Study For GA-Based Fuzzy System

Fuzzy system is a kind of knowledge-based or rule-based system, its heart is the rule base consists of "if-then" rules. Fuzzy system is broadly used in the fields of control, signal process, communication, integrated circuit, Expert System, medical diagnosis and behavior science. This paper proposed the study scheme for GA based fuzzy system from the view of methodology: including fuzzy classification, fuzzy modeling, multi-objective optimization, Agent-based evolution and rules interaction handling. The main contributions of this work are summarized as follows:Firstly, a new method for labeled fuzzy classifier is created by integrating GA to the variable input spread inference-training (VISIT) algorithm. An expert system acts as a fitness function to accomplish the accuracy and interpretability simultaneously. The new fuzzy classification system has some advantages as follows: (1) good interpretability, (2) efficient feature compression,(3) comparative accuracy to the traditional methods.Secondly, a new user-friendly Sugeno fuzzy modeling is proposed. The Sugeno system can use the linear control analytical method to approximate the nonlinear system, where the algorithm determines the all parameters automatically, instead of being given by users in advance. Furthermore, the new algorithm can be extended to the Mamdani fuzzy system.Thirdly, a recursive incremental multiple objective genetic algorithm (RIMOGA) is proposed. The whole evolution process is divided into the same recursive phase as the numbers of objectives and one more objective is added in each phase. In each phase, a new objective is evolved on an independent population. Then the better individuals selected from the single-objective population together with the multi-objective population evolved in the last phase generate the initial population for the added objective set to evolve. The RIMOGA can find more and better solutions than other typical multi-objective genetic algorithm.Fourthly, an Agent-based method for extracting rule-based knowledge is proposed. Each behavior Agent (BA) autonomouslydetermines the number and distribution of the fuzzy sets and improves the interpretability of fuzzy system by using hierarchical chromosome. structure and regulation strategy. Furthermore, the BA can exchange their fuzzy sets information cooperatively in order to maintain the solutions diversity and obtain the global optimal solution. New method can achieve a good tradeoff between the interpretability and accuracy of the fuzzy system.Finally, the traditional uncertain inferences have dealt with the credibility determination of many rules with the same consequents. But the credibility determination of fuzzy rules with the same consequents has not been seen when reasoning with fuzzy system. In order to handling and modeling to the fuzzy rules interactions, this paper suggests using a non-additive set function to replace the weights of rules, to draw the reasoning conclusion and to determine the credibility based on an integral with respect to the non-additive set function. The proposed technique can help to find new knowledge and maintain the knowledge base.