The Fuzzy Modeling Method Based Online-Clustering And The Least Squares Support Vecror Machines

As the combination of fuzzy theory and automatic control technologies, fuzzy control can overcome the disadvantage of traditional control methods which relied on target's accurate model excessively, it can reflect human experiences and knowledge. Due to its simple structure and universal approximation ability, it has been developed to be a effective method of dealing with the nonlinear and uncertainty complex systems. As the most important issue of establishing a fuzzy control system, fuzzy modeling attracts people to research it untiringly. The main task of fuzzy modeling is extracting fuzzy rules and identifying the parameters of membership function effectively. Extraction of fuzzy rules is the key issue of fuzzy modeling, membership function parameters is the important indicators. Unfortunately, so far there is no comprehensive theory for them, this limits the more extensive and more effective application of fuzzy control.This paper focuses on the methods of fuzzy modeling.Support vector machine is a learning algorithm, owned good generalization ability to overcome the shortcomings of traditional learning algorithms. Clustering method has proven to be an effective tool for fuzzy space partition. In this paper, the combinations of two methods for structure identification are studied. Fuzzy neural networks have the advantage of fuzzy system and neural networks. So it is a good method for parameters identification. Main contents of this paper as follows:(1)Expound the basic elements of fuzzy set theory systematically, introduce the fuzzy control theory and the basic structure, introduce fuzzy modeling's contents and methods.(2)Against offline clustering algorithm need to know the input-output data in advance and not take into account the time relationship of input-output, this paper Study a on-line clustering algorithm which is simple but easy to implement, it is an effective method of fuzzy space partition.(3)Explain the basic theory of support vector machine and classification regression principle, introduce the similarity between support vector machine and fuzzy systems, prove the feasibility of extracting fuzzy rules through SVM. Due to the shortcomings of support vector machines, we study the least squares support vector machines and how to use APSO to optimize its parameters of kernel function. At last, we use this method to extract fuzzy rules from input-output data.(4) We shows the need for parameter identification, then introduce the methods of parameter identification and the basic theory of fuzzy neural network.At last, we propose the fuzzy neural network identification algorithm of Mamdani and TS models respectively.(5)Integrate the structure identification and parameters identification method in this paper, we propose a new fuzzy modeling method. Though the analysis of simulation results, we conclude that it is a effective method for fuzzy modeling.