基于递阶进化规划的RBF网络设计新方法

A new design method of RBF neural network based on hierarchical evolutionary programming

针对RBF神经网络的特点 ,提出一种递阶进化规划算法 ,利用此方法同时对网络的拓扑结构和网络参数 (权值、隐节点中心和形状参数 )进行优化 ,克服梯度算法需要求导且易陷入局部极小的弱点 ,分别对单入单出和多入单出非线性函数的建模问题进行了仿真 。

Aimed at the feature of RBF neural network, a hierarchical evolutionary programming for this network design is proposed to optimize its topological configuration and parameters such as the weights, implicit node centers, and shape parameter, so that the malady of necessary derivation and local minimum trap with the gradient algorithm can be overcome. The mapping of SISO and MISO system'nonlinear function is simulated, respectively, by which the effectiveness of the algorithm presented is tested.