Time series prediction using genetically trained wavelet networks

The paper presents a contribution to the analysis of wavelet transfer function use in neural network systems and the discussion of some possible learning algorithms of such structures. Wavelets local properties both in time and frequency domains are stated at first giving motivation for wavelet networks application and providing bases for their initial coefficient estimation described recently. The main part of the paper is devoted to the network coefficients optimization using genetic algorithms as an alternative to the gradient descent method. Principles of the evolution techniques are presented for a simple system and then applied to a given time series modelling and prediction.