摘要
该文提出了一种新的RBF神经网络的设计方法 ,采用遗传算法对RBF神经网络的隐层节点中心值进行进化优选 ,用自适应梯度下降法选择隐层节点高斯函数的宽度 ,用递推的最小二乘法训练RBF神经网络的权值 ,仿真结果证明了该方法的有效性。
A new design of RBF neural networks is proposed. Genetic algorithms is used to optimize the centers of the hidden units of RBF networks. The adaptive grads-dropping algorithms is used to select the widths of the Gauss funtion. The RLS algorithms is used to train the weights of RBF networks. The simulation results testify the effectiveness of the proposed design method.
出处
《计算机仿真》
CSCD
2003年第11期67-69,共3页
Computer Simulation
