JP2000099489A - Method and system for searching product design solution using genetic algorithm - Google Patents

Abstract

PROBLEM TO BE SOLVED: To cancel the trouble of evaluating the evaluation value of any arbitrary evaluation item lower than real one. SOLUTION: This system is composed of a basic design information input part 1, genetic information generating part 2, expression type converting part 3, individual evaluating part 4, next generation solution candidate generating part 5, circulation processing part 6, in-progress information file picture output part 7 and weight control part 8, and the individual evaluating part 4 respectively executes a genetic algorithm only with the evaluation value based on a single evaluation index and performs processing for selecting and storing a maximum value among plural provided evaluation values at that time concerning respective plural evaluation indexes. Next, the genetic algorithm is executed through the evaluation value based on the plural evaluation indexes, the plural evaluation values provided at that time are normalized by the maximum value, and a value found by adding these normalized evaluation values of respective evaluation indexes with the preset ratio of respective evaluation values is defined as total application degree.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for calculating a total fitness of a product based on weights of evaluation values obtained from a plurality of evaluation indices, and selecting and duplicating individuals using the total fitness to generate a next generation solution. The present invention relates to a product design solution search method and a product design solution search system using a genetic algorithm that searches for a product design solution by repeating a process such as generating a candidate individual for a predetermined cycle.

[0002]

2. Description of the Related Art A genetic algorithm is an algorithm based on the inheritance of living organisms, and is one of search algorithms used for finding an optimal approximate solution in a wide solution space.

That is, an initial solution candidate (individual) is
A group is generated, and each individual is evaluated to obtain a fitness. Then, according to the fitness, crossover (selecting and combining two individuals to create a new individual), mutation (changing a part of the individual with random numbers), duplication (creating the same individual as the individual) ) To generate the next (next generation) solution candidate group. By repeating such an operation, the solution candidate group approaches an optimum value or an approximate value thereof.
An example of applying such a genetic algorithm to a structure design problem is, for example, "Generation of design candidates based on a group considering multi-value criteria and its fluctuations (for structures)" Collection No. 954-1 and the like.

[0004] When a genetic algorithm is used for product design having a plurality of evaluation indices (evaluation items), conventionally, a total obtained by multiplying each evaluation item by weight with priority is defined as a total fitness, Individuals were selected or duplicated using the fitness, and a genetic algorithm was executed to derive an approximate solution (suboptimal solution) of the optimal solution.

[0005]

The weight of the evaluation value of each evaluation item when calculating the total fitness is determined by the ratio of the logical maximum value. However, in actuality, when there is an evaluation item that can reach only half of the logical maximum value, the evaluation value of the evaluation item is underestimated, and the overall evaluation value (weighting Fitness). Therefore, there is a problem that a desired search solution cannot be obtained.

More specifically, for example, the logical maximum value for three types of evaluation items (flow path, weight, and center of gravity) is set to 10 points, and each of them is actually used individually (in a single evaluation item) by a genetic algorithm. Assuming that the maximum value of the evaluation values obtained when is executed is 8 points for the flow path, 5 points for the weight, and 10 points for the center of gravity, all the weight coefficients are the same (1: 1: 1).
, The maximum logical total evaluation value is 30 points (=
10 + 10 + 10), and the genetic algorithm operates to approach this value.

In such a system having a plurality of evaluation indices (evaluation items), the respective evaluations are mutually related and often contradictory, and in such a case, the total of the maximum values of the single evaluation values is used. Points (= 8 × 1 + 5 × 1 + 10 × 1)
A lower value (for example, 20 points) is the highest point of the actual overall evaluation value.

Here, for example, an individual A having a score of 7 points, 2 points, and 8 points (a total of 17 points) for the flow path, weight, and center of gravity, respectively.
And the score of the flow path, weight and center of gravity are 5 points, 4 points,
When there are 7 points (16 points in total) of the individual B, the genetic algorithm in the case where the weight is determined by the ratio of the logical maximum value has a high probability that the individual of the A is selected. If you continue to select and continue to optimize, you end up with, for example, 8
Results such as points, three points, and nine points (a total of 20 points) appear as suboptimal solutions.

However, when the actual maximum values (flow path: 8 points, weight: 5 points, center of gravity: 10 points) and the respective weighting factors are considered, the actual maximum value of the weight (5 points) is considered. Close 4
The probability that a B individual having a point should be selected must be high. That is, when the logical maximum value is used, the individual B
Will be underestimated.

The present invention has been made in order to solve such a problem, and an object of the present invention is to solve a problem that an evaluation value of an arbitrary evaluation item is underestimated below the actual value. An object of the present invention is to provide a product design solution search method and search system using a genetic algorithm.

[0011]

In order to solve the above-mentioned problems, a method for searching for a product design solution using a genetic algorithm according to claim 1 of the present invention comprises the steps of: The product's design fitness is calculated by repeating the process of calculating the total fitness of the product based on the evaluation index of the product, selecting and duplicating individuals using the total fitness to generate individuals of next-generation solution candidates, and repeating it in a predetermined cycle. A product design solution search method that uses a genetic algorithm to search for a product algorithm, in which the genetic algorithm is executed using only the evaluation values based on a single evaluation index, and normalized from among a plurality of evaluation values obtained at that time. Process of selecting and storing evaluation values for each of the plurality of evaluation indices, and then executing the genetic algorithm with the evaluation values of the plurality of evaluation indices. The normalized plurality of evaluation values are normalized by the evaluation values for normalization, and the values obtained by adding the normalized evaluation values of the respective evaluation indices at a predetermined ratio of the respective evaluation values are obtained by the general application. Degree.

According to a second aspect of the present invention, there is provided a method for searching for a product design solution using a genetic algorithm, comprising the steps of determining the total fitness of a product based on a plurality of evaluation indices for each of a plurality of individuals generated as a group. Product design using a genetic algorithm that searches for product design solutions by repeating the process of calculating, selecting, and duplicating individuals using their overall fitness to generate individuals for next-generation solution candidates for a predetermined cycle A solution search method, in which the genetic algorithm is executed only with the evaluation value based on a single evaluation index, and processing such as storing the maximum value of the plurality of evaluation values obtained at that time is performed for each of the plurality of evaluation indexes. Is performed, and then a genetic algorithm is executed with evaluation values based on a plurality of evaluation indices, and the plurality of evaluation values obtained at that time are normalized with the maximum value. The evaluation value of each evaluation index is a value obtained by adding the ratio of each evaluation value that is set in advance as to the overall applicability.

[0013] A product design solution search system using a genetic algorithm according to claim 3 of the present invention determines the overall fitness of a product based on a plurality of evaluation indices for each of a plurality of individuals generated as a group. A total fitness calculating means for calculating, a next generation solution candidate generating means for selecting and duplicating an individual using the calculated total fitness to generate an individual for a next generation solution candidate, and A product design solution search system using a genetic algorithm comprising a generation solution candidate generating means and a cyclic processing means for repeating a predetermined cycle repeatedly, each executing the genetic algorithm only with an evaluation value by a single evaluation index, Single evaluation in which a process of selecting a normalization evaluation value from a plurality of evaluation values obtained at that time and storing it in a storage unit is performed for each of a plurality of evaluation indices. A total evaluation value calculation unit, a plurality of evaluation values obtained when the genetic algorithm is executed with the evaluation values based on the plurality of evaluation indices, the plurality of evaluation values obtained by the normalization evaluation value stored in the storage unit. , And a value obtained by adding the normalized evaluation value of each evaluation index by a predetermined ratio of each evaluation value is calculated as a total applicability.

According to a fourth aspect of the present invention, there is provided a product design solution search system using a genetic algorithm, wherein a total fitness of a product is determined based on a plurality of evaluation indices for each of a plurality of individuals generated as a group. A total fitness calculating means for calculating, a next generation solution candidate generating means for selecting and duplicating an individual using the calculated total fitness to generate an individual for a next generation solution candidate, and A product design solution search system using a genetic algorithm comprising a generation solution candidate generating means and a cyclic processing means for repeating a predetermined cycle repeatedly, wherein the genetic algorithm is executed only with an evaluation value by a single evaluation index, and A single evaluation processing means for performing a process of storing the maximum value of the plurality of evaluation values obtained in the storage means for each of the plurality of evaluation indices; The sensitivity calculation unit normalizes a plurality of evaluation values obtained when the genetic algorithm is executed with the evaluation values based on the plurality of evaluation indices, using a maximum value stored in the storage unit. The value obtained by adding the evaluation values of the evaluation indices based on a predetermined ratio of each evaluation value is calculated as the overall applicability.

[0015]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the overall configuration of a product design solution search system using a genetic algorithm according to the present invention.

This product design solution search system is roughly divided into a basic design information input unit 1, a genotype information generation unit 2, a phenotype conversion unit 3, an individual evaluation unit 4, a next generation solution candidate generation unit 5,
It comprises a circulation processing unit 6, an intermediate progress information file screen output unit 7, and a weight control unit 8.

The basic design information input unit 1 includes a constraint condition for product design, a required quality serving as a reference of a quality evaluation index, and a weight coefficient (e.g., a flow path) of an evaluation value of each evaluation index controlled by the weight control unit 8. : Weight: Center of gravity =
Basic information such as 1: 1: 1) is input.

The genotype information generator 2 classifies product design issues into information that can be handled as parameters and information that can be handled as options in combination, and labels each of the classified information to indicate the meaning of the information. The gene is attached to the gene, and a plurality of individuals are generated based on the constraint conditions input from the design basic information input unit 1. As shown in FIG. 3, the labeled genotype information generated by the genotype information generation unit 2 is a set of a label (lowercase alphabet) and a numerical value, and forms one piece of information. One individual (1
Product configuration). That is,
The label plays the role of a gene, and a plurality of product components can be treated as one gene sequence. As for the labeled genotype information, the applicant has already filed a separate application (Japanese Patent Application No. 10-212896) and is not an essential component in the present invention. Described, and other detailed descriptions are omitted.

The phenotype conversion unit 3 includes a genotype information generation unit 2
Is converted into a product quality evaluation index and a product shape.

The individual evaluation unit 4 calculates the evaluation value of each product converted by the phenotype conversion unit 3 as an overall fitness in accordance with the weight given to the quality evaluation index, and based on the overall fitness. Calculate the selection probability of the individual.

More specifically, the individual evaluation unit 4
Has a fitness calculating means 41, a scaling means 42, and a selection probability calculating means 43, as shown in FIG. The fitness calculating means 41 calculates the value of the quality evaluation index (value indicating the product quality) for each product converted by the phenotype conversion unit 3, and calculates the required quality (target value, reference value, standard value). Value) is calculated for each quality evaluation item (evaluation index) and used as an individual fitness (individual evaluation value). Then, after the individual fitness (individual evaluation value) is normalized by a method which is a feature of the present invention (this will be described later), a weight coefficient (for example,
The overall fitness (comprehensive evaluation value) is calculated in consideration of the flow path: weight: center of gravity = 1: 1: 1, etc.). Scaling means 42
Scales the total fitness (comprehensive evaluation value) of each individual calculated by the fitness calculating means 41 so that the difference between the individuals increases. Scaling is to change the individual difference of the calculated overall fitness and reflect it on the probability at the time of selection, and various methods (for example, a simple method of squaring the overall fitness of each individual, etc.) are available. However, detailed description is omitted here. In the present embodiment, the selection probability is a partial total of the expected values. The selection probability calculation means 43 calculates the selection probability of each individual based on the total fitness (comprehensive evaluation value) after scaling.

The next-generation solution candidate generation unit 5 selects an individual to be a parent of the next generation based on the selection probabilities of the individual products calculated by the individual evaluation unit 4, and calculates the number of intersections of the genotype information of the individual and A crossover position is determined based on the selection probability and crossover is performed to generate a next-generation solution candidate individual, and the generated genotype information of the next-generation solution candidate individual is converted to a phenotype conversion unit 3.
To supply. In addition, in addition to the crossover, the mutation position of the genotype information of the selected individual is determined based on the selection probability and the mutation is performed to generate an individual of the next generation solution candidate. The genotype information of the individual is supplied to the phenotype conversion unit 3.

The circulation processing unit 6 includes these phenotype conversion units 3,
The circulation processing by the individual evaluation unit 4 and the next-generation solution candidate generation unit 5 is repeated until the number of generations of the individuals generated by the next-generation solution candidate generation unit 5 reaches a preset fixed number (for example, 300 generations). Or until a good solution spreads to some extent in the group. For example, among individuals in a population,
This is repeated until it is determined that a certain degree of solution has been found, such as a case where the total fitness exceeds 80% of the individuals is 40% or more.

The progress information file screen output unit 7 outputs “genotype information” and “product information” for each individual, which are the results calculated by the genotype information generation unit 2, phenotype conversion unit 3, and individual evaluation unit 4. "Each quality evaluation index and shape", "Fitness for each quality evaluation index", "Overall fitness", "Fitness for each quality evaluation index after scaling", "Overall fitness after scaling", and "Selection probability" Is output to the file as each generation information, and based on the total fitness of each individual, the variance of the entire population, the average fitness of the population, the individual number of the highest fitness in the population and the highest "Fitness" and "minimum fitness individual number and minimum fitness in the group" are calculated and output to the file as generation information. Also, the progress information file screen output unit 7
When the number of generations of individuals generated by the circulation processing unit 6 reaches a predetermined fixed number (for example, 300 generations), the individual information generated at this time is output as a product design result. The individual information is displayed as needed (for example, in the form of a numerical value, a figure, or the like) so that a person can see it.

The weight control unit 8 stores the statistical value (maximum value) of the evaluation value of each individual obtained after the execution of the genetic algorithm described later, and appropriately sets the weight of the evaluation item at the time of execution of the genetic algorithm. I do.

The above is a series of processing at the time of execution of the genetic algorithm by the product design solution search system. In the present invention, such a genetic algorithm is first used by the weight control unit 8 to independently evaluate each evaluation item. And a statistical value (maximum value) Mn is obtained from the evaluation value of each individual in each generation, and this is stored for each evaluation item. However, since the genetic algorithm at this time is a single evaluation item, it is natural that the individual fitness (individual evaluation value) calculated by the individual evaluation unit 4 for each evaluation item is not normalized. That is, the statistical value (maximum value) Mn actually obtained when each evaluation item is executed independently is stored in the weight control unit 8.

After the evaluation for all the evaluation items is completed in this way, next, each evaluation item is weighted again using the weight control unit 8, and the genetic algorithm is executed. At this time, the overall fitness (overall evaluation value)
The following formula (1) Σ ((Evaluation value of each evaluation item / Mn) × αn) (1) is used as a selection criterion. Here, αn is a weight coefficient of each evaluation item.

That is, since each evaluation value is normalized by “(Evaluation value of each evaluation item / Mn)” in the expression (1), αn is a pure weight (weight desired by the operator).

Specifically, using the individual A and the individual B taken up in [Problems to be Solved by the Invention], the maximum value Mn of each evaluation item obtained when the evaluation item is executed alone is: , 8 points for the flow path, 5 points for the weight, and 10 points for the center of gravity. Therefore, using the above equation (1), the individual A
The total fitness (comprehensive evaluation value) of the maximum value of the channel is 7 points,
Since the maximum value of the weight is 2 points and the maximum value of the center of gravity is 8 points, (7/8) × 1 + (2/5) × 1 + (8/10) × 1 =
2.075 points.

On the other hand, the overall fitness of individual B (overall evaluation value)
Since the maximum value of the flow path is 5 points, the maximum value of the weight is 4 points, and the maximum value of the center of gravity is 7 points, (5/8) × 1 + (4/5) × 1 + (7/10) × 1 =
2.125 points.

Therefore, in this case, the fixed B having four points close to the actual maximum value (5 points) of the weight is easily selected, and the probability of obtaining the shape of the weight desired by the operator is increased. Become. That is, the operator only has to consider the ratio of each evaluation value and input the weighting factor without worrying about the score of the evaluation value.

[0033]

Next, a description will be given of an embodiment in which a mass manhole, which is a part where a drain pipe is separated and merged, is actually designed in a product design solution search system using a genetic algorithm having the above configuration.

FIG. 4 shows the structure of a basic mass manhole. This mass manhole has a configuration in which an inlet and an outlet are provided on the side wall of a riser. The meaning of each symbol in the figure is as follows: Rmain: the size of the diameter of the riser FuH: the riser shape ( Pox, Poy): riser position FuEin1: shape of inflow port 1 FuEout: shape of outflow port P0 to P5: coordinates (calculated by genes t11 to t41) for expressing the flow path by a Bezier curve.

The design basic information input unit 1 includes an inflow position (P0), an inflow direction, an outflow position (P5) of an inflow pipe, a pipe size, a required quality (flow path, Weight, center of gravity), required quality weighting factors (e.g.,
Various parameters such as flow path: weight: center of gravity = 10: 10: 10) are input. Further, as parameters of a general genetic algorithm, a limit (maximum) number of generations, a crossover rate, a mutation rate, and a strategy (SGA, ER, MGG, etc.) are input.

The gene sequence of the genotype information generated under the conditions input from the design basic information input unit 1 is a sequence in which a label and a numerical value are set, and "(t1i, t)
(T2i, t) (t2θi, θ) (t3i, t) (t3
θi, t) (t4i, t) (FuH, H) (Fu
Eini, Ein) (FuEout, Eout) (Ty
pe, Type) ". Here, i is the number of inflow ports, and i = 1 in the above-mentioned mass manhole.

The sequence for storing the genotype information is as follows. p [individual No] [label No] [0] (label) p [individual No] [label No] [1] (numeric value) Therefore, for example, p [i] [j] [0] is , Indicates the j-th label No of the individual i, and p [i] [j] [1] indicates the individual i
Shows the j-th numerical value of.

FIG. 5 is a diagram showing the genotype information in a case where there are two inflow ports in a table format for easy understanding. Here, the genotype information when the inflow port shown in FIG.
10 only. That is, the massman hole having the shape shown in FIG. 4 is represented by ten pieces of genotype information.

Here, the X-Y coordinate values of each of the inlets P0 and P0 'in the case where there are two inlets (two inflow pipes) shown in FIG. 5 are P0 (0, 700) and P0', respectively. (700,
0), and the XY coordinate value of the outlet P5 is P5 (120
0, 1200), the genetic algorithm is executed, and the intermediate points P1 to P4, the position (center) and the size R of the standing pipe are obtained, and the obtained points are connected by a Bezier curve. Here, the XY coordinate values of P1 to P4 are obtained by the following equations.

[0040]

(Equation 1)

Here, δ and δ ′ are angles formed by P1 and P4, respectively. Each point P0 to P5 obtained in this way
Based on the above, the evaluation values of the three evaluation items of the flow path, the weight, and the center of gravity are obtained, these are scored, and the weighted sum is used as the total evaluation value. For the flow path, the maximum value of the amount of change of the vectors P0 to P5 is determined, and the evaluation value is determined using this maximum value. In other words, the more suddenly the flow path changes, the worse the flow of water becomes. Therefore, the larger the maximum value of the vector change, the lower the evaluation value. In addition, regarding the weight, an evaluation is made that the lighter the better, the less the material is (ie, the shorter the length, the lighter the better). Also, regarding the center of gravity, the route P0-P1-P2-P3-P4-P
The center of gravity of 5 is set as the position of the standing pipe, and the shortest distance from the path to the path is used as an evaluation criterion.

As a result, each of the logical evaluation values is
The flow path has an ideal maximum value of 3.14, a weight of 1, and a center of gravity of 1000.

Here, when normalized by the logical maximum value, the weighting coefficient of each evaluation value is calculated as follows: (flow path: weight: center of gravity = 10:
10:10) is a value obtained by dividing the weighting coefficient by the ideal maximum value, that is, (flow path: weight: center of gravity = 3.1).
8: 10: 0.01). When the genetic algorithm is executed using this coefficient, the score of the best individual is 8.320119 (channel) +9.113703 (weight) +9.192446 (center of gravity) = 26.62627 (total (Evaluation value).

On the other hand, when normalization is performed using the maximum value of each evaluation item obtained when the evaluation is performed for a single evaluation item, the weight coefficient of each evaluation value is expressed as (flow path: weight: gravity center = 10: 10: 10). When the coefficient is calculated as follows: (flow path: weight: center of gravity = 3.727: 10.0
0756: 0.01081), and when the genetic algorithm is executed using this coefficient, the score of the best individual is 8.330579 (flow path) +9.126039 (weight) +9.177536 (centroid) = 26.63415 (overall evaluation value) ). That is, it can be seen that the ratio of each evaluation value is improved and the overall evaluation value is also increased.

By the way, a graph obtained by connecting the XY coordinate values of the points P1 to P5 of the best individual obtained when normalized by the logical maximum value and the obtained points P1 to P5 by a Bezier curve is shown. The X-Y coordinate values of the points P1 to P5 of the best individual obtained when normalized by the maximum value of each evaluation item obtained by executing the single evaluation item shown in FIG. 6 and obtained. FIG. 7 shows a graph in which the points P1 to P5 are connected by a Bezier curve.

In the above embodiment, the applicant searches for a product design solution by a genetic algorithm using the labeled genotype information, which was previously filed by the applicant.
Genetic algorithms are not limited to those with labels,
Various types proposed in the past are applicable.

[0047]

The method and system for searching for a product design solution using a genetic algorithm according to the present invention execute a genetic algorithm only with an evaluation value by a single evaluation index, and provide a plurality of evaluations obtained at that time. A process of selecting and storing an evaluation value for normalization (for example, the maximum value) from the values is performed for each of the plurality of evaluation indices, and then the genetic algorithm is executed with the evaluation values based on the plurality of evaluation indices, The plurality of evaluation values obtained at that time are normalized by the evaluation value for normalization, and the value obtained by adding the normalized evaluation value of each evaluation index by the ratio of each evaluation value set in advance is calculated. The overall applicability is set. For this reason, evaluation items that would be underestimated if normalization was performed with the logical maximum value can be evaluated with the weights as input by weighting, and the search solution that the operator desires more Can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the overall configuration of a product design solution search system using a genetic algorithm according to the present invention.

FIG. 2 is a block diagram illustrating a specific configuration of an individual evaluation unit.

FIG. 3 is an explanatory diagram showing the structure of labeled genotype information and an example of crossover.

FIG. 4 is an explanatory diagram showing a basic configuration of a mass manhole.

FIG. 5 is a diagram divided into tables for easy understanding of genotype information when there are two inflow ports.

FIG. 6 is an explanatory diagram showing an XY coordinate value of each point of the best individual obtained when normalized by a logical maximum value, and a graph connecting the obtained points with a Bezier curve.

FIG. 7 shows the XY coordinate values of each point of the best individual obtained when normalized by the maximum value of each evaluation item obtained when the single evaluation item is executed, and the obtained points are Bezier curves. It is explanatory drawing which shows the graph connected.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Design basic information input part 2 Genotype information generation part 3 Phenotype conversion part 4 Individual evaluation part 5 Next generation solution candidate generation part 6 Circulation processing part 7 Intermediate progress information file screen output part 8 Weight control part

Claims (4)

[Claims] 1. A method for calculating a total fitness of a product based on a plurality of evaluation indices for each of a plurality of individuals generated as a group, selecting and duplicating the individual using the total fitness, and generating a next-generation solution candidate. A product design solution search method that uses a genetic algorithm to search for a product design solution by repeating a process such as generating individual individuals for a predetermined cycle. Each of the genetic algorithms is executed using only an evaluation value based on a single evaluation index Then, a process of selecting and storing an evaluation value for normalization from a plurality of evaluation values obtained at that time is performed for each of the plurality of evaluation indices, and then the genetic value is evaluated using the evaluation values based on the plurality of evaluation indices. Run the algorithm,
The plurality of evaluation values obtained at that time are normalized by the normalization evaluation value, and the normalized evaluation value of each evaluation index is
A method of searching for a product design solution using a genetic algorithm, wherein a value obtained by adding a predetermined ratio of each evaluation value is used as the overall applicability. 2. For each of a plurality of individuals generated as a group, a total fitness of a product is calculated based on a plurality of evaluation indices, and the individual is selected and copied using the total fitness to generate a next-generation solution candidate. A product design solution search method using a genetic algorithm that searches for a product design solution by repeating a process such as generating individual individuals for a predetermined cycle, wherein the genetic algorithm is executed using only evaluation values based on a single evaluation index. , A process of storing the maximum value of the plurality of evaluation values obtained at that time is performed for each of the plurality of evaluation indices, and then the genetic algorithm is executed with the evaluation values based on the plurality of evaluation indices, Are normalized by the maximum value, the normalized evaluation value of each evaluation index, the value obtained by adding at a predetermined ratio of each evaluation value, Product design solution search method using a genetic algorithm, characterized in that a serial overall applicability. 3. A total fitness calculating means for calculating a total fitness of a product based on a plurality of evaluation indices for each of a plurality of individuals generated as a group, and selecting and duplicating the individual using the calculated total fitness. And a genetic algorithm comprising a next-generation solution candidate generating means for generating an individual of a next-generation solution candidate, and a cyclic processing means for repeating these integrated fitness calculating means and the next-generation solution candidate generating means in a predetermined cycle. The product design solution search system used, the genetic algorithm is executed using only the evaluation value by a single evaluation index, and a normalization evaluation value is selected from a plurality of evaluation values obtained at that time. A single evaluation processing unit for performing processing such as storing in a storage unit for each of a plurality of evaluation indices; A plurality of evaluation values obtained when the rhythm is executed are normalized by the evaluation values for normalization stored in the storage means, and the evaluation values of the normalized evaluation indices are converted into the evaluation values set in advance. A product design solution search system using a genetic algorithm, wherein a value obtained by adding values according to a ratio of values is calculated as a total applicability. 4. A total fitness calculating means for calculating a total fitness of a product based on a plurality of evaluation indices for each of a plurality of individuals generated as a group, and selecting and duplicating individuals using the calculated total fitness. And a genetic algorithm comprising a next-generation solution candidate generating means for generating an individual of a next-generation solution candidate, and a cyclic processing means for repeating these integrated fitness calculating means and the next-generation solution candidate generating means in a predetermined cycle. A product design solution search system used, in which a genetic algorithm is executed using only evaluation values based on a single evaluation index, and the maximum value of a plurality of evaluation values obtained at that time is stored in a storage unit. Is provided for each of the plurality of evaluation indices, wherein the overall fitness calculation means is configured to execute a genetic algorithm with evaluation values by the plurality of evaluation indices. A value obtained by normalizing the obtained plurality of evaluation values with the maximum value stored in the storage means, and adding the normalized evaluation values of the respective evaluation indices according to a preset ratio of the respective evaluation values. A product design solution search system using a genetic algorithm characterized by calculating the total applicability.