Nonlinear System Identification Based On Layer-Proceeding Fuzzy Neural Networks

In this paper, we first analyze the deficiencies of traditional fuzzy neural networks and improve the learning algorithm. The Layer-Proceeding BP Neural Networks (LPBPNN) is given based on the hierarchical error and the learning rate which can cooperatively adjust. And then, fuzzy neural networks logic layers are stratified according to experts'experiences. The Layer-Proceeding Fuzzy Neural Networks (LPFNN) is put forward with the combination of the LPBPNN. At last, LPFNN is used in the identification of nonlinear systems. The examples prove the effectiveness of the algorithm.The paper is organized as followed:In Chapter 1, the background of the paper is introduced. The purpose and the signification of the problem are presented.In Chapter 2, the limitations of the traditional BP Neural Networks algorithm are discussed. And then the limitations of the improved algorithms used in the identification of nonlinear systems, such as Simulated Annealing and Genetic Algorithm, are pointed out. Considering the problems of training sequence samples of the scheduling, joining this combination of stratified thinking, cooperatively adjusting the errors and the learning rate, the LPBPNN is given based on the convergence theorem. The simulation examples show the effectiveness of the improved algorithm.In Chapter 3, the deficiencies of the traditional Fuzzy Neural Networks are analyzed; the stratified thinking is given based on the refinement of the experts' experiences. The LPFNN is put forward on the base of the combination of the LPBPNN. At last, a control model of LPFNN is designed.In Chapter 4, the LPFNN is used in the identification of a class of nonlinear systems. A class of the control models of LPFNN is given; the controller of the LPBPNN and the identifier of the LPFNN are designed. The problems of the error propagations, sample sequences, normalization of samples, error controls are also discussed. Finally, an actual example is illustrated.In Chapter 5, the conclusions are given to summarize the thesis work. At the same time, the prospect of the LPFNN is presented.