用进化规划思想优化径向基函数神经网络结构的均衡器

针对训练径向基函数(RBF)神经网络均衡器的随机梯度算法(SG)中,神经网络的结构是指定的并且所用训练样本较长的问题,引入进化规划思想,用进化规划方法确定径向基函数神经网络的结构,用基于最小均方(LMS)误差准则的自适应算法调整神经元到输出端的连接权重.蒙特卡洛仿真表明,用这种方法确定的均衡器可以达到与SG算法相同的性能,而所用训练样本很少,网络结构不需要事先指定.

Equalizer Using Evolutionary Programming to Optimize the Structure of the Radial Basis Function Neural Network

Stochastic gradient (SG) algorithm is used for training the radial basis function (RBF)neural network equalizer. The structure of the neural network is appointed at first, and the training samples used appeared too long. To solve the problem, the evolutionary programming method is introduced to find out the neural network's structure, and the adaptive algorithm based on the least-mean-square(LMS) error criterion is used to adjust the linking weights from the neurons to the output. Monte-Carlo simulations demonstrate that the performance of the proposed algorithm is the same as that of the SG algorithm, the training samples used become much shorter, and the structure of the network need not to be appointed beforehand.