RBF neural network with optimal selection cluster algorithm and its application

In this paper, the RBF neural network (RBFNN) is used as a modeling framework to solve the identification problem of nonlinear systems. First, it is proposed a kind of optimal selection cluster algorithm. By this algorithm, it is optimally gained the hidden layer node number of RBFNN in terms of input samples. At the same time, the initial parameter values of RBF are obtained. Then, the parameters of RBF are estimated by the gradient algorithm with momentum terms, and the weights of RBFNN are identified by the recursive least square algorithm. The above two algorithms are alternately iterated. By the above hybrid algorithms, it is not only raised identification precision of RBFNN, but also improved the generalization property of the net. The validity of the scheme described is proved.