NONLINEAR PLANT IDENTIFICATION BY WAVELETS

This article combines the wavelet theory with the basic concept of neural network to define a new mapping network called adaptive wavelet neural network (wavenet). It is an alternative to feedforward neural networks for approximating arbitrary nonlinear functions. The wavenet algorithm consist of self-construction of the network and the minimisation of the error. In the first step, the neural network structure is defined,the network gradually recruits hidden units to effectively and sufficiently cover the time-frequency region occupied by a given target. The network parameters are updated simultaneously preserving the network topology. In the second step, the instantaneous errors are minimised using an adaptation technique based on the LMS algorithm. The approach is applied to nonlinear plants identification, using simulated data.