stage of the period, and the search convergence is weighted at the later stage of the period.

Also in the twelfth embodiment, the cross-pair selection condition is converted in accordance with the search progress during the period of the set-pair selection condition used with the second selection reference value DSTb 2. That is, the second selection reference value DSTb2 at the initial stage of the period is set to be smaller than that at the later stage of the period, so that the change in the group control performance is weighted at the initial stage of the period, and the convergence of the search is weighted at the later stage of the period. Therefore, the system can improve the search efficiency by converting the selection condition. Thirteenth embodiment (another embodiment of cross parameter selection)

In a first embodiment, the cross-over parameter (parameter position) in which the parameter values are permuted is randomly selected for both mother sets. In contrast, the generator in the thirteenth embodiment is characterized in that the difference between the parameter values (parameter deviation) is used as the parameter selection condition, and the parameter selection condition is modified in accordance with the search case.

The generator is described with reference to fig. 41 and 42. In the present thirteenth embodiment, the portions different from the eleventh embodiment are mainly described.

Fig. 41 shows the contents of step 318 in the new set generation program 31 (see fig. 14) of the thirteenth embodiment. In fig. 41, in step 318a, the value of the counter RC is initialized to "0". In the present embodiment, the counter RC is used to count the number of times to determine a parameter deviation condition as one of the cross parameter selection conditions. Step 318b generates a random number in the range of 0 to 25 to determine the parameter PX associated therewith. This is the same operation as described for the first embodiment.

Step 318c compares the number of times the parameter deviation condition is determined to a certain number of times. When the cross parameter PX satisfying the parameter deviation condition cannot be found even when the steps 318d to 318h are repeated a certain number of times or more (e.g., 10 times), it is judged that the selection in consideration of the parameter deviation has been sufficiently completed, and the process ends at step 318 to determine the parameter PX selected at step 318b as the cross parameter.

On the other hand, in step 318c, if it is determined that the number of times the parameter deviation condition is less than a certain number of times (RC < 10), the process proceeds to step 318c, and then to step 318d, the counter RC is incremented by 1. In a next step 318e, the difference between the PX-th values of the selected two sets PS1, PS2 [ | EPS (PS1) < PX > -EPS (PS2) < PX > ], is calculated to find the distance DSTP. Each parameter is converted to a value between 0 and 100 and normalized. At step 387 (see fig. 20) of the optimal set extractor 38, each parameter is reconverted to a useful value in the group control apparatus 1 when the optimal set data BPD is generated.

As described above, the difference DSTP of the PX-th parameter values of the two sets PS1, PS2 is calculated at step 318e, and then it is determined whether the deviation condition specified by the selection is satisfied at steps 318f to 318 h.

When the first selection condition is specified as the parameter deviation condition (SELS =1), the procedure proceeds to step 318f, and then to step 318g, at which it is determined whether the deviation DSTP is equal to or greater than the first selection reference value DSTc 1. If DSTP < DSTc1, the crossover parameter PX that does not meet the parameter bias condition is discarded and the process returns to step 318b to repeat the same run from the beginning.

Also, when the second selection condition is specified as the parameter deviation condition (SELS =2), it is judged at step 318h whether the deviation DSTP is equal to or less than the second selection reference value DSTc 2. If the selection condition is not met (DSTP > DSTc2), the process proceeds to step 318b and starts over again.

The operation of steps 318b to 318h is repeated until the parameter deviation condition is determined a certain number of times (10 times) or more, or until the crossing parameter satisfying the parameter deviation condition is found. If the crossing parameter PX meeting DSTP ≧ DSTc1 or DSTP ≦ DSTc2 is detected, the parameter PX is determined as the regular crossing parameter PX, and the process proceeds to the next step 319. Since the steps after step 319 are the same as those in the first embodiment, a description thereof will be omitted.

Fig. 42 shows one method of modifying the selection conditions described above based on the search progress.

Fig. 42 shows a selection condition modification routine at step 52 (see fig. 37) of the operation routine 100.

The selection condition modifying program of the thirteenth embodiment is the same as the selection condition modifying program (fig. 39) of the eleventh embodiment except for steps 531, 532. Therefore, the operations of steps 529 to 532 in which the selection reference values DSTc1, DSTc2 are corrected will be mainly described, assuming that at step 283 (see fig. 13) of the initialization program 28, the specified data SELS = 1; selecting a reference value DSTc1= 50; the reference value DSTc2=50 is selected.

In the initial stage of the search (NE ≦ NEd1), if the successful index RSC is smaller than the desired value RSCD (RSC ≦ RSCD) while the first selection condition (SELS =1) is selected, it is judged that the value of the cross parameter satisfying the condition is too small and the successful index RSC is prevented from becoming larger because the currently set first selection reference value DSTc1 is too limited and inappropriate. Then, the process proceeds to step 529 and then to step 531, where the first selection reference value DSTc1 is set to be slightly smaller (for example, by 5%), and the operation of the selection condition modification program 52 is ended.

On the other hand, at the last stage of the search (NE ≧ NEd2), if the success index RSC is smaller than the desired value RSCf (RSC ≦ RSCf) while the second selection condition (SELS =2) is selected, it is judged that the cross parameter that meets this condition is too small, and the success index RSC is prevented from becoming larger because the currently set second selection reference value DSTc2 is too limited and inappropriate. The process then proceeds to step 530 and then to 532, at which point the second selection reference value DSTc2 is set slightly larger (e.g., 5% larger).

If the success index RSC is greater than the desired value RSCd, it is judged that the selection reference value of the currently selected selection condition is increased over the current value without question, and the operation of the selection condition modification program 52 is ended.

As described above, in the thirteenth embodiment, since the parameter deviation representing the similarity between the two selected sets is obtained, and since the parameter selection condition is thereafter set in accordance with the parameter deviation, it is possible to make the generation converge in consideration of the variation of the group control performance and the search.

That is, if a control parameter whose parameter deviation is equal to or greater than the first selection reference value DSTc1 gives priority in selection, and if the intersection can be matched to parameters whose characteristics are different from each other as much as possible, the probability of generating a new set with excellent group control performance can be improved although the selection may suffer from trouble or the performance is lost and the convergence is poor. If the control parameters whose parameter deviation is equal to or less than the second selection reference value DSTc2 take the selection priority and if the intersection can be matched to the parameters whose characteristics are as identical to each other as possible, the generation probability of the new set having relatively excellent group control performance can be increased although it is possible to reduce the generation probability of the new set having excellent group control performance.

In the thirteenth embodiment, the initial stage and the final stage of the search are judged according to the estimated times; in the initial stage of searching, executing a searching process for weighting the group control performance variation, wherein the control parameters with the parameter deviation equal to or greater than the first selection reference field value DSTc1 obtain the selection priority; on the contrary, in the final stage of the search, a search process for weighting the search convergence is performed, wherein the control parameters whose parameter deviation is equal to or less than the second selection reference value DSTc2 take the selection priority. Therefore, generation considering the group control performance variation and the search convergence can be performed according to the search period.

In the thirteenth embodiment, in performing the search by applying the cross parameter selection condition used by the first selection reference value DSTc1, the first selection reference value DSTb1 is set to a value smaller than its current value, so that the selection condition becomes more gradual if the success index RSC becomes smaller than the desired value at the initial stage of the search. As a result, the lookup is in a low state due to the assigned first selection reference value DSTb1 being inappropriate, the value of which is automatically modified to an appropriate value. In the thirteenth embodiment, when the search is performed with the cross parameter selection condition to which the second selection reference value DSTc2 is applied, the second selection reference value DSTb2 is set to a value smaller than its current value, and therefore, at the final stage of the search, if the success index RSC becomes smaller than the desired value, the selection condition becomes more gradual. As a result, when the search is in the low state due to the assigned second selection reference value DSTb2 being inappropriate, the value thereof is automatically corrected to an appropriate value.

In the thirteenth embodiment, when a search is performed with a first cross parameter selection condition applying the first selection reference value DSTc1 in the middle of the search, if the success index RSC becomes smaller than a desired value, it is converted into a second cross parameter selection condition applying the second selection reference value DSTc 2. As a result, when the search is in a low state because the first cross parameter selection condition is not suitable for the present situation, the condition can be automatically shifted to a suitable selection condition.

In the thirteenth embodiment, when a search is performed with the second cross parameter selection condition applying the second selection reference value DSTc2 in the middle of the search, if the success index RSC becomes smaller than a desired value, it is converted into the first cross parameter selection condition applying the first selection reference value DSTc 1. As a result, when the search is in a low state because the second cross parameter selection condition is not suitable for the present situation, the condition can be automatically converted into a suitable selection condition.

In this way, the system according to the thirteenth embodiment can clear the search of the low state and thereby improve the search efficiency by changing the selection reference value and converting the selection condition according to the search situation. [ fourteenth embodiment (another embodiment of selecting condition modification) ]

Referring to fig. 43, another embodiment of the selection condition modifier 5 will be described below. Fig. 43 shows the operation steps of the selection condition modification program 52, which is a partially modified program in the thirteenth embodiment (see fig. 42).

In fig. 43, step 522 determines whether the estimated number of times is a fixed number NEb (e.g., 50 times). Step 526 determines whether it is a second period (post-stage), and if it is not the second period (NE < NE2d), the first condition is designated as a cross parameter selection condition in step 527; in step 531, the first selection reference value DSTb1 is set slightly smaller (e.g., 2% smaller).

Conversely, if it is determined in step 526 that the second period (NE ≧ NE2d), then the second condition is designated as a cross parameter selection condition in step 528; in step 532, the second selection reference value DSTb2 is set slightly smaller (e.g., 2% smaller).

In this way, the condition relating to the parameter deviation is changed according to the search situation. The operation is the same as that of the thirteenth embodiment except for the above.

As described above, in the fourteenth embodiment, in the period of time for which the condition is selected using the crossing parameter to which the first selection reference value DSTb1 is applied, the first selection reference value DSTb1 at the initial stage of the period is set to be larger than that at the later stage of the period, and for this reason, the selection condition is set such that the initial stage of the period is more strict than that at the later stage, so that, at the initial stage of the period, the group control performance variation is weighted and, at the final stage of the period, the search convergence is weighted.

Also, in the fourteenth embodiment, in the period of the selection condition using the crossing parameter to which the second selection reference value DSTb2 is applied, the second selection reference value DSTb2 of the latter stage of the period is set to be smaller than the value of the initial stage of the period, and for this reason, the selection condition is set so that the latter stage of the period becomes stricter than the initial stage. As a result, the cluster control performance variation is weighted during the initial phase of the epoch, and the search convergence is weighted during the final phase of the epoch.

Although in the ninth to fourteenth embodiments, it is determined whether it is the first half or the second half of the search, or the initial stage or the final stage of the search, based on the estimated number of times NE, it may be determined whether it is the first half or the second half of the search, or the initial stage or the final stage of the search, by replacing the estimated number of times NE with the additional recording number of times NR. Fifteenth embodiment (another embodiment of parameter selection)

Next, another embodiment of the generator will be described using fig. 44 and 45. In the fifteenth embodiment, the selection probability (occurrence rate) of each parameter is judged according to the degree of correlation of the traffic flow characteristics and the degree of correlation of the group control performance estimation items. The basic structure of the fifteenth embodiment is the same as that of the second embodiment, and therefore, the following will mainly describe a portion thereof different from the second embodiment.

Fig. 44 shows the contents of step 318 (see fig. 14) in the new set generation program 31.

In fig. 44, steps 318j to 318q judge the parameter occurrence rates RPA (1) to RPA (25) of the 25 parameters according to the traffic flow specification. In particular, in steps 318j to 3181, the form of the traffic flow is distinguished according to the contents of the number of passengers, the proportion of the bottom traffic, the proportion of the upward traffic, the proportion of the downward traffic, and the like, and the similar proportion included in the traffic flow specification data TRS. That is, it is determined which of the peak time period, the upper top, the lower bottom, and the regular time period is the current traffic situation.

If it is determined as the regular time period, the occurrence rates RPA1(1) to RPA1(25) of the respective parameters previously prepared for the regular time period are set as the parameter occurrence rates RPA (1) to RPA (25) at step 318 m. Also, when it is determined as the peak time, RPA2(1) to RPA2(25) are set as the parameter occurrence rates RPA (1) to RPA (25) at step 318 n; when it is judged as the top time, RPA3(1) to RPA3(25) are set as parameter occurrence rates RPA (1) to RPA (25) at step 318P; when the lower floor time is determined, RPA4(1) to RPA4(25) are set as the parameter occurrence rates RPA (1) to RPA (25) at step 318 q. Column 10B of fig. 45 shows the parameter occurrence rates RPA1(1) to RPA1(25), RPA2(1) to RPA2(25), RPA3(1) to RPA3(25), RPA4(1) to RPA4(25) prepared for various traffic flows.

In fig. 45, the feature of the traffic flow is for a regular period, with respect to the parameter occurrence rates RPA (1) to RPA (25), the occurrence rate is set to "10" for the parameters having close relation with the traffic flow (parameter numbers =1 to 9,22 to 25), and the occurrence rate is set to "0" for the parameters having little relation with the traffic flow (parameter numbers =18 to 21). For control parameters having a moderate relationship with the traffic flow (parameter number =10 to 17), the occurrence rate is set to "5".

When the characteristics of the traffic flow are for the peak time period, the occurrence rates are set to "10" with respect to the parameters RPA2(1) to RPA2(25) for the parameters having close relation with the traffic flow (parameter numbers =1 to 9,18 to 21), and are set to "0" with respect to the parameters having little relation with the traffic flow (parameter numbers =10 to 17). For control parameters with a moderate relationship to traffic flow (parameter number =22 to 25), the occurrence rate is set at "5".

When the characteristics of the traffic flow are for the top time period, regarding the parameters of occurrence rates RPA3(1) to RPA3(25), the occurrence rate is set to "10" for the parameters having close relation with the traffic flow (parameter numbers =1 to 13), and the occurrence rate is set to "0" for the parameters having little relation with the traffic flow (parameter numbers 14 to 21). For control parameters with a moderate relationship to traffic flow (parameter numbers 22 to 25), the occurrence rate is set at "5".

When the characteristics of the traffic flow are directed to the lower floor time period, regarding the parameters RPA4(1) to RPA4(25), the occurrence rate is set to "10" for the parameters having close relation with the traffic flow (parameter numbers =1 to 9,14 to 17), and the occurrence rate is set to "0" for the parameters having little relation with the traffic flow (parameter numbers =18 to 21). For control parameters with a moderate relationship to traffic flow (parameter number =22 to 25), the occurrence rate is set at "5".

As described above, steps 318j to 318q determine the parameter occurrence rates RPA (1) to RPA (25) of the 25 control parameters according to the traffic flow specification. The values of the occurrence rates of the respective traffic flows RPA1(1) to RPA1(25), RPA2(1) to RPA2(25), RPA3(1) to RPA3(25), and RPA4(1) to RPA4(25) are not limited to the values shown in fig. 45. It is possible to set any value if the comparison shows the degree relating to each traffic flow characteristic. There may be a small difference in the occurrence rate between the control parameters.

Referring next to fig. 44, step 318r corrects the parameter occurrence rates RPA (1) to RPA (25) according to the correction values RPAA (1) to RPAA (25), where the correction values RPAA (1) to RPAA (25) are proportional to the degree of correlation (e.g., average latency) of the group control performance estimation terms.

The correction values RPAA (1) to RPAA (25) are set by the above-described performance reference value setting means 3. As described in the second embodiment, the performance reference value setting means 3 provides as its output the [ target value ] of the average waiting time, the [ specified value ] of the estimated reference value BX, and in the present fifteenth embodiment, in addition to the above, the [ degree of correlation ] with the average waiting time as the estimation term is provided as the correction value, as in the second embodiment.

Therefore, the reference value data TGT contained in the seek condition signal 1a is supplied from the group control device 1 to the seek device 10, which contains the waiting time target value TAW, the additional reference designation value TCB, and the correction values RPAA (1) to RPAA (25). RPAA (1) to RPAA (25) shown in fig. 45 indicate correction values of the target values of the average waiting time.

Regarding the correction values RPAA (1) to RPAA (25), the correction value is set to "10" for the control parameters (parameter number =8,22,23) having close relation to the average waiting time as the estimation items, and is set to "0" for the control parameters (parameter number =10 to 21) having little relation thereto. For the control parameters (parameter numbers =1 to 7,9,24,25) having a moderate relationship therewith, the correction value is set to "5".

On the other hand, if the estimation item is not the average waiting time but the power saving, with respect to the correction values RPAA (1) to RPAA (25), the correction value is set to "10" for the control parameters (parameter numbers =4 to 7,22 to 25) having close relation to the average waiting time as the estimation item, and the correction value is set to "0" for the control parameters (parameter numbers =9 to 21) having little relation thereto. For the control parameters (parameter numbers =1 to 3,8) having a moderate relationship therewith, the correction value is set to "5".

Even if the estimation items are not as described above, the correction values RPAA (1) to RPAA (25) can be determined according to the degree of correlation as well. Any value may be set for the correction values RPAA (1) to RPAA (25) if the comparison table shows the degree of correlation with the estimation items. There may be small differences in the correction values between the control parameters.

Next, at step 318S, a random number having a value between [0] and [ sum of parameter occurrence rates RPA (1) to RPA (25) ] is generated to determine the parameter number PX for performing the crossover or mutation. Then, the process proceeds to the next step 319, and since the steps after step 319 are the same as those of the first embodiment, a description thereof will be omitted.

As described above, in the fifteenth embodiment, the parameter values having close relation with the specific traffic flow characteristics take priority in the change because the degree of correlation between the parameter and the traffic flow characteristics is set as the parameter selection condition, thereby increasing the probability that a new set having excellent group control performance will be generated.

In the fifteenth embodiment, the occurrence rate of the degree of correlation generation ratio with the traffic flow characteristics is set for each parameter, and the parameter is selected based on the occurrence rate, so that it is easy to select a parameter that has a close relationship with a specific traffic flow characteristic and easily affects the group control performance, and the probability of generating a new set having excellent group control performance becomes high.

In the fifteenth embodiment, since the control parameter unrelated to the traffic-flow characteristic is not selected by setting its occurrence rate to zero when the group control is performed, it is possible to completely prevent the application of the intersection or variation to the parameter unrelated to the traffic-flow characteristic.

In the fifteenth embodiment, since the degree of correlation between the parameter and the estimation item as the estimation purpose is set as the parameter selection condition, the parameter values having close relation with the estimation item take priority in the change, thereby increasing the probability that a new set having excellent group control performance will be generated.

In the fifteenth embodiment, an occurrence rate is set for each parameter, which is proportional to the degree of correlation of the estimation item as the estimation purpose, and the parameter is selected based on the occurrence rate, so that it tends to be easy to select a parameter that easily affects the group control performance, so that the probability of generating a new set having excellent group control performance is increased.

In the fifteenth embodiment, since the control parameters unrelated to the estimation items are not selected by setting the occurrence rates thereof to zero, it is possible to completely prevent the application of the crossover or the variation to the parameters unrelated to the estimation items as the estimation targets.

Furthermore, in the fifteenth embodiment, the correlation between the parameter and the estimation item as the estimation target and the correlation between the control parameter and the traffic flow characteristics are combined to form the parameter condition, so that the probability of generating a new set having excellent group control performance is increased.

Thus, according to the fifteenth embodiment, generation of useless new sets, estimation, judgment of additional records, and the like are reduced, enabling efficient search.

Note that, if the estimation function of the crowd control performance includes a plurality of estimation items, the respective appearance rates are weighted, or correction values are added, according to the importance of the respective estimation items. Sixteenth embodiment (another embodiment of cross-pair selection) ]

Next, another embodiment of the generator will be described. In this embodiment, the crossing sets are selected according to the number of identical sets. With regard to the sixteenth embodiment, the following will mainly describe the portions thereof different from the first embodiment.

The operation of step 317 to select the cross-pair PS1 and PS2 according to this embodiment (see fig. 14) in the new set generation program 31 (see fig. 11) is significantly different from that of the first embodiment, the operation of which is described below with fig. 46.

In FIG. 46, first the distance DST (i, j) between the various sets (if i, j =1, 2.. once, P; i ≠ j) is calculated at step 317j according to equation [31 ]. The calculation is the same as that of step 414 (see fig. 26) in the deletion program 35 of the third embodiment. Each parameter value is normalized to a value between 0 and 100.

Step 317K initializes the number of sets i to 1, and repeats the operations of steps 317l to 317n until the occurrence rates RSA (1) to RSA (P) set by all sets (i =1, 2.. ·, P) are detected at step 317P. In step 317l, j ≠ i, while j =1, 2.. The P-totals find the number of sets MDST (i) of DST (i, j) ≦ DSTa (the same number of sets). DSTa is a judgment value for judging whether or not the two sets are identical to each other, and in the sixteenth embodiment, it is placed at 25 as in the third embodiment.

In step 317m, the occurrence rate rsa (i) of the set i is calculated by the formula rsa (i) =1 ÷ { mdst (i) +1} based on the same set number mdst (i). That is, as the same number of sets becomes smaller, the occurrence rate is set larger. To calculate the occurrence for the next step, step 317n increments the number of sets by 1.

Thus, when the occurrence rates RSA (1) to RSA (p) are determined for all sets, finally two random numbers with a value between [0] and the sum of [ occurrence rates (RSA (1) to RSA (p)) are generated in step 317q, and then two parent sets PS1, PS2 are selected on the basis of the respective random numbers and the occurrence rates RUA (1) to RUA (p).

This is the end of the run of step 317 and the two sets PS1 and PS2 are determined to be standard crossing pairs. The process then proceeds to the next step 318. The steps after step 318 are the same as those of the first embodiment, and a description thereof is omitted.

As described above, in the sixteenth embodiment, since the crossing pairs are selected in accordance with the same number of sets, the probability of crossing the pairs of sets having mutually different characteristics is improved. Thus, the probability of generating a new set with excellent group control performance is improved.

Note that, in the above-described embodiment, although only one cross parameter is selected (this parameter selection method is generally referred to as "one-dot cross"), other methods may be employed. A method of simultaneously selecting two or more crossing parameters ("multipoint crossing") may be employed. A method called "uniform crossover" may also be employed in which program columns (masks) having the same length as the parameter numbers are prepared in advance, and which parent set can convert various genes (parameter values) into child sets is determined based on the values of the individual bits specified by the masks. This is the same as [ variant ].

Although this set includes 25 parameters, the number and content thereof are only examples, and the present invention is applicable to any type of parameter set used in a group control algorithm. Seventeenth embodiment (another embodiment of the system configuration)

In the above embodiments, although the group control device 1 and the search device 10 are installed in the elevator machine room of the building and the optimal set is obtained by online operation, other methods may be adopted.

For example, in embodiments 1 to 6 and 8 to 15, as shown in fig. 47, it is also possible to install the search device 10 and the simulator 2 in the monitoring center of the elevator maintenance company and connect the search device 10 and the group control device 1 by the communication devices 4A and 4B through telephone lines. In this case, the communication device 4A can perform data communication with another building in which the communication device 4B is installed. With this configuration, a set of search devices and a simulator are typically used for a plurality of group control devices. The finding device 10, the simulator 2 and the communication device 4A may be installed in a manager's room or a security center.

According to the seventeenth embodiment, the cost of the system can be reduced by sharing one expensive finding device 10 and one simulator 2.

Particularly in the eighth embodiment, as shown in fig. 48, it is also possible to install the search apparatus 10 in a manager room of a building or a monitoring center of an elevator maintenance company, and connect the search apparatus 10 and the group control apparatus 1 by using the communication apparatuses 4A and 4B through telephone lines.

The finding apparatus 10 may be used for developing a group control algorithm, i.e. a case for selecting an optimal group control algorithm scheme from a plurality of group control algorithm schemes. In short, when a new group control algorithm is to be developed, a simulator can be used for simulation according to the group control performance data PRF obtained at that time, the performance group control algorithm is estimated, or an optimal set is found. In this case, the simulator 2 is connected to the finding apparatus 10 as shown in fig. 49.

The finding device 10 is also useful when the cluster control device 1 is shipped from its factory, when the best set is found by the simulator 2 shown in fig. 49 and then recorded, or when recording the initialized set clusters GPS1 to GPS 4. The finding means 10 and the simulator 2 may be implemented by one microcomputer. The cluster control device 1, the search device 10, and the simulator 2 may be a single microcomputer.

Claims (37)

1. A system for group control of a plurality of elevator cars according to a group control algorithm, said group control algorithm comprising a plurality of parameters, said system comprising finding means for finding an optimal set among a set provided to said group control algorithm as a combination of parameter values, characterized in that said finding means comprises:

a storage device to store a plurality of sets;

generating means for selecting one or more sets from the storage means as one or more parent sets and generating one or more new sets that inherit properties of the partial parent sets;

estimating means for finding an execution result as a group control performance value each time the group control algorithm is executed with each new set;

selecting means for improving the plurality of sets stored in said storage means by adding said new set to said storage means and deleting defective sets from said storage means; and

and an extracting means for extracting an optimum set from the plurality of sets which have been improved and stored in the storage means, based on the group control performance value.

2. The system of claim 1, wherein said generating means comprises:

numerical value exchange means for generating two new sets by exchanging numerical value parts of the two sets selected from the storage means;

new value replacement means for generating a new set by replacing a part of parameter values of a set selected from the storage means with new values generated in a random manner; and

generating method selection means for selecting between value exchange and new value permutation according to a certain probability.

3. The system of claim 1, wherein said generating means comprises:

parent set selection means for selecting one or more sets from said storage means;

parameter selection means for selecting, together with said one or both sets, a parameter by which to exchange a value or to permute a new value;

numerical value exchanging means for generating two new sets by exchanging the numerical value parts of the parameters selected by the parameter selecting means between the two sets selected by the parent set selecting means;

new value replacing means for generating a new set by replacing the parameter values of the set selected by the parameter selecting means and selected by the parent set selecting means with new values generated in a random manner; and

generating method selection means for selecting between value exchange and new value permutation according to a certain probability.

4. The system of claim 3, wherein the parent set selecting means performs parent set selection based on the parent set selection reference information to provide a probability of generation of a good new set.

5. The system according to claim 4, wherein the parent set selection reference information is distances between sets, the parent set selection means calculates the distances between the sets, and randomly selects a pair of sets from the storage means, in which the distances between the sets satisfy a condition.

6. The system according to claim 4, wherein the parent set selection reference information is the group control performance value, and the parent set selection means weights a selection probability of each set according to the group control performance value, thereby randomly selecting one or both sets from the storage means.

7. The system according to claim 4, wherein the parent set selection reference information is the same set number, the parent set selection means calculates the same set number for each set, weights a selection probability of each set according to the same set number, and thereby randomly selects one or two sets from the storage means.

8. The system of claim 3, further comprising means for modifying the parent set selection criteria based on the progress of the search.

9. The system of claim 3, wherein the parameter selection means selects the parameter based on parameter selection reference information to increase a probability of generating a good new set.

10. The system according to claim 9, wherein the parameter selection reference information is a difference between two parameter values to be exchanged between the two sets, and the parameter selection means calculates the difference and randomly selects a parameter whose the difference satisfies a certain condition.

11. The system of claim 9, wherein the parameter selection reference information is a degree of correlation between the utilization of the elevator car and each parameter, and the parameter selection means weights the selection probability of each parameter according to the degree of correlation and randomly selects the parameter accordingly.

12. The system of claim 9, wherein the parameter selection reference information is a degree of correlation between the content of the performance estimation value and each parameter, and the parameter selection means weights a selection probability of each parameter according to the degree of correlation, and thereby randomly selects the parameter.

13. A system as claimed in claim 3, further comprising modifying means for modifying the parameter selection condition in dependence on the progress of the search.

14. The system of claim 2, further comprising probability modifying means for modifying the selection probability of each generation method in accordance with the progress of the search.

15. The system of claim 14 wherein said probability modifying means calculates a success index based on a ratio of the number of collections added to said storage means to the estimated number of collections, and modifies said selection probability based on said success index.

16. A system for group control of a plurality of elevator cars according to a group control algorithm, said group control algorithm comprising a plurality of parameters, said system comprising finding means for finding an optimal set among a set provided to said group control algorithm as a combination of parameter values, characterized in that said finding means comprises:

a storage device to store a plurality of sets;

a value exchanging means for generating a new set in which two parts inherit the properties of the parent set thereof by exchanging partial parameter values between the two sets selected as the parent set from the storage means;

new value replacement means for generating a new set partially inheriting the property of its parent set by replacing the partial parameter values of a set selected as the parent set from the memory means with new values generated in a random manner;

a generation method selection device for selecting a numerical value exchange method and a new value replacement method in combination with the probability of each method;

estimating means for finding an execution result as a group control performance value whenever the group control algorithm is executed with one or more new sets;

adding means for additionally storing only good new sets satisfying a certain adding condition to the storage means;

deleting means for deleting a defect set satisfying a certain deletion condition from the storage means; and

extracting means for extracting an optimum set from among the plurality of sets which have been improved and stored in the storage means, based on the group control performance value.

17. The system of claim 16, further comprising means for modifying the one additional condition or the plurality of additional conditions.

18. The system of claim 17, wherein the additional conditions are determined based on the group control reference values for the respective sets and are modified based on a look-up progress.

19. The system of claim 16, wherein the means for removing removes the set whose performance estimates are impairments.

20. The system of claim 19, wherein the means for deleting deletes a set that is the same as another set based on a distance between sets.

21. The system of claim 16, further comprising means for initializing a lookup.

22. The system of claim 21, wherein the initialization means comprises a first initialization mode and a second initialization mode; in the first initialization mode, a plurality of sets prepared previously are used as initialization; in the second initialization mode, a plurality of sets that have improved at the time of the last search cycle are used as initialization, whereby the first initialization mode and the second initialization mode are selected according to a search start condition.

23. The system of claim 16, further comprising an end determining means for determining an end of the search based on the search condition.

24. The system of claim 23, wherein the end determination means determines the end of the search based on the estimated number of sets.

25. The system of claim 23, wherein the end judgment means judges the end of the search based on the number of added sets.

26. The system of claim 23 wherein said end determining means determines the end of the search based on a success index which is a ratio of the number of added sets to the number of estimated sets.

27. The system according to claim 23, wherein the end judgment means calculates distances between sets with respect to a plurality of sets stored in the storage means, and judges the end of the search based on the distances between sets.

28. The system of claim 16, further comprising means for re-lookup determination for determining re-lookup based on changes in various preconditions found at the start of each lookup, the preconditions including elevator car specifications, traffic flow specifications, performance reference values, and control reference values.

29. The system of claim 16, wherein the storage means stores the group control performance values for assignment according to each set.

30. The system of claim 16, wherein the searching means is coupled to a target value setting means for setting a target value in conjunction with the searching process.

31. The system of claim 16, wherein the lookup device is coupled to a group control device that includes the group control algorithm and controls operation of the plurality of elevator cars, and the lookup device is coupled to a simulator that includes the same group control algorithm as the group control device, the estimation device setting the result of the performed estimation to a group control performance value.

32. The system of claim 31, wherein the look-up device and the simulator are remotely located from the cluster control device, the look-up device and the cluster control device being connected by a communication line.

33. The system of claim 16, wherein the lookup device is coupled to a group control device that includes the group control algorithm and controls operation of the plurality of elevator cars, and the estimation device sets the result of the execution to a group control performance value each time the group control algorithm is simulated by the group control device.

34. The system of claim 33, wherein the lookup device is remotely located from the group control device, the lookup device being coupled to the group control device via a communication line.

35. A system for group controlling a plurality of elevator cars, comprising:

a simulator comprising a group control algorithm for group controlling a plurality of said elevator cars; and

a finding means connected to said simulator for finding a set with optimal parameter values according to said group control algorithm, said finding means comprising:

a storage device to store a plurality of sets;

numerical value exchange means for generating two new sets that partially inherit the properties of their parent sets by interchanging partial parameter values between the two sets selected as the parent sets from the storage means;

new value replacement means for generating a new set having a part of the property of the mother set inherited thereto by replacing a part of the parameter values of the set selected from the storage means as the mother set with a new value generated in a random manner;

a generation method selection device for selecting a numerical value exchange method and a new value replacement method in combination with the probability of each method;

estimating means for finding the result of the execution as a group control performance value each time the group control algorithm is executed with one or more new sets;

adding means for additionally storing only a good new set satisfying a certain addition condition to the storage means;

deleting means for deleting a defect set satisfying a certain deletion condition from the storage means; and

and an extracting means for extracting an optimum set from among the plurality of sets which have been improved and stored in the storage means, based on the group control performance value.

36. A system for group control of a plurality of elevator cars according to a group control algorithm, said group control algorithm comprising a plurality of parameters, said system comprising finding means for finding an optimal set among a set provided to said group control algorithm as a combination of parameter values, characterized in that said finding means comprises:

a storage device to store a plurality of sets;

a cross type generating means for generating two new sets of which the parent set property is inherited by a part of the values exchanged between the two sets selected as the parent set from the storing means;

estimating means for finding the result of the execution as a group control performance value each time the group control algorithm is executed with one or more new sets;

selecting means for improving the plurality of sets stored in said storage means by adding said new set to said storage means and by deleting defective sets from said storage means; and

and an extracting means for extracting an optimum set from the plurality of sets which have been improved and stored in the storage means, based on the group control performance value.

37. A system for group control of a plurality of elevator cars according to a group control algorithm, said group control algorithm comprising a plurality of parameters, said system comprising finding means for finding an optimal set among a set provided to said group control algorithm as a combination of parameter values, characterized in that said finding means comprises:

a storage device to store a plurality of sets;

a variation generating means for generating a new set having a part of the property of the mother set inherited thereto by replacing a part of parameter values in a set selected from the storing means as the mother set with a new value generated at random;

estimating means for finding the result of the execution as a group control performance value each time the group control algorithm is executed with one or more new sets;

selecting means for improving the plurality of sets stored in said storage means by adding said new set to said storage means and by deleting a defective set from said storage means; and

and an extracting means for extracting an optimum set from the plurality of sets which have been improved and stored in the storage means, based on the group control performance value.

Images