全局优化RBF网络的一种新算法

神经网络的输入变量、隐含层结点以及中心的选择对模型的性能都有重大的影响,以前的研究一般只考虑优化网络的参数或其结点数。为解决这个问题,提出了一种新的全局优化算法来自动选择RBF神经网络的输入变量和结点数目,并同时优化其参数。在提出的算法中,RBF网络的结点数目、输入变量的选择和参数都采用二进制编码,并用遗传算法来优化。为提高算法的性能和收敛速度,在遗传算法优化的同时引入了一种高性能的基于梯度的局部搜索算子(结构化的非线性参数优化方法)来优化RBF网络中的参数。Box-Jenkins煤气炉标准时间序列的预测问题被用来检验算法的性能。实验结果表明,提出的算法可以得到非常"紧凑"的RBF网络,且其性能优于其他一些算法。

A Novel Algorithm for Global Optimization of RBF Neural Networks

The input vector,the number of hidden nodes and the centres of neural networks have significant effects on the modeling performance.The majority of the previous work only focused on determining the parameters of the networks or the number of hidden nodes.In this paper,anovel algorithm is proposed for automatic selection of the number of nodes and the proper input variables,and simultaneously optimizing the parameters of the RBF neural networks.In the proposed algorithm,the nodes,inputs and parameters of the RBF networks are represented in one chromosome and evolved simultaneously by genetic algorithm.To improve the performance of the propose algorithm and accelerate the computational convergence,a gradient-based local search strategy is introduced to optimize the parmeters of the RBF networks.The performance of the presented approach is evaluated by the Box-Jenkins time series prediction problem.It is shown by the simulation results that the proposed algortihm produces parsimonious RBF networks,and obtains better results than some other existing methods.