基于RBF神经网络的非线性系统的预测

Prediction of Nonlinear Systems Based on RBF Neural Networks

对于非线性系统的预测辨识,提出用动态节点生成构造性RBF神经网络作为预测模型,且RBF神经网络的学习算法采用一种新的全监督式学习算法,即神经网络隐层引入新节点时,通过使新节点的输出尽可能逼近残差序列的方向来获取网络参数,从而减少学习误差,使网络输出能够较好的跟踪系统输出。仿真表明该学习算法的有效性。

Prediction identification for nonlinear systems is proposed, dynamic node creating training RBF neural network is adopted for predictive model, and the training algorithm of RBF neural network is a new training algorithm. During the training process , a new is created one by one to compensate the training error. The parameters of the new node are obtained by approximating the direction of new node＇ s output to that of the training error so that network output can track system output well. Simulations indicate the efficiency of the new algorithm.