Artificial fish-swarm algorithm (AFSA) is a novel method to search global optimal valuebased on the behavior of autonomous animals proposed by Li Xiao-lei in 2002. The algorithmhas the upstanding ability of conquering local optimum and reaching global optimum. Theartificial fish-swarm neural networks (AFSNN) model is constructed, which is the artificialneural networks (ANN) trained by AFSA. The AFNN has the satisfied global ability of search,compared with the ANN trained by BP algorithm, SA, EA. AFNN is then used to theshort-term load forecasting in electronic system, and the artificial fish-swarm neural networksforecasting model (AFSNNFM) is formed. In order to improve the stability of the algorithmand its ability to search the global optimum, we propose an improved AFSA. When theartificial fish swarm's optimum value is not variational after several generations, we addleaping behavior and change the parameter of artificial fish. By this way, we can avoid localoptimum and increase the probability to get the global optimum. An improved ArtificialFish-swarm Algorithm (IAFSA) for the optimization of feed-forward neural networks(IAFSNN) and a forecasting model (IAFSNNFM) based on this method are presented for thefirst time in this paper. The calculating result demonstrates that the improved model has betterglobal convergence and stability. |