JP2003331003A - Component selection system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently retrieve the optimal solution by combining an immune system with a chain network of questions and solutions. <P>SOLUTION: When a chain such that one or a plurality of solutions i-j (j=1, 2,...) of a question i (i=1, 2,...) generate another question (i+1)-j, one or plurality of solutions (i+1)-j-k (k=j=1, 2,...) of the question become necessary and the solutions (i+1)-j-k further generate questions is continued by defining the questions for selecting components constituting a product and defining the product to be calculated as the solutions, such a network as the chain of questions and solutions is calculated by the immune system 10 in which chiasma, mutation, degrees of antigenic affinity are introduced by regarding the problems as antigens and regarding solutions as antibodies. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention analyzes the market from past data to infer the product the customer is looking for, and when procuring the parts of the product based on this inference, The present invention relates to a component selection system by an immune system that decides whether to use components of performance / cost (selects an optimal combination of components).

[0002]

2. Description of the Related Art (1): Description of Genetic Algorithm A genetic algorithm is a technique in which the genetic mechanism of an organism is imitated and applied in an engineering manner.

In the process of evolution of living things, existing individuals (parents)
When a new individual (child) is born, crossover of the chromosomes of the individual, mutation of the gene on the chromosome, etc. occur. Individuals who do not adapt to the environment are eliminated, individuals who adapt to the environment survive, become new parents, and create new offspring.

In this way, a group of individuals adapted to the environment survive. The degree to which each individual adapts to the environment is determined by the chromosome (one-dimensional string of genes). In the genetic algorithm, candidate solutions of the optimization problem are expressed as chromosomes, which are one-dimensional strings of genes, and selection / self-reproduction / crossover / mutation ( The optimum solution is searched by repeatedly performing operations such as mutation).

FIG. 11 is an explanatory diagram of a genetic algorithm. In FIG. 11, in the population A of the chromosomes a, b, c ... d, by crossing over the chromosomes a and b and c and d, respectively, as shown in B, the chromosomes ab, ba,
It is possible to obtain cd and dc. Further, as shown in C, chromosomes a ', b', c ', d'can be obtained by mutating the genes a ₀ , b ₀ , c ₀ , d ₀ . Although not shown in FIG. 11, a group of chromosomes a, a, b, c ... Can be obtained by self-replicating a part of chromosomes, for example, chromosome a in FIG. 11 (A).

An objective function of the optimization problem corresponds to the environment, and a fitness function is defined for the chromosome so that it takes a larger value as the objective function is optimized.

The selection / self-replication is an operation of selecting an individual having a chromosome having a high fitness in a population with a higher probability to be a next-generation parent. FIG. 12 is an explanatory diagram of selection in the genetic algorithm. In FIG. 12, for example, an operation of selecting a chromosome (Chromosome) having a high fitness as a parent of the next generation.

[0008] Crossover is an operation of replacing a part of two chromosomes (parents) with each other to create a new individual (child). FIG. 13 is an explanatory diagram of crossover in the genetic algorithm. In FIG. 13, for example, the operation is to replace some of the two parental chromosomes P ₁ and P ₂ with each other to create new child chromosomes C ₁ and C ₂ .

Mutation is an operation of randomly replacing a part of genes on one chromosome. FIG. 14 is an explanatory diagram of mutations in the genetic algorithm. In FIG. 14, for example, an operation of replacing a part of the genes of one chromosome M with a random gene m.

By repeating these operations, it is possible to obtain a solution that further optimizes the chromosome having a high fitness, that is, the objective function.

(2): Explanation of data mining Data mining means mining (mining) data stored in a data warehouse (data warehouse) to obtain information / hypothesis / knowledge / issues that are treasures. It is a method / process to find such things. Since this method has been actively used mainly in the field of business, the method has been formulated at the beginning of its application to business problems.

However, processing a large amount of data by making full use of a computer and an algorithm to discover knowledge that is meaningful to humans can be applied to fields other than the business field. In the business field, data mining rarely uses itself as it is, and KDD (Kn
owledge Discovery in Databases). Here, KDD is based on large-scale data, and the information desired by the user
This is the process of extracting hypotheses / knowledge / issues and extracting information / hypotheses / knowledge / issues that the user did not expect.

KDD generally begins by selecting data when requested by a user. In general, the selected data will be inconsistent, duplicate, or not unique.
A process called data cleaning that removes such a state is executed so that data can be handled consistently. If necessary, external information should be introduced to supplement the data. It is rare for a user to perform processing including such pre-processing, and a data warehouse as shown in FIG. 15 is often in charge. Data mining is the next process. Since this part is the core processing of KDD, this part may be called KDD. Since the data itself is diverse, the data mining method is actually devised for a wide variety of problems. The following four are known as main algorithms.

A decision tree method that is simple but known to be powerful in processing large amounts of data.

A neural network known for its ability to classify and learn large amounts of data.

A genetic algorithm that is expected to exert its power in the field of practical problem solving.

Inductive logic programming, which is more expressive than decision trees and is used for knowledge acquisition.

Other methods include a query language,
There are statistical technology and visualization technology. SQL (structured query languag) which is one of the keyword searches for data
It may be possible to obtain valuable information just by using a query language such as e). There are also cases where humans can easily extract knowledge from data by using a visualization technique that represents the data in a graph or the like. In addition, a statistical method may be used as a preprocessing to obtain deeper knowledge.

FIG. 15 is an explanatory diagram of data mining. In FIG. 15, a data warehouse 24 is, for example, an existing database (DB) 21, PO of a financial institution or the like.
Data 22 such as S data generated at the site, experimental data and survey data 23 such as opinion polls are washed.
In the KDD 25, the data warehouse 24 is searched to obtain information, and knowledge, hypotheses, and problems are extracted. Here, "search" can be searched using various search engines. “Getting information” is, for example, information obtained by dividing data into categories. “Obtaining knowledge” is knowledge that, for example, when a certain CD (compact disc) song is popular, it can be sold by arranging this product in a store. “Finding a hypothesis” means finding a hypothesis such that A is B. "Find an issue"
Is to find common problems from customer complaints, for example. The user verifies the knowledge obtained from the KDD 25 and the hypothesis found, makes a solution to the found problem, and utilizes it for management and business.

[0020]

In the conventional products, it has been difficult to select the optimum parts and construct a product / product that meets the various needs of customers who are already known.

According to the present invention, by combining the immune system with a network of chains, the solution 1 for a certain problem 1 causes a further problem 2, the solution 2 for this problem 2 causes a further problem 3, etc. The purpose is to efficiently search the optimal solution for selecting.

[0022]

FIG. 2 is an explanatory diagram of the immune type network of the present invention. The present invention has the following means in order to solve the above conventional problems.

(1): The problem i (i = 1, i = 1,
2, ...) One or more solutions i-j (j = 1, 2, ...)
..) yields another problem (i + 1) -j, and one or more solutions (i + 1) -j-k (k = j = 1,
2, ...) is required, and this solution (i + 1) -j-k
If the chain continues to cause further problems, the immunity system 10 in which a network of such a chain of problems and solutions is introduced by introducing crossovers, mutations, and antigen affinity by imitating the problem as an antigen and the solution as an antibody Ask by. For this reason, it is possible to efficiently handle a problem in which many problems occur such that the solution of a certain problem causes a problem different from the original problem (because the solution is originally stored), and crossover and sudden Mutation,
The solution is effectively generated by introducing affinity,
The optimal solution can be searched efficiently.

(2): In the component selection system of the above (1), the market is analyzed as the problem 1, the product required by the market is obtained as the solution 1, and the solution 2 to the problem 2 for selecting the components constituting this product is described above. Determined by the immune system 10. Therefore, it is possible to effectively select a component from a large number of procurable components.

(3): In the component selection system of (1) or (2), when the market trend is cost-oriented, the specific component of the product represented by the antibody as the antigen affinity used when seeking a solution. A value obtained by subtracting the total sum of the weighted costs of the above from a positive constant determined so that the antigen affinity is positive is used. Therefore, it is possible to efficiently select the parts of the product in which cost is important.

(4): In the component selection system of (1) or (2), when the market trend is performance-oriented,
The sum of the standardized numerical values showing the weighted performance of the specific parts of the product represented by the antibody is used as the antigen affinity used when obtaining the solution. Therefore, it is possible to efficiently select the parts of the product that place importance on performance.

(5): In the component selection system of (1) or (2) above, when the market trend emphasizes balance, a specific component of the product represented by the antibody as the antigen affinity used when seeking a solution. And the sum of the numerical values obtained by dividing the standardized numerical value indicating the weighted performance of the above by the cost of the corresponding specific component. For this reason, it is possible to efficiently select the parts of the product for which balance is important.

[0028]

BEST MODE FOR CARRYING OUT THE INVENTION (1): Description of Basic Concept It is assumed that one clearly defined customer set “narrow market” is currently known. At this time, this "narrow market" is used to find another "narrow market", that is, a group "narrow market" of customers who need another product or merchandise. This allows companies to discover adaptable "narrow markets" and survive (survival)
The probability of doing it increases. This method is called market mining.

An optimization method inspired by the immune system is used to find the optimal combination of the components that make up the product corresponding to the "narrow market" obtained by the market mining method (the previous patent application. 2001-15557
In Section 6, we assumed a layered network that hierarchically determines the possibility of combining parts that make up a product. This patent removes this assumption and proposes a method that can be applied to more general cases).

The previous Japanese Patent Application No. 2001-313028
Then, it was possible to determine what kind of combination of parts is the product / product that the market requires by conducting a market analysis using the immune system. This method raises the question of which parts to use in the resulting product. For example, a personal computer (P
In case of C), the question arises whether to use the in-house product for the hard disk or the product of another vendor. This time, we propose an immune type system that determines what kind of performance and cost should be used by combining these parts.

FIG. 1 is an explanatory diagram (1) of a network formed by a chain of problems and solutions. In FIG. 1, a solution 1 for one problem 1 causes another problem 2 and a solution 2 for this problem 2 is needed. Solution 2 of this problem 2 yields yet another problem i, and a solution i for this problem i is needed. Thus the solution of the problem gives rise to yet another problem and a solution to this problem is needed. A network of such chains will be created. Such a network has a correspondence with the network hypothesis of the immune system.

FIG. 2 is an explanatory diagram of the immune type network. In FIG. 2, the immune system 10 for problem 1
Solution 1 due to ## EQU1 ## causes yet another problem 2. This problem 2
Solution 2 by the immune-type system 10 against A.

FIG. 3 is an explanatory diagram (2) of a network formed by a chain of problems and solutions. A problem may have multiple solutions. In FIG. 3, there are a solution 1-1, a solution 1-2, and a solution 1-3 for a certain problem 1. For example, the solution 1-1
Causes another problem 2-1 and solutions 2-1-1 and 2-1-2 are required as solutions to the problem 2-1.
For example, this solution 2-1-1 gives rise to another problem i,
A solution i to this problem i is needed. A network of such chains will be created. Such a network has a correspondence with the network hypothesis of the immune system.

(2): Explanation of Market Mining (Basic Concept of Market Mining) The "narrow market" is a narrow market for only specific customers. This market is characterized by a clearly defined market. This is the case when the market is limited to a small number of specific customers such as airline users, such as airline mileage services.

Here, "narrow" does not necessarily mean a small market. For example, some electronics stores in Akihabara
Although it targets only a specific market called PC, it cannot be said that the scale is small. In addition, the restaurant serves specific customers who prefer "Italian food" and "Chinese food", but the market size is not small.

There is Japanese Patent Application No. 2001-155576 (prior application) showing that an effective business model may be constructed by dealing with such a market. Suppose this way you have decided which products to bring to the market (eg, the products that your customers are likely to buy). Next comes the problem of selecting the parts that make up this product. The present invention addresses this issue.

(3): Explanation of the immune type system As an optimization method inspired by biology, a genetic algorithm is known, but the optimization method inspired by the immune system is also used as a countermeasure against computer viruses. Has been done. Here, a candidate solution for the optimization problem is generated at high speed by using an immune system that is inspired by the features of the immune system, "acquisition of immune tolerance" and "clone selection theory." The system has at least one of the following features.

The solution candidates satisfying the main constraint conditions of the IM-1 optimization problem are generated.

IM-2 A large number of solution candidates for the optimization problem are generated.

We consider "acquiring immunological tolerance" as the main constraint that the immune system must satisfy. Therefore, in IM-1 above,
Inspired by the acquisition of immune tolerance that does not attack itself. In the immune system that satisfies IM-1, a large number of solution candidates that satisfy the main constraint conditions are generated, and a solution candidate that is close to the optimum solution of the optimization problem is selected from among them. After that, a final optimum solution is obtained by using this solution candidate as an initial solution and an optimization method other than the immune system.

What is meant by the "clone selection theory" is that the immune system prepares in advance antibodies corresponding to antigens. Some constraints of the optimization problem correspond to "antigens". The solution candidates generated by the immune system correspond to “antibodies”. Just as the organism's immune system selects "antibodies" that match "antigens", the immune system selects solution candidates that satisfy each of the constraints of the optimization problem.

Suppose that the immune system of the previous earlier application has identified a market-fitting product. The next issue is what kind of performance and cost should be selected as the specific parts necessary for constructing this product. Again, the market is regarded as an antigen, and the parts with performance / cost that match this are regarded as antibodies. The solution is sought by producing antibodies with high affinity.

FIG. 4 is a block diagram of the immune system. In FIG. 4, the immune type system, which is a device for executing the evolutionary calculation method (device for selecting a part), has an immune mechanism I.
M and immune control means 15 are provided. The immune system is a system in which a problem is regarded as an antigen, a solution adapted to the problem is regarded as an antibody, and a solution candidate adapted to a changing problem is obtained by the immune mechanism IM. Immune mechanism IM,
The parent antibody pab is derived from the antibody group pool means 11 for storing the initial antibody group data iab set based on the past and present problem characteristic data md, and the antibody data abd (j) stored in the antibody group pool means 11. Parent selection means 12 for selection, new generation antibody production means 13 for producing a new generation antibody nab from the parent antibody pab, and the new generation antibody n
ab and an antigen affinity evaluation means 14 for evaluating the affinity of each antibody data abd (j) stored in the antibody group pool means 11 with an antigen.

The immunity mechanism IM solves the optimization problem by an immunological algorithm simulating the mechanism of immunity of an organism as described later in detail. The operation will be described here.

FIG. 5 is a flow chart of the immune type system. The immunity mechanism IM receives an execution start instruction from the external immunity control means 15 and starts execution, but continues execution until the system state meets a stop condition s given in advance. Hereinafter, the operation of the immune system will be described in step S of FIG.
1 to step S9.

S1: The antibody group pool means 11 acquires the initial antibody group data iab from the characteristic data md in question.
Further, the antigen affinity evaluation means 14 acquires the target antigen data tag from the market characteristic data md in question, and proceeds to step S2. Here, the target antigen data tag is given as data that maps the state of this problem, targeting the expected state of the market into which the product is put.

S2: The immunity control means 15 determines whether or not the system state matches a given stop condition s. In this determination, if the system state matches the stop condition s given in advance, the process proceeds to step S9, and if not, the process proceeds to step S3.

S3: The parent selection means 12 selects the parent antibody pab according to the selection algorithm and moves to step S4.

S4: The new generation antibody producing means 13 acquires the request for the solution, determines the selection operator op, and moves to step S5.

S5: The new-generation antibody production means 13 produces a w-piece of new-generation antibody nab by causing the selected operator op to act on the set parent antibody pab, and then proceeds to step S6.

S6: The immune control means 15 performs step S5.
Then, it is judged whether or not the new generation antibody nab could be produced. In this determination, if the new-generation antibody nab can be produced, the process proceeds to step S7, and if not, step S3.
Then, the parent antibody pab and the selection operator op are selected again to make the production successful.

S7: The antigen affinity evaluation means 14 uses the target antigen data tag and the antibody data nab of the new generation antibody nab to determine the antigen affinity f (ta) of the new generation antibody nab.
g, nab) is calculated. In addition, the antigen affinity f (t) of other existing antibody data abd (j) in the antibody group pool means 11
ag, abd (j)) is obtained, and the process proceeds to step S8. Here, j = 1, 2, ..., J, and a total of J pieces of antibody data abd (j) are stored in the antibody group pool means 11. Generally, J can change every time the loop of steps S2 to S8 is repeated. When this step S7 is first executed, the antibody data abd (j) is equal to that given as the initial antibody group data iab.

S8: The antibody group pool means 11 is the above J +
w antibody data abd (j), nab antigen affinity f
In accordance with the antibody group pool storage algorithm, the antibody data abd (j) to be stored in the antibody group pool 11 is selected and stored, other antibodies are deleted, and the process returns to step S2. As a result, J'pieces of antibody data abd (j) are pooled. This antibody group pool storage algorithm naturally selects antibody data abd (j) having a higher antigen affinity f (tag, abd (j)).

Thus, the antibody group pool means 11 has J '.
With the updated antibody data abd (j), the production of the first-generation new-generation antibody ends, and the process returns to step S2 to start the second generation. In this way, the content of the antibody group pool gradually changes to the antigen affinity f (tag, abd (j)).
High antibody data abd (j) group. By the above operation, the immune mechanism IM can produce and pool the antibody having a higher affinity for the target antigen data tag.

S9: If the stopping condition s is met in step S2, the immune control means 5 selects the highest antigen affinity f (tag, abd) from the pool antibodies of the antibody group pool means 11.
An antibody with high (j)) is selected as the optimal antibody mfab,
The evaluation value ev (that is, the antigen affinity f (tag, mf
ab)) is obtained and the operation ends.

In this way, the component selection system of the present embodiment can present the product having the optimal component configuration for the target market together with the fitness evaluation value for the market.

(4): Description of component selection method (description of target product) The product handled here is a product in which major components are connected by a bus, such as a PC. One of the characteristics of these products is that components (parts of a hard disk such as a hard disk of a PC) provided by various vendors are collected to form a final product or system. For example, "B (Business) to B" on the e-marketplace on the Internet.
When procuring the components of such a product by “(Business)”, the number of types of parts required is very large and it becomes difficult to find the optimal solution by calculation. You need a system.

(Description of Product Model) It is assumed that a product is made by combining several components. The type of each component is MC1, MC2, MC3, ...
・, MCN. Here, for example, in the case of a PC, M
C1 is a CPU (Central Processing Unit), MC2
Is a hard disk. Furthermore, concrete components corresponding to the types of each component are expressed as follows.

[0059]

[Equation 1]

Specifically, VC1 (i) is a CP of Intel Corp.
CPU of U and AMD (Advanced Micro Devices, Inc.)
Is. The VC2 (i) is, for example, a hard disk of Fujitsu Limited or a hard disk of Hitachi Ltd. However,
No order relationship is assumed between the component types.

(Explanation of Market Data) From the past experience, the following trends mainly exist in products.

Performance Priority, Cost Emphasis, Balance Emphasis (For General Customers) The trend of performance emphasis is that performance is prioritized over cost in the case of a computer like a large machine or a super computer. The emphasis on cost is a trend that emphasizes cost reduction rather than performance, and corresponds to personal computers for low-end users. Balance-oriented is a trend for general customers who place importance on the balance between cost and performance.

The immune system determines which trend from the input data and sets an affinity with an appropriate selection operator. As a result, the immune system can provide a product using appropriate parts required by the user.

(5) Description of Specific Operation of Immune Type System (Explanation of Code Method for Antibody Data) The code method for antibody data is given by using the parts constituting the product. The following two main coding methods are convenient.

Chromosome expression in which parts are arranged in a list.

Chromosome representation by a graph in which parts are nodes and linking relationships between parts are links.

In this case, it may not be possible to combine arbitrary parts. Using the terminology in the product model description above, the concrete component VCi (j) of the product component type MCi cannot be combined with any other concrete component. However, it is assumed that the immune system can already refer to the data that can determine what specific components (ie, parts) are combined. Since the layered network is not assumed unlike the previous application, it can be applied to a wider range.

FIG. 6 is an illustration of genes, and FIG. 6 (a) is an illustration of genes that can be combined with possible genes. In FIG. 6A, for example, the component VC
1 is VC1 (1), VC1 (2), VC1 (3), VC1
There are (4) and VC1 (5), and the component VC2 has
VC2 (1), VC2 (2), VC2 (3), VC2 (4), V
There is C2 (5). In addition, genes that cannot be combined include "VC1 (1), VC2 (1), VC3 (4), VC4
(2), VC5 (2) "and" VC1 (2), VC2 (1), VC
3 (1), VC4 (3), VC5 (1) ".

FIG. 6 (b) is an explanation of an example of a chromosome. In FIG. 6B, as an example of the chromosome, "VC1 (2),
VC2 (2), VC3 (4), VC4 (3), VC5 (1) "are shown.

(Explanation of Initial Population of Antibodies) As the initial population of antibody data, random selection is made within the range of parts in which specific components assigned to the types of product components can be combined. Such a combination
When the size is smaller than the size of the designated initial population, the randomly selected antibody data is duplicated to prepare the initial population of the designated size.

(Explanation of crossover) As the crossover which acts on the above-mentioned antibody data, the following methods can be considered.

A method of pasting a part of the parent 1 antibody data to the corresponding part of the parent 2 antibody data. In this case, a child is normally created when the combination of the parts represented by the antibody data of the resulting child is possible. Otherwise, if it is not possible to combine the parts represented by the obtained child antibody data, the child antibody data is a copy of the parent antibody data.

An improved version of the method of pasting a part of the parent 1 antibody data to the corresponding part of the parent 2 antibody data. In this case, a child is normally created when the combination of the parts represented by the antibody data of the resulting child is possible. Otherwise, if it is not possible to combine the parts represented by the child antibody data obtained, the specified number of times is specified until the parts represented by the child antibody data can be combined using the same parent. Carry out a crossover. If the child antibody data does not represent a possible combination of parts even after performing the crossover a specified number of times, the child antibody data is a copy of the parent antibody data.

FIG. 7 is an explanatory diagram of the crossover, and FIG. 7A is an explanatory diagram of a successful pasting crossover. In FIG. 7 (a), part of the antibody data VC3 (4), VC4 (3) of the parent P1
Is attached to the corresponding portion of the parent P2 antibody data. As a result, the attachment crossover is performed and the children C1 and C2 are normally created.

FIG. 7B is an illustration of a failure example of pasting crossover. In FIG. 7B, part of the antibody data VC3 (4), VC4 (3) of the parent P1 is attached to the corresponding part of the antibody data of the parent P2. This pasting crossover results in antibody data C1 made up of genes that cannot be combined.
Therefore, the antibody data for child C1 in this case is the parent P1.
A copy of the antibody data for "VC1 (1), VC2 (1), VC
3 (4), VC4 (3), VC5 (2) ".

(Explanation of Mutation) As the mutation acting on the above-mentioned antibody data, the following method can be considered.

R-type mutations: mutations that randomly change genes (specific components assigned to product component types) at randomly selected loci in antibody data. If a combination of parts represented by the obtained antibody data is possible, this is put in the antibody data population. Otherwise, it will not be included in the antibody data population.

FIG. 8 is an explanatory diagram of mutations.
(A) is an explanation of a successful mutation example. In FIG. 8 (a), a part of the antibody data VC3 (4) of the gene M is randomly changed (VC3 (2)) and mutated to the gene M '. This combination is a successful example and is included in the population of antibody data.

FIG. 8 (b) is an illustration of a failure example of mutation. In FIG. 8B, a part of the antibody data VC3 (1) of the gene M is randomly changed (VC3 (4)) and mutated to the gene M '. Since this combination is an impossible gene (failure example), it cannot be included in the antibody data population.

IR type mutations: mutations that randomly change genes (specific components assigned to product component types) at randomly selected loci in antibody data. When the combination of parts represented by the obtained antibody data is possible, this is put in the population of antibody data. Otherwise, the same antibody data is used to perform the mutation a pre-specified number of times until a combination of parts represented by the mutated antibody data is possible. If the antibody data does not represent a possible combination of parts even after performing the mutation a specified number of times, the antibody data will not be included in the antibody data population.

SR type mutations: Genes that can be exchanged for genes (specific components assigned to product component types) at randomly selected gene loci in antibody data (combination of parts capable of antibody data) A mutation that replaces the selected gene with a gene on this list.

(Description of Antigen Affinity) The antigen affinity defined as follows can be used. Where cost (i)
Represents the cost of the specific component assigned to the i-th component of the product represented by the antibody data. Similarly, spec (i) is also a standardized numerical value expressing the performance of the specific component assigned to the i-th component of the product represented by the antibody data. ui, vi, and wi are positive numbers representing appropriate weights. C is an appropriately selected positive constant for making the antigen affinity positive.

[0083]

[Equation 2]

The immune type system determines which antigen affinity to use based on the market data entered. For example, taking a computer as an example, if a super computer that emphasizes performance over cost is desired, f
Select _spec as the antigen affinity. For a market that demands low-cost computers, f _cost may be selected as the antigen affinity. In a general market that requires a proper balance between performance and price, it is necessary to select f _consumer as the antigen affinity.

The component selection method described in the above embodiment is
It is a way to implement which parts should be used in a product in a clear and effective manner. The application of the immune system in the component selection problem is a method that contributes to the increase in affinity with the market and increases the speed of obtaining a solution adapted to the market. By the method proposed here, it is possible to more efficiently propose a product required by the market in the component selection problem by the search method of the immune system.

(6) Description by Specific Example: Application to Computer Parts Selection Problem Consider that the immune system is applied to a computer (product) part selection problem. Here, the target computer parts are a CPU, a memory (Mem), and an HD (hard disk). More precisely, the types of product components are CPU, memory, and HD. CPU1, CPU2, CPU3, Mem1, Me are specific components corresponding to these types of components.
Suppose there are m2, Mem3, HD1, HD2, HD3.

FIG. 9 is an explanatory diagram of a computer part selection problem, and FIG. 9A is an explanation of components, costs and specifications. In FIG. 9A, each specific component (CPU1, CPU2, CPU3, Mem)
It is assumed that the cost and specifications (performance) of 1, 1, Mem2, Mem3, HD1, HD2, HD3) are as shown in the figure.

FIG. 9B is an explanation of the application to the component selection problem. In FIG. 9B, the computer with the cost and specifications of the components (possible genes) of FIG. 9A is sold to general consumers. This is input to the immune system as antigen data. Therefore, the immune system uses f as the antigen affinity in order to assemble a product with appropriate performance and price balance from past data.
Select _consumer . A specific set of components that cannot be combined is "CPU2, Mem3, H
D2 "," CPU2, Mem3, HD3 "(assumed). The antigen affinity is f _consumer
It was used. Where α = 10, β = 1, C = 30, ui
= 1 and wi = 1. An example of chromosomes that can be combined is "CPU2, Mem3, HD1".

The result of this simulation is 100
Antigen data with an antibody affinity of 100 in the generation "CPU3,
Mem2, HD1 "was presented as a solution candidate.

Description of Application to Medicine The immunotype system presented here can also be used when a medicine is used in combination. In this case, "product" corresponds to "drug", "part" corresponds to "drug combination", and "cost"
Corresponds to the strength of "side effects" and "performance" is the "effect of the drug"
It corresponds to. If the “side effects” and “effects of drugs” are quantified and standardized, the optimal drug combination can be obtained depending on the immune system. Of course, the combination of drugs that are harmful and dangerous to the human body when combined is given as input data to the immune system as data. Drugs that are tailored to the needs of the individual (if side effects are good, side effects may be strong (effect emphasis), no side effects are good (side effects emphasis), and moderate effects are expected (balance importance)) Can be treated.

FIG. 10 is an explanatory diagram of application to medicine. Figure 1
In 0, as a part when using drugs in combination,
Use as cold medicine, cough medicine, and antibiotics. As specific examples of possible genes corresponding to the types of these combinations,
Cold medicine 1, cold medicine 2, cold medicine 3, cough medicine 1, cough medicine 2,
Suppose you have cough medicine 3, antibiotic 1, antibiotic 2, antibiotic 3. Uncombinable genes "cold medicine 2,
Suppose cough medicine 3 and antibiotic 2 ". As an example of chromosomes that can be combined, "cold medicine 2, cough medicine 3,
Antibiotics 1 "etc.

In this case, if the side effect intensity of each drug and the effect of each drug are quantified, the optimal combination of drugs can be obtained depending on the immune system. For example, a combination of these drugs is input to the immune type system as antigen data. The immune system, for example, selects f _consumer as an antigen affinity to prescribe a drug having a well-balanced moderate side effect and drug effect from past data, presents it as a solution candidate by performing simulation. You can

In the application to the drug, "side effect" and "effect of drug" were used, but "cost" and "effect of drug"
Alternatively, “cost”, “side effect”, “effect of drug” and the like can be used.

As described in the above embodiments, by combining the network hypothesis in the immune system and the immune type system, it is possible to efficiently deal with the problem that the solution of a certain problem causes a problem different from the original problem. You can do it (because the solution is stored in the beginning, it is efficient for those with many problems). Further, by introducing crossover, mutation, and affinity suitable for the component selection problem into the immune system, the solution can be effectively generated and the optimum solution can be searched efficiently.

(7): Description of program installation Antibody group pool means 11, parent selection means 12, new generation antibody production means 13, antigen affinity evaluation means 14, immune control means 1
5. The immunity mechanism IM and the like can be configured by a program, are executed by the main control unit (CPU), and are stored in the main memory. This program is processed by a general computer. This computer is configured with hardware such as a main control unit, a main memory, a file device, an output device such as a display device and a printing device, and an input device.

The program of the present invention is installed in this computer. This installation is performed on a portable recording medium such as a floppy disk or magneto-optical disk.
These programs are stored and installed in a file device provided in the computer via a drive device for accessing a recording medium included in the computer or via a network such as a LAN. It As a result, it is possible to easily provide a component selection system, device, or method capable of efficiently handling a problem in which a large number of problems occur such that the solution of a certain problem causes a problem different from the original problem.

[Additional Notes] (Additional Note 1) The problem i (i = 1, 2, ...
.) One or more solutions i-j (j = 1, 2, ...) Generate another problem (i + 1) -j, and one or more solutions (i + 1) -j-k (of this problem k = j = 1, 2, ...) is required, and if the chain of solutions (i + 1) -j-k causes further problems, the problem is regarded as an antigen and the solution as an antibody. A part selection method or part selection system, characterized in that a network of such problem-solution chains is obtained by an immune type system in which crossover, mutation, and antigen affinity are introduced.

(Supplementary Note 2) It is characterized in that the market is analyzed as Problem 1, the product required by the market is obtained as Solution 1, and Solution 2 for Problem 2 for selecting the components constituting this product is obtained by the immune system. The component selection method or component selection system according to attachment 1.

(Supplementary Note 3) Input data characterized by using a trend of a narrow market in the market as input data in the immune system of Supplementary Note 2.

(Supplementary Note 4) The immune type system of Supplementary Note 2 is characterized by using a method similar to “antibody” that changes the selection method of product parts according to the trend of the narrow market which is the input data of Supplementary note 3. Immune type system.

(Supplementary Note 5) An immune type system, characterized in that, in the component selection method or component selection system according to Supplementary Note 1 or 2, this and an evolutionary calculation method execution device are used in combination.

(Supplementary Note 6) As a coding method for an antibody when a solution is obtained by a method similar to the “antibody” in Supplementary Note 5, the parts that make up the product are used as nodes, and the parts that can be combined are connected by links. An immune system characterized by using an evolutionary calculation method using a generated graph.

(Supplementary Note 7) A feature is that an evolutionary calculation method using a list whose components are components of a product is used as an antibody coding method when a solution is obtained by a method similar to the “antibody” in Supplementary Note 5. Immune type system.

(Supplementary Note 8) In a method similar to “Antibody” in Supplementary Note 5, when the antibody coding method in Supplementary Note 7 is used,
A system characterized by using a table for determining whether or not a previously created combination is possible for determining whether or not coded parts can be combined with each other.

(Supplementary Note 9) A specific component (part) that can combine the initial population having a specified size in a method similar to the “antibody” in Supplementary Note 5
An immune type system characterized by being configured by randomly selecting within the range.

(Supplementary Note 10) In the method of constructing the initial population of Supplementary Note 9, when the initial population having a specified size cannot be satisfied by randomly selecting it within the range of concrete components (parts) that can be combined. Is an immune type system characterized by constructing an initial population using a copy of antibody data already created.

(Supplementary note 11) A method similar to the method of "Antibody" in Supplementary note 5, characterized by using sticking crossover as an operator used when obtaining a solution by evolutionary calculation.

(Supplementary Note 12) In the pasting crossover of Supplementary Note 11, if a specific combination of components (parts) represented by the antibody of the child is impossible, no child is produced.

(Supplementary note 13) In the pasting crossover of Supplementary note 11, the pasting crossover is repeatedly applied until a specific combination of components (parts) represented by the antibody of the child becomes possible.

(Supplementary Note 14) A method similar to the method of “Antibody” in Supplementary Note 5, characterized by using an R-type mutation as an operator used when obtaining a solution by evolutionary calculation.

(Supplementary Note 15) A method characterized by using an IR-type mutation as an operator used in obtaining a solution by evolutionary calculation in a method similar to the “antibody” in Supplementary Note 5.

(Supplementary Note 16) A method similar to the method of “Antibody” in Supplementary Note 5, characterized by using an SR type mutation as an operator used when obtaining a solution by evolutionary calculation.

(Supplementary Note 17) An immunotype characterized in that, in the method similar to “Antibody” in Supplementary Note 5, the input data showing the market trend in Supplementary Note 3 is regarded as “Antigen” and the affinity of the antigen to be used is selected. system.

(Supplementary Note 18) An immune type system characterized in that, as the market trend of Supplementary Note 5, performance priority, cost emphasis, and balance emphasis are used as input data of the market trend.

(Supplementary Note 19) A specific component (part) of a product represented by “antibody” as an antigen affinity used when obtaining a solution when the market trend is cost-oriented.
6. The component selection system according to appendix 5, wherein a value obtained by subtracting the total sum of the weighted costs of the above from a positive constant determined so that the antigen affinity is positive is used.

(Supplementary Note 20) When the market trend is performance-oriented, the standardized weighted performance of the specific component (part) of the product represented by the antibody is used as the antigen affinity used when seeking a solution. 6. The component selection system according to appendix 5, wherein the sum of numerical values is used.

(Supplementary Note 21) When the market trend emphasizes balance, a standardized value indicating the weighted performance of the specific component (part) of the product represented by the antibody as the antigen affinity used when seeking a solution. 6. The component selection system according to appendix 5, characterized by using a sum of numerical values obtained by dividing a numerical value by the cost of a corresponding specific component.

(Supplementary note 22) The problem i (i = i =
1, 2, ..., One or more solutions i-j (j = 1,
, ...) yields another problem (i + 1) -j, and one or more solutions (i + 1) -j-k (k = j = 1,
2, ...) is required, and this solution (i + 1) -j-k
If the chain continues to cause further problems, the problem is regarded as an antigen and the solution is regarded as an antibody, and the network of such a chain of problems and solutions is crossed, mutated, or by an immune type system that introduces antigen affinity. A computer-readable recording medium having a program or a program for causing a computer to execute a desired function.

[0119]

As described above, the present invention has the following effects.

(1): The problem of selecting the components that make up the product and the product to be sought are taken as solutions, and the network of problem-solution chains is defined by an immune system that introduces crossover, mutation, and antigen affinity. Therefore, it is possible to efficiently deal with the problem that many problems occur such that the solution of a certain problem causes a different problem from the original problem, and the solution by introducing crossover, mutation, and affinity. Can be effectively generated and the optimum solution can be efficiently searched.

(2): The market is analyzed as Problem 1, the product required by the market is found as Solution 1, and Solution 2 for Problem 2 that selects the components that make up this product is obtained by the immune system, so that procurement is possible. A component can be effectively selected from a large number of components.

(3): When the market trend emphasizes cost, the sum of weighted costs of specific parts of the product represented by the antibody is used as the antigen affinity used when obtaining a solution, and the antigen affinity is positive. Since a value obtained by subtracting from a positive constant determined so as to be used, it is possible to efficiently select the parts of the product in which cost is important.

(4): The sum of the standardized numerical values showing the weighted performance of the specific parts of the product represented by the antibody as the antigen affinity used when seeking a solution when the market trend emphasizes performance. Therefore, it is possible to efficiently select the parts of the product whose performance is important.

(5): Corresponds to a standardized numerical value indicating the weighted performance of the specific component of the product represented by the antibody as the antigen affinity used when seeking a solution when the market trend emphasizes balance. Since the sum of the numerical values divided by the cost of the specific parts to be used is used, it is possible to efficiently select the parts of the product for which importance is attached to the balance.

[Brief description of drawings]

FIG. 1 is an explanatory diagram (1) of a network formed by a chain of problems and solutions in the embodiment.

FIG. 2 is an explanatory diagram of an immune network according to the embodiment.

FIG. 3 is an explanatory diagram (2) of a network formed by a chain of problems and solutions in the embodiment.

FIG. 4 is a configuration diagram of an immune system according to the embodiment.

FIG. 5 is a flowchart of an immune system according to an embodiment.

FIG. 6 is an explanatory diagram of genes in the embodiment.

FIG. 7 is an explanatory diagram of crossover in the embodiment.

FIG. 8 is an explanatory diagram of mutation in the embodiment.

FIG. 9 is an explanatory diagram of a computer component selection problem in the embodiment.

FIG. 10 is an explanatory diagram of application to medicine in the embodiment.

FIG. 11 is an explanatory diagram of a conventional genetic algorithm.

FIG. 12 is an explanatory diagram of selection in a conventional genetic algorithm.

FIG. 13 is an explanatory diagram of crossover in a conventional genetic algorithm.

FIG. 14 is an explanatory diagram of mutations in a conventional genetic algorithm.

FIG. 15 is an explanatory diagram of conventional data mining.

[Explanation of symbols]

10 Immune type system 11 Antibody Group Pool Means 12 Parent selection means 13 New Generation Antibody Production Means 14 Antigen affinity evaluation means 15 Immune control means IM immune system

Claims (5)

[Claims] 1. A solution for selecting a component constituting a product and a product to be obtained as a solution, and one or more solutions i-j of a problem i (i = 1, 2, ...).
(J = 1, 2, ...) Gives another problem (i + 1) -j, and one or more solutions (i + 1) -jk of this other problem
(K = j = 1, 2, ...) Is required, and this solution (i
If the chain continues such that +1) -j-k causes more problems, the problem is regarded as an antigen, and the solution is regarded as an antibody. A component selection system characterized by the immune system that has been introduced. 2. A method for analyzing a market as a problem 1, finding a product required by the market as a solution 1, and finding a solution 2 for the problem 2 for selecting parts constituting the product by the immune system. 1. The component selection system described in 1. 3. When the market trend is cost-oriented, the sum of weighted costs of specific parts of the product represented by the antibody is used as the antigen affinity used when obtaining a solution so that the antigen affinity becomes positive. 3. The component selection system according to claim 1, wherein a value obtained by subtracting from the positive constant defined in is used. 4. When the market trend is performance-oriented, the sum of standardized numerical values showing the weighted performance of specific parts of the product represented by the antibody is used as the antigen affinity used when obtaining a solution. The component selection system according to claim 1 or 2, wherein. 5. When a market trend emphasizes balance, a standardized numerical value indicating a weighted performance of a specific component of a product represented by an antibody is used as an antigen affinity used when obtaining a solution. 3. The component selection system according to claim 1 or 2, wherein a total sum of numerical values divided by the cost of specific components is used.