一种进化RBF神经网络的模型及其训练算法

A MODEL OF EVOLUTIONARY-RBF NEURAL NETWORKS AND ITS TRAINING ALGORITHM

径向基函数神经网络 (RBFNN)具有最优逼近和全局逼近的特性 ,在函数拟合方面优于传统的 BP网络 ,因此被广泛应用于非线性时间序列分析等领域 .本文针对时间序列中的非平稳数据 ,结合差分平稳化与分阶遗传的思想 ,提出一个新的进化 RBF神经网络的模型及其训练算法 .通过实例分析表明 。

Because the radial basis function neural networks (RBFNN) have the performances of best approximation and universal approximation, they are widely applied in those fields of nonlinear time series analysis and so on. In this paper, a novel model called Evolutionary RBF and its training algorithm are presented for the non stationary data in time series, combined with the methods of difference and hierarchical genetic algorithm. The simulation results confirm the superior performance of the evolutionary RBF over the classical neural networks, and the former is more fit for the non stationary time series problems.