Fuzzy Wavelet Neural Networks with hybrid algorithm in nonlinear system identification

This paper presents a hybrid learning algorithm for Fuzzy Wavelet Neural Network (FWNN) and uses it in nonlinear system identification. The algorithm gives the initial parameters by clustering algorithm, then updates them with a combination of Back-Propagation and Recursive Least Square methods. The proposed approach is tested for identification of nonlinear systems commonly used in the literature. It is shown that with the proposed approach the number of rules and complexity of the structure will be reduced while the performance is better than the previous works. In order to comparison, Gradient Descent algorithm is applied in the same conditions. The results indicate a superior convergence speed for the proposed algorithm in comparison to Gradient Descent method which is commonly used in the literature.