Novel Algorithms Of Fuzzy Inference And Adaptive Control Based On Fuzzy Wavelet Neural Networks

Intelligent Control has emerged as an advanced control under the development of researching so-called '3C' problems. At present, the well-known theories and methods of Intelligent control are: Expert control, Fuzzy Control, Neural Network Control, Learning Control (including Iterative Learning Control), Hybrid control and general theory of Intelligent Control System.As a branch of Intelligent Control, Fuzzy Control plays an important role in industrial fields, especially in series of family-electric production. In past thirty years, Fuzzy Control is based on CR1 inference algorithm presented by Zadeh, which is a composition operation of fuzzy relation and included single inference. From logic semantic, there exists some limitations on the CR1 algorithm, and also the algorithm has not characteristic of reverse. To these limitations of CR1 algorithm, Guo-Jun Wang has proposed the triple I principle in literature [3] and given a R0-type Triple I algorithm.In order to find out a novel fuzzy inference algorithm being superior to CR1 algorithm and a novel fuzzy neural network, the author deduces nine kind of triple I algorithms by nine fuzzy operators. Further more, the author compares these algorithms with the ordinary intuitional rules respectively and gives a conclusion that RL-type Triple I algorithm is rather more closely to the intuitional rules.Because fuzzy neural networks require the conditions of continuous and differential on membership functions of fuzzy sets and fuzzy operators of fuzzy inference, by means of comparison study, the author obtains a conclusion that the fuzzy neural networks with production operator and CR1 inference algorithm are very fit to study the problem of control. In the last part, for a class of nonlinear dynamic system, the author has designed an 'Adaptive Fuzzy wavelet Neural Network Controller' based on T-S model and slide mode control, also the asymptotical stability of the closed-loop system is proved. By simulation study for a vehicle-carried inverse pendulum, the result shows that the tracking errors asymptotical converge to zero quickly than the presented method.