Abstract:Kalman filter demands prior knowledge of mathematical model and noise characteristics about the known system,but usually the physical system can't meet the demand.In view of this problem,based on the function approximation capability of the neural network and local and multi-resolution characteristics of wavelet transform,a design method of wavelet neural network filter is proposed to estimate the state of the system.Simulation shows that tracking accuracy of this method is better than that of Kalman filter.

Design Method of Filter Based on Wavelet Neural Network