Short term load forecasting by using neural networks with variable activation functions and Embedded chaos algorithm

Abstract

In this paper a novel variant activation (transform) sigmoid function with three parameters is proposed, and then the improved BP algorithm based on it is educed and discussed, then Embedded Chaos-BP algorithm is proposed by means of combining the new fast BP algorithm and chaos optimization algorithm, Embedded chaos-BP algorithm converges fast and globally, and has no local minimum. The efficiency and advantage of our method is proved by simulation results of nonlinear function and prediction results of short-term load based on the improved and traditional BP ANNs. © Springer-Verlag Berlin Heidelberg 2006.