多层RBF网络的自适应遗传算法及其在实函数逼近中的应用

Adaptive Genetic Algorithm for Multilayer RBF Network and It’s Application on Real Function Approximation

将自适应遗传算法应用于多层RBF神经网络的学习,对隐层核函数的中心和宽度值进行同时优化,并用正则最小二乘法求权重,以完成网络的构建。应用该学习法进行实函数的逼近,实验证明了该算法比多层RBF网络的聚类学习法具有更高的实函数逼近精度。

With the application of adaptive genetic algorithm to the training of multi-layer RBF networks and the optimization of the hidden layer centers and width values and using regularized least squares method, weight vectors is obtained. Computer simulation shows that the precision of real function approximation by this algorithm is much higher than the precision by clustering algorithm for multi-layer RBF networks.