T-S Fuzzy System Structure Identification Based On Support Vector Machine And Algrithm Research

In the past 20 years, fuzzy system identification is the active research and application in intelligent control field. Since Zadeh proposed the fuzzy set theory in 1965, fuzzy sets and fuzzy control of the theoretical study and practical application have been carried out widely. Because of the complexity of the current information technology, the mathematic modeling of traditional method has powerless. With the uncertainty of the controlled device, the traditional theory deals with this problem difficultly. Therefore, the fuzzy system identification is one of the key of intelligent control theory.In particular, nearly three decades, fuzzy system modeling and identification have been studied by many scholars, but there are still many questions that need to be further explored. Fuzzy system identification includes structure identification and parameter estimation. Among them, the structure identification is core and challenge. When dealing with complex nonlinear system, the fuzzy system depends heavily on the input space dimension.Therefore, it is apt to "dimension disaster" and its generalization ability is not strong. How to get a compromise in fuzzy modeling is the main reaserch.The starting point of this paper is how to use a simpler and effective identification algorithm for improving the model generalization ability and reducing the complexity of the fuzzy modeling.In this paper, the algorithm based on sample data for fast extracting fuzzy rules has been studied. The algorithm used the nuclear part of the Mahalanobis distance to select probable support vectors in order to reduc the training sets size of support vector machine. In order to achieve the fast extracting T-S fuzzy rules, the way used the support vector regression and T-S fuzzy modeling to establish its equivalence. This method solved a certain problems, such as structure complexity of the fuzzy system identification, "dimension disaster", poor generalization ability and robustness. In this paper, the model was constucted though support vector machine at first, then the parameter learning of fuzzy modeling has been accomplished by the back-propagation algorithm and genetic algorithm. Furthermore, using a one-dimensio- nal, two-dimensional nonlinear function and chaotic sequences tests this model and based on self-organizing neural network and fuzzy clustering algorithm model are compared. Finally, fuzzy modeling based on support vector machine applied in the beam-ball system and handstand pendulum system. Results proved the effectiveness and reliability of this method.