Abstract
The idea of combining both wavelets and neural networks has resulted in the formulation of wavelet network, whose basic functions are drawn from a family of orthonormal wavelets(1), which absorbs the advantage of high resolution of wavelets and the advantages of learning and feedforward of neural networks. The usual method to train wavelet networks is the backpropagation (BP) algorithm described by Rumelhart et al. However, this algorithm converges slowly for large or complex problems. In this paper, we propose to train wavelet network for nonlinear time series prediction by using the Unscented Kalman filter (UKF), which outperforms the conventional BP method and several other reference methods. Several simulation results are presented to validate the proposed method.
