A new fuzzy wavelet neural network design for time series prediction

In this thesis, two different neuro-fuzzy models which use wavelet basis functions in its processing units are proposed for time series prediction and system identification problems. The structure of introduced models comes from the idea of adaptive neuro-fuzzy inference system (ANFIS) which is used for obtaining fuzzy rule base from the input-output data of an unknown function. The first model in this thesis is called as adaptive wavelet network (AWN) in which wavelet basis functions are used as membership functions in antecedent part of the rules whereas mostly Gaussian type membership functions are used in the ANFIS. In the second model which is called as fuzzy wavelet neural network (FWNN), wavelet basis functions are used in consequent part of the rules instead of zero or first order polynomial function in the ANFIS. A fast training gradient algorithm based on quasi-Newton methods is used to obtain the optimal values for unknown parameters of the FWNN models. The AWN models are trained by a hybrid algorithm which combines gradient algorithm with least square estimation. Simulation examples of some benchmark problems in the literature are also given to illustrate the effectiveness of models.