基于粒子群优化的积单元网络预测混沌序列

Predicting Chaotic Time Series Using a Product Unit Network Based on PSO

提出一种基于粒子群优化 (PSO)的积单元神经网络 (PU NN )预测混沌时间序列的方法 .PUNN信息存储能力强 ,但是它的训练却很困难 .PSO是一类基于群智能的随机全局优化技术 ,故该文用 PSO算法训练 PUNN.对 Mackey-Glass混沌序列分别用 PU NN和模糊神经网络方法做的单步及多步预测对比实验结果说明不仅用 PSO算法训练PUNN是有效的 ,而且用 PU NN预测混沌时间序列是一种有效的方法 .

A new method of predicting chaotic time series using a product unit neural network (PUNN), which is trained by a particle swarm optimizer (PSO) is proposed in this work. PUNNs have power information storage capacity, but the usual optimization algorithms like gradient descent cannot train the PUNNs efficiently. PSO is a population based stochastic global optimization technique, so PUNNs are trained by a PSO in this paper. The application considered is Mackey-Glass chaotic time series. Experiment results for single step and multi step prediction are obtained both from PUNN and fuzzy neural networks methods. The work not only demonstrates that PSO is efficient for training PUNNs, but it also highlights the advantages of the proposed method.