Research On System Identification And Control Based On Support Vector Machine And Fuzzy Inference

Fuzzy inference system, or called fuzzy system for short, is one advanced computing frame which sets its foundations on fuzzy sets theory, fuzzy if-then rules and fuzzy logical inference. Fuzzy system simulates the thinking modes of people, and realizes the different decision mechanisms under different circumstances for the uncertain, imprecise problems. Fuzzy inference is based on fuzzy rule database, which is just the original point of this thesis. Support vector machines (SVM), which set their mathematical foundation on Statistical Learning Theory, provide a new pattern identification and method for small data learning. Focusing on the present methods of extracting fuzzy rules, a new fuzzy inference model based on SVM is proposed, and some research and application works have been committed as listed followings:1)The mathematical foundation and realization theory of SVM are studies. The mathematical foundation of SVM, that is Statistical Learning Theory, is introduced, and the advantages and realization algorithm in solving small data problems are explained, as well the geometrical characteristics and the choices of kernel function are discussed.2)A fuzzy inference model based on SVM is proposed. According to the common points of fuzzy inference and SVM, fuzzy inference and SVM are combined to construct a new fuzzy inference model based on SVM, and the realization algorithm is brought forward. Then a simple simulation experiment is given to demonstrate its geometrical explanation by comparing with the common methods on fuzzy rules extraction.3)The new fuzzy inference model based on SVM is applied to nonlinear system identification. Taking the prediction of chaos time-series and three-stages nonlinear system identification for examples, it is induced that the proposed model can obtain higherprediction and identification accuracy by comparing with the common intelligent methods.4)The new fuzzy inference model based on SVM is applied to nonlinear system control. Taking beam-ball system control and double inverted pendulum control for examples, it is induced that the proposed method can obtain higher control accuracy and better stability performance, and has much application values.