混沌序列的T-SK反馈模糊神经网络预测

Prediction of chaotic time series using T-SK recurrent fuzzy network

针对模糊神经网络预测混沌系统输入节点数目的确定随意性较大及敛速度慢的缺点,提出T SK反馈模糊神经网络(T SKrecurrentfuzzynetwork,TRFN)。同时采用两阶段学习算法:先进行结构学习来确定TRFN的最佳结构,再利用基于混沌动态量的BP算法对神经网络进行参数学习,提高了收敛速度和预测精度。应用此网络和相应的学习算法,对Henton序列进行了预测,与传统的模糊神经网络相比,在节点数目较少的情况下,取得了更快更精确的预测结果,验证了该方法的有效性。

Aimed at the problem that there is no criterion to determine the number of input nodes of a fuzzy network which is used to predict chaotic series, a T-SK recurrent fuzzy network (TRFN) is proposed, and a two-step learning arithmetic is adopted: the first step is structure learning to decide the best structure of the network and avoid the waste of nodes; the second one is parameter learning to improve BP arithmetic and increase the convergence rate. By using the TRFN and its arithmetic to predict the Henton chaotic time series, a better result is obtained, which proves that the method proposed is valid.