JP2008009621A - Method for processing genetic algorithm, and method for simulating train operation using the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for processing a genetic algorithm capable of reducing a time for searching for optimal solutions. <P>SOLUTION: When evaluation value results of more than a half of first generation data [1] to [2m-1] are returned from slaves, a master performs genetic operation of current generation data on the basis of the returned evaluation value results to generate new data and assigns the generated data [1] to [m] and current generation data [m+1] to [2m-1] whose evaluation values are not calculated as a second generation data to the respective slaves. When evaluation value results of second generation data [m] generated by genetic operation of the current generation data having the best evaluation values and the second generation data [m+1] to [2m-1] whose evaluation values are not calculated are returned from the slaves, third generation data are generated in a manner similar to data generation of the second generation, and generations are repeatedly exchanged. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a genetic algorithm processing method, and more particularly, to a genetic algorithm processing method capable of efficiently generating next-generation data and shortening an optimal solution search time. The present invention also relates to a train operation simulation method using this genetic algorithm processing method.

  In the combinatorial optimization problem, there is a genetic algorithm as one of the methods for obtaining an optimal solution in a relatively short time, and a plurality of processing devices are used as a method for improving the processing speed of the optimal solution search of this genetic algorithm. There is a method of calculating evaluation values in parallel by a master / slave system.

  The master / slave system performs genetic operations (selection, crossover, mutation, etc.) of the genetic algorithm on the master, generates evaluation data (individual data of each generation), assigns each to multiple slaves, The slave calculates the evaluation value of the assigned evaluation data in parallel and returns it to the master. The master performs a genetic operation again on the returned evaluation value, and assigns the generated evaluation data to each slave. This operation is repeated until the preset condition is satisfied, and the generation change is repeated. However, this system must wait for a reply from the slowest slave when the evaluation value calculation time of each slave varies.

Therefore, in order to efficiently perform the processing of the genetic algorithm using the master / slave system, the performance of each slave is registered in the master as described in Patent Document 1, and each registered slave By varying the number of evaluation data to be allocated according to the performance of the slave, allocating a large amount of evaluation data to slaves with high performance and assigning a small amount of evaluation data to slaves with low performance, the calculation time of all slaves can be reduced. A method has been proposed in which the processing time until the search for the optimum solution is made uniform is shortened.
JP 2005-190372 A

  By the way, in the case of the method of Patent Document 1, when it is guaranteed that the registered performance of each slave can always be exhibited, the evaluation value calculation time can be made uniform and the processing time until the optimum solution search can be shortened. . However, for example, when each slave has a unique local process assigned in advance in addition to the evaluation value calculation requested by the master, and the evaluation value calculation is performed between local processes, the load status of the local process Accordingly, the processing time for evaluation value calculation of each slave changes. In addition, when there is a slave that cannot exhibit the registered performance due to a decrease in performance, the reply of the calculation result from the slave is greatly delayed. For this reason, the evaluation value calculation time of each slave varies. Furthermore, if a failure occurs in any of the slaves and the calculation result is not returned, the evaluation value data for which the calculation result has not been returned must be assigned to another slave and recalculated. In such a case, since the method of Patent Document 1 needs to wait for the reply of all evaluation value calculation results, there is a problem that the processing speed of the optimum solution search is greatly reduced.

  The present invention has been made paying attention to the above problems, and a genetic algorithm processing method capable of reducing the processing time for searching for an optimal solution without being affected by differences or changes in processing capabilities of a plurality of processing devices. The purpose is to provide. It is another object of the present invention to provide a train operation simulation method using the genetic algorithm processing method of the present invention.

  For this reason, the invention of claim 1 is an evaluation value calculation process for calculating evaluation values of a plurality of individual data of the current generation, and the inheritance of the individual data of the current generation based on the evaluation values obtained by the evaluation value calculation process. In a genetic algorithm processing method for searching for an optimal solution by performing generation change by repeatedly performing individual data generation processing for generating a plurality of next generation individual data by performing a general operation, the current generation in the evaluation value calculation processing The individual data generation process is executed for the current generation individual data of the majority of the evaluation values obtained when the evaluation value of the majority of the individual data is calculated, and the individual generated by the solid data generation processing The data and the current generation individual data for which the evaluation value is not calculated are used as the next generation individual data, the evaluation value calculation processing is executed for the next generation individual data, and the current generation individual data with the best evaluation value is genetically manipulated. When the evaluation value of the majority including the evaluation value result of the next generation individual data generated as described above and the next generation individual data for which the evaluation value has not been calculated is calculated, the next generation is generated in the same manner as the generation of the next generation individual data. It is characterized in that generational change is repeated by generating individual data.

  According to a second aspect of the present invention, there is provided a master for performing an individual data generation process for generating a plurality of next-generation individual data by performing a genetic operation on the individual data based on each evaluation value of a plurality of individual data of the current generation. Using the plurality of slaves that execute the evaluation value calculation process for calculating the evaluation value of the individual data allocated from the master at their own timing and send back to the master, and the individual data generation process and the evaluation value In the genetic algorithm processing method of searching for an optimal solution by repeating generation processing and repeating generation processing, when the master returns an evaluation value calculation result of more than a majority of the individual data of the current generation from the slave. , The individual data generation processing is executed for the current generation individual data of more than a majority of the evaluation values returned, and generated by the solid data generation processing Next generation individual data generated by genetically manipulating body data and current generation individual data for which evaluation values have not been calculated as next generation individual data and assigned to the plurality of slaves, and having the best evaluation value current generation individual data and the evaluation When more than half of the evaluation value calculation results including the evaluation value result of the next generation individual data whose value has not been calculated are returned from the slave, individual data for the next generation is generated in the same manner as the generation of the next generation individual data. In this way, the generation change is repeated.

  The next generation according to claim 2, wherein the master generates the current generation individual data having the best evaluation value by genetic manipulation with respect to the slaves that returned the evaluation value calculation result of the majority or more as in claim 3. More than half of the next generation individual data including the individual data and the next generation individual data for which the evaluation value has not been calculated may be assigned.

  The next-generation individual data generated by genetically manipulating the current generation data having the best evaluation value in the order of slaves in which the master is expected to have a high evaluation value calculation processing speed. More than half of the next generation individual data including the next generation individual data for which the evaluation value has not been calculated may be assigned.

  As in claim 5, the master determines that the total value of the next generation individual data generated by genetic operation of the current generation individual data having the best evaluation value and the next generation individual data for which the evaluation value has not been calculated is all the next generations. It is better to make it a majority of individual data.

  The invention of claim 6 is a train operation simulation method for simulating train operation using the genetic algorithm processing method according to any one of claims 2 to 5, wherein the master is a train operation The change pattern of the train operation diagram of the section is generated as the individual data of the current generation and assigned to each slave, each slave calculates the total value of the crossing time for the entire train operation section as an evaluation value of the individual data, Using the genetic algorithm processing method, genetic operation is performed to minimize the evaluation value, and the change in the total value of the level crossing interruption time is simulated. It is characterized by searching.

  The invention of claim 7 is a train operation simulation method for simulating train operation using the genetic algorithm processing method according to any one of claims 2 to 5, wherein the master is train operation The change pattern of the train operation diagram of the section is generated as the individual data of the current generation and assigned to each slave, each slave calculates the total value of the level crossing interruption time of the entire train operation section as an evaluation value of the individual data, Using a genetic algorithm processing method, genetic operations are performed so that the evaluation value is maximized, and the change in the total value of the level crossing interruption time is simulated. It is characterized by searching for the maximum value.

  According to the processing method of the genetic algorithm of the present invention, when the evaluation value calculation result of the majority or more is obtained, the individual data generation processing of the next generation is executed, so the evaluation values of all the individual data of the same generation Generation changes can proceed without waiting for calculation results, and the optimal solution search processing speed can be improved without degrading the characteristics of inheriting the properties of the best individual data of the current generation to the next generation, reducing the optimal solution search time. it can.

  Also, in the master / slave system, there is no need to wait for the evaluation value calculation results of all the slaves. Therefore, the slave whose evaluation value calculation is slow due to the influence of local processing or performance degradation, or the slave whose evaluation value calculation result cannot be returned due to a failure. Even if there is, the genetic algorithm can be executed without being affected by these slaves, and the processing speed of the optimum solution search will not be reduced.

  According to the train operation simulation method of the present invention, it is possible to easily search for a train operation diagram that minimizes the total railroad crossing interruption time in the train operation section. Moreover, it becomes possible to easily search for the maximum value of the total railroad crossing interruption time in the train operation section when the train operation diagram is disturbed.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing an embodiment of an optimal solution search system to which the genetic algorithm processing method of the present invention is applied.
In FIG. 1, an optimal solution search system according to this embodiment includes one master computer (hereinafter referred to as a master) 1 and N (N = 2m−1: m) connected to the master 1 via a network. ≧ 2) slave computers (hereinafter referred to as slaves) 2 ₁ to _{2 2m−1} . The slaves 2 ₁ to _{2 2m-1} are divided into slave groups 2 ₁ to 2 _m that are expected to have a high processing speed and slave groups 2 _{m + 1} to _{2 2m-1} that are not expected to have a high processing speed. Register with Master 1.

The master 1 generates N pieces of individual data [1] to [2m−1] of the initial generation using, for example, random numbers, and the individual data based on the evaluation value calculation result of the individual data returned from the slave. An individual data generation process for generating next-generation individual data by performing a genetic operation (selection, crossover, mutation, etc.) is executed. The master 1 assigns the generated individual data to each slave 2 ₁ to 2 _2m-1.

Each slave 2 ₁ to 2 _2m-1, in addition to the evaluation value calculation processing for calculating the evaluation value of the individual data allocated from the master 1, executes the specific local processes assigned in advance. The evaluation value calculation process is executed at each timing between the local processes and is returned to the master 1.

The optimal solution searching system, repeating the evaluation value calculation processing for parallel computation of the evaluation value by the individual data generation processing and N of slaves 2 ₁ to 2 _2m-1 by genetic operations based on the evaluation value calculation result in the master 1 The optimal solution is searched using a genetic algorithm that repeats generation changes.

Specifically, when the evaluation value calculation result of more than a majority of the individual data of the current generation is returned from the slave, the master 1 performs genetic operation on the current generation individual data of the majority of the evaluation values returned. run the individual data generation processing, assigned a current generation individual data of the evaluation value uncalculated an individual data generated in this process as a next generation individual data to the slave 2 ₁ ~2 _2m-1, slave 2 ₁ to 2 _{2m -1} calculates the evaluation value of the next generation individual data. The master 1 receives the evaluation value calculation result of more than a majority including the next generation individual data generated by genetic manipulation of the current generation individual data with the best evaluation value and the evaluation value result of the next generation individual data whose evaluation value is not calculated from the slave. At the time of the reply, the generation change is repeated so as to generate the next generation individual data in the same manner as the generation of the next generation individual data, and the optimum solution is searched.

Next, the optimum solution search operation of this embodiment will be described in detail with reference to FIGS.
FIG. 2 is a flowchart for explaining the processing operation of the master 1, and FIG. 3 is an explanatory diagram showing the process of generation change.
In step 1 of FIG. 2 (indicated by S1 in the figure, the same shall apply hereinafter), the master 1 generates individual data [1] to [2m−1] of the initial generation as shown in FIG. 3 using random numbers, Assign to each slave 2 ₁ to _{2 2m-1} . Each of the slaves 2 ₁ to _{2 2m-1} calculates the evaluation value of the individual data [1] to [2m-1] assigned at its own timing during the local processing, and sends it back to the master 1.

In Step 2, it is determined whether the calculation result of the evaluation values from all the slave 2 ₁ to 2 _2m-1 sent back, the flow proceeds to Step 3 If it is determined that YES.
If there is a slave to which the evaluation value calculation result is not returned even after the predetermined time has elapsed, another slave that has already returned the evaluation value calculation result performs the evaluation value calculation of the individual data, and Get the calculation result.

In step 3, genetic operations are performed to generate new 2m-1 first generation individual data [1] to [2m-1] as shown in FIG. Specifically, for example, 2m-1 first generation individual data [1] to [2m-1] are generated by a genetic operation for crossing two initial generation individual data according to the obtained evaluation value. Assign the generated first generation individuals Data [1] ~ [2m-1 ] to each slave 2 ₁ to 2 _2m-1, and waits for a reply from the slave 2 ₁ to 2 _2m-1.
In step 4, it is determined whether or not m evaluation value calculation results, which are the majority, are returned, and when m evaluation value calculation results are returned, the determination is YES and the process proceeds to step 5. Normally, an evaluation value result is returned from the slave group expected to have a high processing speed in FIG.

In step 5, the same genetic operation as in step 3 is performed on the majority m individual data to which the evaluation value calculation results are returned without waiting for the remaining evaluation value calculation results, and a new one as shown in FIG. M individual data [1] to [m] are generated. Here, the data [m] is the individual data generated by genetically operating the first generation individual data of the best evaluation value among the returned evaluation values (for example, the individual data with the best evaluation value calculation result and 2 (Individual data generated by crossing the second best individual data). Further, the remaining (m−1) first generation individual data whose evaluation values have not been calculated are assumed to be data [m + 1] to [2m−1]. Then, new individual data [1] to [m] generated by genetic manipulation and individual data [m + 1] to [2m−1] which are first generation individual data whose evaluation values are not calculated are shown in FIG. a (2m-1) individual data [1] the second generation of the number and ~ [2m-1], assigned to each slave 2 ₁ to 2 _2m-1, the reply from the slave 2 ₁ to 2 _2m-1 wait.

As a method of assigning the second generation individual data [1] to [2m−1] to the slaves 2 ₁ to _{2 2m−1} , data [m + 1] to [2m−1] whose evaluation values are not calculated from the data [m] are used. The majority of the second generation individual data up to m are assigned to the m slaves that have already returned the evaluation value calculation results, and the (m-1) th of the remaining data [1] to [m-1]. Allocate the 2nd generation individual data to the remaining slaves that could not wait for the reply of the evaluation value calculation result.
Further, a majority of m second generation individual data from data [m] to data [m + 1] to [2m−1] whose evaluation values have not been calculated are assigned in the order of slaves expected to have a high processing speed, and the remaining ( m-1) The second generation individual data may be assigned to the remaining slaves.

In step 6, it is determined whether evaluation value calculation results of the second generation data [m] and evaluation value uncalculated data [m + 1] to [2m−1] are returned, and the evaluation values of these individual data are determined. When the reply is made, the determination is YES, and the process proceeds to step 7.
In step 7, it is determined whether there is an evaluation value that satisfies a preset condition among the returned evaluation values. Then, until there is an evaluation value that satisfies a preset condition and the determination is YES, the operations in steps 5 to 7 are repeated to repeat the generation change. Although FIG. 3 shows up to the third generation, generation change is performed in the same manner from the third generation onward.

  Note that FIG. 3 shows that the majority of m individual data from the data [m] to the data [m + 1] to [2m−1] whose evaluation values have not been calculated is from the remaining individual data [1] to [m−1]. Although shown as being sent back quickly, the evaluation values of other individual data are sent back before the data [m + 1] to [2m-1] from which the evaluation value has not been calculated is sent back. It may be possible. In this case, in step 5, individual data is generated by performing a genetic operation including the other individual data. For example, when the number of returned other individual data is α, (m + α) individual data is newly generated by genetic operation. Then, ((m−1) −α) individual data for which the evaluation value has not been calculated and the individual data having the best evaluation value among the returned evaluation values, which could not wait for the reply of the evaluation value calculation result, are genetically determined. (Α + 1) out of the (m + α) generation data is generated using the individual data in the order of good evaluation value so that the total of the individual data generated by operation becomes a majority of m. Then, m pieces of individual data composed of the (α + 1) evaluation value best individual data and the evaluation value uncalculated ((m−1) −α) individual data are arranged in the order of slaves with a high processing speed, or Then, the generation change is performed by assigning the evaluation value to the returned slave and assigning the remaining individual data to other slaves.

FIG. 4 is a diagram showing a comparison result between the conventional method and the method of the present invention, taking as an example a search for a train diagram that minimizes the total cut-off time of all railroad crossings existing in one train operation section.
In this case, the evaluation value is the total value (total interruption time) of the interruption times at each level crossing. The individual data is a list of station departure times and station arrival times of all trains traveling in the corresponding train operation section, and a plurality of pieces of data are generated by changing the time variously with respect to the plan diagram.

  In FIG. 4, the vertical axis represents the evaluation value, and the horizontal axis represents the total evaluation value calculation number. N represents the number of individual data, and m represents the number of evaluation value calculations. In the figure, N = 30, m = 30 and N = 59, m = 59 indicate the case of the conventional method in which genetic operations are performed after waiting for the reply of the evaluation value calculation results of all individual data, N = 59, m = 30 indicates the case of the method of the present invention in which the genetic operation is performed when 30 evaluation value calculation results, which are the majority of the 59 individual data numbers, are returned. In this example, the smaller the evaluation value (total crossing time of the crossing) is better.

  As can be seen from FIG. 4, the method of the present invention can obtain a smaller evaluation value with a smaller number of evaluation value calculations than the conventional method, and can search for the optimum solution in a shorter time than the conventional method. It can also be seen that the convergence of the evaluation value is not inferior to the conventional method.

  When the processing speed of the slowest slave in the slave group that is expected to be high in processing speed is T1, and the processing speed of the slowest slave in the slave group that is not expected to be high in processing speed is T2, at least T1 / When T2 = 0.5 or less, in other words, when the processing speed of the slowest slave group is higher than the processing speed of the slowest slave group of the slowest processing group, the optimal solution is obtained. Search efficiency is improved.

FIG. 5 shows a comparison result between the method of the present invention and the conventional method with respect to convergence when the number m of evaluation value calculations is constant and the number N of individual data is changed.
In FIG. 5, N = 60, m = 30 indicates a case where 60 individual data are divided into two without overlapping, and a genetic operation is performed on each independently, and N = 40, m = 30 N = 48 and m = 30 indicate cases where the number of evaluation values calculated at the start of the genetic operation is larger than the majority.

  From FIG. 5, it can be seen that when the number of evaluation value calculations m is constant, the evaluation value decreases the earliest when the number of individual data N is (2m−1), which is optimal. In other words, in the method of the present invention, the optimal solution search efficiency is best when the genetic operation is performed when the evaluation value of the majority of the individual data is calculated.

FIG. 6 shows a comparison of convergence when the genetic data is executed based on the same calculation method as in FIG. 3, with the number of individual data N being an odd number and the number m of evaluation value calculations being just a majority (N = 2m−1). Is shown.
Focusing on the region A close to the approximate convergence value in FIG. 6, it can be seen that the convergence is good when the number m of evaluation value calculations per generation is reduced and the number of generations is increased. In the convergence value region B, there is no difference in convergence between when the number of evaluation value calculations is reduced and the number of generations is increased, and when the number of evaluation value calculations is increased and the number of generations is reduced. The total evaluation value calculation number is (evaluation number calculation number per generation) × (number of generations).

  As is clear from the above, the use of the processing method of the present invention makes it possible to shorten the optimal solution search time compared to the conventional method and improve the efficiency of optimal solution search using a genetic algorithm. In addition, even if there are slaves whose evaluation value calculation results are delayed or cannot be returned on the slave side that performs evaluation value calculation, the genetic algorithm can be executed without being affected by these slaves, and the optimal solution search process The speed does not decrease.

  In the above embodiment, a configuration example in which a plurality of computers are connected via a network has been described. However, when a plurality of CPUs are mounted in one computer, a plurality of processes are substantially performed by one CPU. In the present invention, any computer system capable of substantially parallel processing by a plurality of processing means (CPUs) can be applied.

Next, the train simulation method of the present invention using the genetic algorithm processing method described above will be described.
As a train simulation apparatus to which the train simulation method of the present invention is applied, for example, data such as a train operation diagram may be input from the train operation management apparatus to the master 1 in FIG. Each slave may use a station terminal device installed at each station, for example. According to such a configuration, there is an advantage that the train simulation system can be easily introduced, the existing facilities can be used effectively, and the cost for introducing the system can be reduced.

A train simulation method for searching for a train schedule that minimizes the total cut-off time of all railroad crossings in the train operation section will be described below.
In this case, as shown in FIG. 7, the individual data is, for example, a change pattern of departure time at each station of all trains traveling in the target train operation section. Specifically, as shown in FIG. 7, all the individual data strings indicating the plan diagram are set to 0, and a change pattern in which the departure time of each station is slightly changed is generated and used as the individual data string. In FIG. 7, “0” indicates the plan diagram as it is, “+1” indicates that it is delayed by 1 second from the plan diagram, and “−1” indicates that it is 1 second earlier than the plan diagram. In FIG. 7, there are only three pieces of individual data, but (2m-1) pieces of such data are generated. The (2m-1) pieces of individual data generated in this way are used to repeat the generation change of the individual data according to the flowchart of FIG. Search for a train schedule that minimizes the total crossing time at the level crossing.

  In the train simulation method of this embodiment for searching for a train schedule that minimizes the total crossing time of a level crossing, a change pattern of a train operation diagram as shown in FIG. 7 is generated as individual data as the initial generation of FIG. To do. The evaluation value calculation result is the total value of the level crossing interruption time, and the smaller the better. Therefore, the second generation individual data [m] is generated by crossing the individual data having the smallest evaluation value and the next smallest individual data. Then, for example, a threshold value of an evaluation value is set in advance, and when an evaluation value equal to or lower than the threshold value is obtained, the determination in step 7 is YES and the genetic algorithm is terminated. In this case, the individual data having an evaluation value equal to or less than the threshold value is a train diagram that minimizes the total crossing time for a crossing.

Next, a simulation method for calculating the fluctuation range of the railroad crossing interruption time due to diamond disturbance will be described. In this case, the maximum total crossing time of the crossing in the train operation section is searched.
Below, the example which simulates the crossing time reduction diagram which makes the total crossing interruption time searched by the above-mentioned train simulation the minimum is demonstrated.
In this case, for the individual data, the crossing time reduction diagram searched by the above-mentioned simulation is used as a reference diagram. Therefore, the data sequence of the schedule diagram in FIG. 7 may be replaced with the data sequence of the crossing time reduction diagram, and the individual data sequence indicating the crossing time reduction diagram is set to 0, and the departure time of each station is slightly changed. A change pattern is generated as an individual data string.

  After searching for a railroad crossing time reduction diagram by train simulation, (2m-1) pieces of individual data generated based on the railroad crossing time reduction diagram are processed using the genetic algorithm processing method of the present invention described above with reference to FIG. According to the flow chart, the generational change of the individual data is repeated, and the change in the crossing interruption time is calculated. In this case, since the maximum crossing time is searched, the larger the evaluation value, the better. Therefore, the second generation individual data [m] is generated by crossing the individual data with the largest evaluation value and the next largest individual data. Then, for example, an evaluation value threshold value is set in advance, and when an evaluation value equal to or higher than the threshold value is obtained, the determination in step 7 is YES and the genetic algorithm is terminated. In this case, an evaluation value equal to or greater than the threshold value can be searched with the total crossing time as a maximum value.

FIG. 8 shows a comparison result between the conventional genetic algorithm processing method and the processing method of the present invention for calculating the fluctuation range of the railroad crossing interruption time due to the diamond disturbance.
FIG. 8 shows the conventional method and the present invention with respect to the convergence when searching for the maximum level crossing cut-off time in the case where a time disruption occurs in a time crossing time reduction diagram searched using the genetic algorithm processing method of the present invention. The comparison result with the method is shown. In the figure, the evaluation value is the level crossing interruption time (seconds).

In the figure, N = 99, m = 99 indicates the case of a conventional method in which genetic operations are performed after reply of evaluation value calculation results of all individual data, and N = 99, m = 50 indicates the number of individual data The case of the method of the present invention in which the genetic operation is performed when a majority of the evaluation value calculation results are returned.
As shown in FIG. 8, there is no difference in convergence between the conventional method and the method of the present invention, and it can be seen that the processing method of the present invention is effective for this search case.

The block diagram which shows one Embodiment of the optimal solution search system to which the processing method of the genetic algorithm of this invention is applied. Flowchart explaining the processing operation of the master of the same embodiment Explanatory drawing which shows the process of generation change by embodiment same as the above. The figure which shows the comparison result of the conventional method and the method of this invention in the search example of the train diagram which minimizes the total value of level crossing interruption time The figure which shows the comparison result of the method of this invention and the conventional method about the convergence when the number of evaluation value calculation is fixed and the number of individual data is changed The figure which showed the comparison of convergence when the number of individual data in the method of the present invention is changed Explanatory drawing of an example of individual data used for train diagram search that minimizes the total value of level crossing interruption time The figure which shows the comparison result of the conventional method and the method of this invention about the convergence property at the time of the fluctuation range search of the level crossing interruption time at the time of diamond disturbance

Explanation of symbols

1 Master 2 ₁ to _{2 2m-1} Slave

Claims (7)

An evaluation value calculation process for calculating evaluation values of a plurality of individual data of the current generation, and performing a genetic operation on the individual data of the current generation based on the evaluation values obtained by the evaluation value calculation process, In the genetic algorithm processing method of searching for the optimal solution by performing generation change by repeating individual data generation processing for generating individual data,
When calculating the evaluation value of the majority of the current generation individual data in the evaluation value calculation processing, the individual data generation processing is executed for the current generation individual data of the majority of the evaluation values obtained, The individual data generated by the solid data generation process and the current generation individual data for which the evaluation value has not been calculated are set as the next generation individual data, the evaluation value calculation process is executed for the next generation individual data, and the current generation individual having the best evaluation value When calculating more than half of the evaluation values including the next generation individual data generated by genetic manipulation of the data and the evaluation value result of the next generation individual data for which the evaluation value has not been calculated, the same as the generation of the next generation individual data. In addition, the genetic algorithm processing method is characterized in that generational change is repeated by generating individual data of the next generation. A master that performs individual data generation processing for generating a plurality of next-generation individual data by performing a genetic operation of the individual data based on each evaluation value of a plurality of individual data of the current generation, and assigned from the master Using the plurality of slaves that execute the evaluation value calculation process for calculating the evaluation value of the individual data at their own timings and send back to the master, the generation data is changed by repeating the individual data generation process and the evaluation value calculation process. In the processing method of the genetic algorithm that searches for the optimal solution by performing
When the master returns an evaluation value calculation result of more than a majority of the individual data of the current generation from the slave, the individual data generation process for the current generation individual data of the majority of evaluation values returned The individual data generated by the solid data generation process and the current generation individual data for which the evaluation value is not calculated are assigned to the plurality of slaves as the next generation individual data, and the current generation individual data with the best evaluation value is genetically When the next generation individual data generated by operation and the evaluation value calculation result of the majority including the evaluation value result of the next generation individual data for which the evaluation value has not been calculated are returned from the slave, the next generation individual data generation and Similarly, the genetic algorithm processing method is characterized in that generational change is repeated by generating individual data of the next generation.   Next-generation individual data generated by genetically manipulating the current generation individual data with the best evaluation value and the next-generation individual for which the evaluation value has not been calculated, for the slave to which the master has returned the evaluation value calculation result of the majority or more The genetic algorithm processing method according to claim 2, wherein a majority of next-generation individual data including data is assigned.   Next, the next generation individual data generated by genetically manipulating the current generation data having the best evaluation value and the next generation individual data for which the evaluation value has not yet been calculated, in the order of slaves in which the master is expected to have a high evaluation value calculation processing speed. 3. The genetic algorithm processing method according to claim 2, wherein a majority of the next generation individual data including at least one is allocated.   In the master, the total value of the next generation individual data generated by genetic operation of the current generation individual data with the best evaluation value and the next generation individual data for which the evaluation value has not been calculated is the majority of all the next generation individual data. The genetic algorithm processing method according to any one of claims 2 to 4, characterized in that: A train operation simulation method for simulating train operation using the genetic algorithm processing method according to any one of claims 2 to 5,
The master generates a change pattern of the train operation diagram of the train operation section as the individual data of the current generation and assigns it to each slave, and each slave is a total of the crossing interruption time of the entire train operation section as an evaluation value of the individual data The value is calculated, and a genetic operation is performed using the genetic algorithm processing method so as to minimize the evaluation value, thereby simulating a change in the total value of the level crossing interruption time. A train operation simulation method characterized by searching for a train operation diagram. A train operation simulation method for simulating train operation using the genetic algorithm processing method according to any one of claims 2 to 5,
The master generates a change pattern of the train operation diagram of the train operation section as the individual data of the current generation and assigns it to each slave, and each slave is a total of the crossing interruption time of the entire train operation section as an evaluation value of the individual data A level crossing in the event of a diagram disruption is calculated by simulating a change in the total value of the level crossing cut-off time by performing a genetic operation to maximize the evaluation value using the processing method of the genetic algorithm. A train operation simulation method characterized by searching for the maximum value of total shut-off time.

Images