Estimation of the parameters of wavelet neural networks using simultaneous use of genetic algorithm and recursive least square

In this paper, a novel identification scheme based on wavelet neural network structure is proposed. The objective function for identification considered in this paper is the sum of squared error. In order to optimize this objective, the genetic algorithm (GA) which is a global optimization is used for the parameters which appear nonlinearly in the wavelet structure. Recursive least square algorithm is used for the parameters which appear linearly in the output of wavelet neural network because it is known to be an optimal estimator for these parameters. The proposed training algorithm is used to identify chaotic system and a highly nonlinear dynamical system. Simulation results show that the proposed method identifies input/output data with higher performance in terms of sum of squared error when it is compared to gradient descent method.