JPH11272638A - Data processor utilizing genetic algorithm and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a genetic algorithm considering a building block by setting the plural combinations of the building blocks being the partial array pattern of chromosomes to be the processing objects of the genetic algorithm. SOLUTION: The plural combinations of the building blocks being the partial array pattern of chromosomes are set to be the processing objects of the genetic algorithm. In this processing, a step 11 sets the initial value of Meta Chromosomes by random numbers. A step 12 decodes the Meta Chromosomes to chromosomes, decodes the chromosomes to the parameter of an optimizing system in addition and calculates an evaluation value concerning the system of the parameter. A step 14 selects the Meta Chromosomes by the evaluation value. A step 15 crosses the Meta Chromosomes selected by the step 14. The step 16 expresses a part to mutate with respect to the Meta Chromosomes of a new generation generated y being crossed by the step 15.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a genetic algorithm which is an established solution of an optimization problem. Many intellectual processes such as timetable creation using a computer, vehicle assignment problems, and personnel planning problems can be treated as optimization problems. Among these optimization problems, a number of stochastic optimization methods (simulated annealing, genetic algorithm) have been devised as new methods to solve large-scale problems that were difficult to solve with conventional expert systems. However, genetic algorithms have received special attention and are growing into practical methods.

In recent years, with the cost reduction of computers, it is possible to effectively use abundant CPU power. However, genetic algorithms are regarded as a promising method to overcome problems with computer power.

[0003]

2. Description of the Related Art In a conventional genetic algorithm, various parts of a chromosome (referred to as schemata) evolve in parallel in a population, in which schemata (particularly such schemata) contribute to the improvement of fitness. Building a simple schema
(Called blocks) spread more into the population. It is said that the successful combination of these building blocks evolves the chromosome population within the genetic algorithm, ultimately leading to suboptimal solutions. This is the building block hypothesis of the genetic algorithm.

[0004]

However, the genetic algorithm based on the conventional building blocks has the following problems. Building blocks appear spontaneously in the chromosome but are not explicitly present, and they are not actively controllable objects.

Furthermore, what constitutes a building block depends largely on the problem and coding, and is also greatly influenced by the arrangement of the alleles on the chromosome. For example, if two related alleles are next to each other, the two sets are more likely to evolve in a good combination, but apart from each other, more cross-separation occurs and the building block grows. The likelihood of doing so is lower than the former.

[0006] Also, as in this example, if it is known in advance what should be related, coding and gene sequence can be successfully determined for a simple problem. Are correlated, and it is difficult to properly set the coding and gene sequence. In more complex problems, it is not even possible to know what constitutes a building block and how, and it is not possible to properly configure coding and gene sequences to help the building block grow Becomes

[0007] The present invention has the above-mentioned weaknesses of the building blocks, that is, it does not know what forms the building blocks and cannot perform explicit control. -The building block depends on the coding, and there is a difference in the easiness of formation with the building block length. It actively complements this problem.

[0008]

Means for Solving the Problems A plurality of combinations of building blocks, which are partial sequence patterns of chromosomes, are processed by a genetic algorithm, whereby the building blocks can be controlled.

[0009]

FIG. 1 is a flow chart of the present invention. This figure is a process similar to a normal genetic algorithm, except that a target is a chromosome (hereinafter, referred to as a chromosome) to a meta chromosome (hereinafter, referred to as a meta chromosome).

In the figure, reference numeral 11 sets an initial value of a meta chromosome by a random number. In the figure, 12 decodes the meta chromosome into a chromosome,
Further, this section decodes the parameters of the system for optimizing the chromosome, and calculates the evaluation value of the system for the parameters. In the figure, reference numeral 13 denotes a part for receiving the evaluation value at 12 and determining the end of optimization. In the figure, reference numeral 14 denotes a portion for selecting a meta chromosome based on an evaluation value. In the figure, 15
Is the portion where the crossover of the metachromosome selected in 14 is performed. In the figure, reference numeral 16 denotes a portion for mutating a new generation meta chromosome generated by crossover at 15.

FIG. 2 is a diagram for explaining the principle of the present invention. In the figure, reference numeral 21 denotes a group of meta chromosomes, each meta chromosome designates one Base Chromosome (reference chromosome, hereinafter referred to as a base chromosome) and a plurality of Building Blocks (hereinafter, referred to as building blocks). ing. The number of groups is irrelevant to the present invention, and it is desirable to increase the number, but it is determined by a trade-off with the calculation time.

In the figure, reference numeral 22 denotes a base chromosome table, which is a table of base chromosomes from which a chromosome is generated at the time of meta-decoding (hereinafter referred to as meta-decoding) for converting a meta chromosome into a chromosome. The base chromosome has the same format as the chromosome 24, and is copied from the chromosome 24 according to the set rules, a new one is registered in the table, and is deleted from the table according to the set rules.

In the figure, reference numeral 23 denotes a building block table which collects parts for overwriting the base chromosome designated by the meta chromosome when generating the chromosome at the time of meta-decoding for converting the meta chromosome into a chromosome. It is a table. The building blocks are located on chromosome 24 according to the rules set.
Is extracted as a building block, a new one is registered in the table, and is deleted from the table according to the set rules.

In the figure, reference numeral 24 denotes a chromosome, which is obtained by meta-decoding the meta chromosome 21 and corresponds to a chromosome of a normal genetic algorithm. In the figure, 25 is P
arameters (hereinafter referred to as parameters), which are parameters of the system to be optimized. Obtained by decoding chromosome 24. FIG. 3 is a principle diagram-2 showing the details of 21, 22, 23 and 24 in FIG. In the figure, the head of each individual of the meta chromosome 31 indicates which base chromosome in the base chromosome table to use, and the part excluding the head indicates which building block in the building block table to use. ing.

The base chromosome table 32 is a set of chromosomes distinguished by subscripts. Each base chromosome has, as a counter (hereinafter, referred to as a counter), the number of references from the meta chromosome or a reference degree calculated by adding the degree of application to the number, and the base chromosome table is managed by this. The building block table 33
A set of building blocks distinguished by a subscript. In each of the building blocks, the portion for which the value is not determined is represented by “*”.

Each building block has, as a counter, the number referenced from the meta chromosome or the reference degree calculated in consideration of the degree of application, and manages the building block table. Chromosome 34 shows an example of meta-decoding of a meta chromosome, based on base chromosome 1 (a), building block 3 applied (b), 0 applied (c), 1 An example in which the number is applied (d) is shown.

In the present invention, meta-chromosomes are not treated as chromosomes as shown in FIG. In the meta chromosome evaluation, the portion shown in FIG. 2 operates, and each meta chromosome is meta-
Decoding is performed, and the generated chromosome population is converted into a parameter population in the same manner as a normal genetic algorithm to obtain a system evaluation function value.

This evaluation function value is returned as a metachromosome evaluation value. Meta-chromosomes are subjected to operations such as selection, crossover, mutation, and the like in the same manner as ordinary genetic algorithms. Each time a new generation is created, the counters in the base chromosome table 32 and the building block table 33 are recalculated, and missing or low-reference counters are removed from their respective tables, and the new Replenished in a way.

With such a mechanism, the basic part of the chromosome is updated with reference relation to the base chromosome, and the building blocks that are said to play a central role in optimization are stored in the building block table 33. Save with
Updated and combined through meta-decoding to help create something closer to the optimal solution. Building blocks are explicitly described in the building block table, and are controlled by frequency (number of references or degree of reference). In addition, since the crossover is not performed on the building block, it is irrelevant to the fragility depending on the building block length.

Hereinafter, a description will be given using specific data. As a simple example, consider the problem of maximizing a 4-bit chromosome. The evaluation function fitness is defined as fitness = x * y, where x, y = ｛0 (00), 1 (01), 2
(10), 3 (11)｝ At this time, the fitness range is an integer from 0 to 9. FIG. 4 illustrates an embodiment.

When the system is started, a series of initialization processing from [STEP 1] is performed. First, a metachromosome population is generated by random numbers [STEP2] (In this example, 0: 0,0: 21,0: 10
Was generated). At the same time, a chromosome population is generated with random numbers [STEP8] and evaluated [STEP9] (1001 = 2, 0110 =
2, 0001 = 0).

Base chromosome registration based on [STEP9] [STEP1]
0], and the base chromosome table 42 is initialized (here, one individual 0110). Based on [STEP9], register the building block [STEP11] and initialize the building block 41 (* 11 *, 0 ** 1, 1 * 0
*). With the above, the initialization processing ends, and the optimization processing starts. The optimization is performed by performing the same processing as the genetic algorithm on the meta chromosome.

In the meta-chromosome evaluation [STEP 3], the details are shown in FIG.
Individuals specified in the base chromosome table 42 described in part of the meta chromosome (the base chromosome of 0: 0 is 0110)
Is first copied to the decoded chromosome 51. next,
The building block is overwritten with the copied base chromosome according to the designation of the metachromosome building block, and a chromosome is created in which all designated building blocks are overwritten, which is the chromosome to be evaluated . This evaluation value is returned as an evaluation for each metachromosome.

Here, an end determination [STEP 4] is made. If the conditions are satisfied, the process proceeds to end [STEP 12], and the system ends. Meta-chromosome selection [STEP5] is performed (0:21, 0:10), and meta-chromosome crossover [STEP6] and meta-chromosome mutation [STEP7] are performed in the same manner as in the conventional genetic algorithm (here: In the example above, the meta chromosome 0:20 is overwritten with the building block 1 * 0 *, * 11 * on the base chromosome 0110, and becomes 0110 → 1100 → 1110, which shows or shows a better evaluation value. There is). This cycle is repeated to evolve metachromosomes with better sets of building blocks.

On the other hand, a specified generation (for example, 10 generations)
Each time, a building block is extracted from the decoded chromosome 51, registered, and the building block table 41 is updated [STEP 21]. Similarly, the base chromosome table 42 is updated [STEP 22]. FIG.
The procedure for updating the building block table 41 in [STEP 21] is shown below. The building block table 41 counts a specified number of times (in the example of the figure, the sum of the evaluation functions for each decoded chromosome is used as a reference level), and counts a low evaluation (in the example, the reference level). Discard building blocks below a certain threshold [S
TEP31]. Then, the plurality of decoded chromosomes 51 are statistically processed, and a new building block is extracted stochastically [STEP32] (for example, a building block having the highest frequency among a plurality of decoded chromosomes is selected. ), And register it as a new building block in the building block table 41 [STEP3
3].

FIG. 7 shows a procedure for updating the base chromosome table 42 in [STEP 22]. Base chromosome table 42
For each base chromosome, the number of times referred to during the evaluation of the meta chromosome in [STEP 3] is counted, and the base chromosome with a low evaluation is discarded [STEP 41]. Then, a plurality of decoded chromosomes 51 are statistically processed to stochastically extract a new base chromosome [STEP 42] and register it as a new base chromosome in the base chromosome table 42 [STEP 43].

[0027]

As described above, according to the present invention, a building system which cannot be controlled by the existing genetic algorithm is used.
Blocks can be explicitly handled, and at the same time, by extracting and handling building blocks, there is an effect that the difference in formability due to the building block length is small and the coding dependency of optimization performance is reduced.
In addition, since there is a mechanism for combining a plurality of building blocks, the optimization ability is improved.

[Brief description of the drawings]

FIG. 1 is an outline of a flowchart of the present invention.

FIG. 2 is a diagram illustrating the principle of the present invention.

FIG. 3 is a detailed explanatory view of the principle of the present invention.

FIG. 4 is an embodiment of the present invention.

FIG. 5 shows the details of the evaluation of metachromosomes in the example of the present invention.

FIG. 6 is a flowchart of updating a building block table according to the present invention.

FIG. 7 is a flowchart of updating a base chromosome table of the present invention.

[Explanation of symbols]

 21 Meta chromosome 22 Base chromosome table 23 Meta chromosome 24 Chromosome 25 Parameter

Claims (4)

[Claims] 1. A data processing apparatus using a genetic algorithm, wherein data of a plurality of combinations of building blocks, which are partial sequence patterns of chromosomes as genetic data, are to be processed by the genetic algorithm. A data processing device using a genetic algorithm. 2. A method according to claim 1, further comprising: frequency calculating means for determining a use frequency or a reference degree of said building block; and means for determining whether to adopt the building block based on the use frequency or the reference degree calculated by said frequency calculating means. A data processing apparatus using a genetic algorithm according to claim 1. 3. The data processing apparatus according to claim 1, wherein data to be processed by said genetic algorithm is obtained from a combination of a reference chromosome and said building block. 4. A computer-readable storage medium storing a program for performing data processing using a genetic algorithm, comprising: storing data of a plurality of combinations of building blocks, which are partial sequence patterns of chromosomes; A computer-readable storage medium storing a program for performing data processing using a genetic algorithm, which is a processing target of a genetic algorithm.