Wavelet Network based Online Sequential Extreme Learning Machine for Dynamic System Modeling

Wavelet network (WN) has been introduced in many applications of dynamic systems modeling with different learning algorithms. In this paper an online sequential extreme learning machine (OSELM) algorithm adopted as training procedure for wavelet network based on serial-parallel nonlinear autoregressive exogenous (NARX) model. The proposed model used as system identification for nonlinear dynamic systems. The main advantage of OSELM over conventional algorithms is the ability of updating network weights sequentially through data sample-by-sample in a single learning step. This attains good performance at extremely fast learning. The initial kernel parameters of WN played a big role to ensure fast and better learning performance. Simulation of the proposed scheme applied to nonlinear dynamic systems validates that WN-OSELM is superior in terms of identification accuracy and fast learning ability compared to NN-OSELM.