The Research Of The Fuzzy Classification Algorithm And Its Application In The Mining Of Forging Model Design Criteria

In data classification, facing a large number of original data which thecorresponding mathematical models are extracted from, on this basis, hidden rules arefound. These problems are difficulty to be described by traditional deterministicmodels and to be solved based on accurate mathematical methods. Yet it is moresuitable that they be described as fuzzy optimization problems.Fuzzy optimization is essentially a multi-objective optimization problem, whilegenetic algorithm (GA) is an effective method which is used to solve multi-objectiveproblems. The GA in fuzzy optimization, especially the one based on dynamic models,has its special essence, therefore it is necessary that it be studied more deeply.Two typical intelligent algorithms, fuzzy optimization and GA, are studied inthis dissertation. Applying fuzzy optimization algorithms based on GA to dataclassification based on regression equation, flash size design criteria are obtained,while applying fuzzy optimization algorithms based on GA to data classificationbased on fuzzy neural networks, flash metal consumption design criteria are obtained.Thereby the effectiveness of fuzzy optimization algorithms based on GA is verified.The main content of this dissertation is as follows.(1) Several main classification algorithms are elaborated and the researchingprogress of classification algorithms is researched. Finally the rule expressions offuzzy classification algorithms are probed, including such rule expressions as onesbased on regression equation, neural networks and nature-inspired computation.(2) The related concepts of fuzzy optimization are probed and the traditionalfuzzy optimization problems are discussed. The fuzzy optimization problems onclassification algorithms are chiefly focused on, which include fuzzy optimizationmodels based on regression equation and neutral networks，and their realizationmethods and steps are discussed. Aimed at the above two fuzzy optimization problems,fuzzy optimization models are constructed by the form of fuzzy logic.(3) The basic idea of GA is elaborated and the simple GA is described. Thensome improvement strategies are proposed, which mainly include the methods ofadaptive genetic parameter adjustment and the methods of the mutation along theweighted gradient direction.(4) The application of fuzzy classification algorithms based on regression equation to the design criterion mining of flash sizes is carried out. The miningproblem of flash size design criteria is solved by fuzzy classification algorithm basedon regression equation. The fuzzy classification algorithm is compared and analyzedas follows: checking the algorithm stability and comparing the algorithm with themining algorithm realized by least-square method, and the results show that thisalgorithm is faster and has a higher accuracy and better stability.The application of fuzzy classification algorithm based on fuzzy optimization BPneural networks to the design criterion mining of flash metal consumption is carriedout. The mining problem of flash metal consumption design criteria is solved by thefuzzy classification algorithm based on neutral networks. The fuzzy optimizationalgorithm is compared and analyzed as follows: checking the algorithm stability andcomparing the algorithm with the mining algorithm realized by gradient decentmethod. The results show that this algorithm is faster and has higher accuracy andbetter stability.At last, the content of this dissertation is summarized and further feasibleresearch problems are discussed.