In ink composition analysis applications, it is often the case that an ink does not meet the requirements of use. If the ink samples that do not meet the requirements of the application can be made into qualified ink by adding additives, blending, etc., the limited ink resources can be fully utilized. This is a huge economic benefit of the ink application business. .
Due to the complexity of the ink composition, it is difficult to predict the direction of change in the additive quality and ink quality after blending. Therefore, there is a great deal of blindness in specific ink and specific index processing methods. In this paper, an analogy inference system based on data mining technology, the Ink Analysis Computer Management System (OACMS), is proposed to guide the correction of the ink index.
First, the basic idea and solution model
The heuristic search technology based on analogy reasoning is to obtain the solution process of the past problem similar to the new problem through analogy, as the heuristic information to guide the solution of the new problem, which can reduce the blindness of the search, narrow the search scope, and reduce Solve the difficulty of the problem.
In general, heuristic search based on analogical reasoning directly uses the case library as the search space. The result of this processing will lead to two problems:
1. The performance of the system will continue to decrease during the operation of the system. With the increase of the number of solutions, the continuous accumulation of solution cases, the search space will continue to expand, although this can make the system have ample opportunity to obtain a solution to past problems similar to the solution of the problem, but it will also lead to heuristic search information. Saturation reduces system efficiency and performance.
2. The performance of the system will depend to a large extent on the compromise between search costs and case inspiration. The contents of the casebase record are incomplete, which can reduce the requirements for the degree of similarity of the case. The matching process may be easier, but the enlightening effect of the past case guidance solution will be weakened; the content of the case library record is too detailed, the degree of similarity of the case requirements It is relatively high, the matching process is difficult, the search cost is increased, and the system may also rely on weak methods to solve the problem.
In real life, human beings do not store their past experience in the mind, but use a common data structure to store it. That is, there is a process of mining useful information from past case data. This is an effective way to solve problems, accumulate experience, and enhance problem solving skills. The solution method considered in this paper is to use data mining techniques to abstractly abstract the association rules with hierarchical structure from a large number of case database data. The system solves the problem by searching for matching rules from the rule base. Since data mining produces hierarchical information for solution, the system can reduce the search cost while minimizing the impact of information loss and maintain the system's search quality, thus ensuring that the system's solution efficiency increases with the increase of running time. Will continue to improve. .
The empirical knowledge accumulated in the past can be divided into two categories: shallow case libraries and rule bases that are generated through data mining in case databases. At the beginning of system operation, there is a lack of solving cases in the case library. There are a few simple past experience rules in the rule base, so the system must use weak methods to search for solutions. After the system has accumulated certain case experiences, the solution to the new problem is first searched in the rule base. If the rules are applied successfully,
: The casebase records the process of solving the past cases; the rule bank stores the processing rules resulting from the abstract processing of the data in the casebase; the control strategy specifies how to determine the similarity of the events; and the analogy inference mechanism stipulates how to conduct the business at a specific degree of similarity. Process, and complete the addition, modification, and deletion of the case base, rule base, and control strategy.
The solution is successful. Otherwise, the search case base, if there is a similar problem solving, is used as a heuristic information to guide the solution of the new problem. If there is still no similar solution to the past problems, the weak method is used to solve and the new problem solving process is added to the case library. The more experience accumulated in the past, the more effective the application of the rule base; the higher the degree of similarity between the new problem and the past problem, the fewer search times in the solution process.
Second, system implementation
In the implementation process, we need to solve 1 how to generate solution cases; 2 how to define and judge the similarity of two cases; 3 how to abstract general rules from a large number of cases; 4 how to maintain the rule base and case base; 5 How to use heuristic information to guide new issues and solve five issues.
When generating a solution case, during the process of recording and saving the problem solving, search for future solutions to search for feature information with heuristics. The feature information defining the search process includes the sequence of operations on the search path and the state value of the solution problem at the time of selecting each operation. Due to the introduction of the rule base, feature information can be described as exhaustively as possible. In the OACMS, the characteristic information of the search process includes the description of the processing operation performed on the ink, the influencing factors of the processing operation, and the change of the index of the ink after each treatment.
Due to the need to find the case guidance solution with the highest degree of similarity to the new problem in the case base, it is necessary to determine the degree of similarity between the two cases. OACMS determines the degree of similarity of two cases by assigning weights to each feature information. This search cost to find the best similarity matching process will be higher when the case base is inflated. Therefore, the system introduces a rule base to reduce the cost. Search fee.
If the occurrence of a certain type of case is more frequent, using a case library to search for a specific case to deal with, will increase the search cost, use a generalized example of a certain type, solve the problem according to the generalized rules, can be reasonably reduced Search costs, without compromising guidance on solving new problems. In the OACMS, the author uses the method of dividing the attribute value of the feature information into certain segments and then mapping it into Boolean, and obtains generalized rules of certain degree of support and certain credibility. Confidence constraint generalization rules represent the frequency of occurrence of a certain type of cases, and it is only necessary to provide generalization rules for cases that reach a certain degree of frequentity; the degree of support constraints generalizes the accuracy of the rules, only to a certain degree of accuracy. The rules can guide the problem solving.
The maintenance operations on the case base and rule base include additions, modifications, and deletions. There are two cases of search for the case base, search success and search failure. If the search fails, the new problem is solved by using a weak method to fill in the feature information directly in the process of solving the problem and complete the increase of the case library. The search success will produce two kinds of situations, one is that the feature information is exactly the same, and the solution path is the same; in another case, the analogy inference may fail because the feature information cannot cover all the factors that affect the problem solving. In the second case, the casebase needs to be modified and the solution to the new problem added to the casebase as a new case. In OACMS, rewards and penalties are used to determine which solution path to use to guide future solutions to similar problems. Similarly, a reward and penalty function and a certain threshold are used to determine if a case needs to be deleted to reduce future invalid search costs. The reason for the maintenance of the rule base mainly comes from changes in the case base, changes in support, and credibility. Because the update of the rule base requires scanning of the entire case base, it is not appropriate to use the change of the case base to trigger the update of the rule base. For this purpose, the OACMS uses the work mode of regularly updating the rule base.
The main problem of using heuristic information to guide the solution of new problems is to solve the problem of mismatch in the solution process. If the search process cannot obtain heuristic information, then the weak method is used to solve and the case library is added; if the search process obtains exactly the same heuristic information but the solution fails, the rewards and penalties are used to extend the case as a counterexample to the original solution case; if the search The process obtains partially similar heuristic information (it is not appropriate to apply rules to partially similar situations), heuristic information and weak methods are used to solve, and new cases are added to the case base.
(author: Liu Zhipeng and Ke Zhi, Xu Shaohua, Xi'an University of Technology)