A rapid learning and dynamic stepwise updating algorithm for flatneural networks and the application to time-series prediction

摘要

A fast learning algorithm is proposed to find an optimal weightsnof the flat neural networks (especially, the functional-link network).nAlthough the flat networks are used for nonlinear functionnapproximation, they can be formulated as linear systems. Thus, thenweights of the networks can be solved easily using a linear least-squarenmethod. This formulation makes it easier to update the weights instantlynfor both a new added pattern and a new added enhancement node. A dynamicnstepwise updating algorithm is proposed to update the weights of thensystem on-the-fly. The model is tested on several time-series datanincluding an infrared laser data set, a chaotic time-series, a monthlynflour price data set, and a nonlinear system identification problem. Thensimulation results are compared to existing models in which more complexnarchitectures and more costly training are needed. The results indicatenthat the proposed model is very attractive to real-time processes