Analysis And Research On Data-Driven Fuzzy System Modeling

Data-driven fuzzy system modeling is a very active area in current research, which has great potential in solving the problem of nonlinear and serious uncertainty system. Based on fuzzy control, clustering algorithm and artificial neural network, this paper studied how to build a model from a large number of input and output data, mainly studied T-S fuzzy system modeling methods. The principal tasks are as follows:Firstly, the fuzzy system composition and common types were detailed analyzed, and the universal approximation of the T-S model was discussed; at the same time, the general data-driven fuzzy system modeling methods were introduced and analyzed.Secondly, the stability of T-S model was analyzed. According to the determination of the number of categories and the selection of initial cluster centers, the fuzzy C-means clustering algorithm was improved, an improved fuzzy clustering algorithm was proposed. Used subtract clustering algorithm to get the initial value of cluster centers and the cluster number, then divided the input space by fuzzy C-means clustering method, after that extracted the fuzzy rules from the input and output data, which could made identification model expressed by a number of local line models. Then recursive least squares was used to identify the consequent parameters, thereby established a non-linear system of T-S fuzzy model, avoided the blindness and randomness of the membership functions determine and the fuzzy rules extraction, improved the efficiency and accuracy of the fuzzy rules. Simulation results showed that the new method can effectively reduce the ambiguity of space division and the time of parameter test, and also the system is more precise.At last, combined the knowledge expression ability of the fuzzy control with the self-learning adaptive ability of the neural network, a kind of T-S fuzzy neural network modeling method was proposed. Used the improved fuzzy clustering algorithm to adaptively obtain the precise cluster number and membership parameters, identify the former RBF networks. Used improved BP algorithm to train the after network weights and optimal parameters, so as to establish the T-S fuzzy neural network model, which had made full use of the advantages of self-training speed of RBF neural network and the strong ability of nonlinear fitting of BP neural network. Finally, this method was applied to nonlinear system identification, simulation results showed the effectiveness and feasibility of the model.