Nonlinear system identification by feedback GMDH-type neural network with architecture self-selecting function

Abstract

The feedback Group Method of Data Handling (GMDH)-type neural network algorithm is proposed and is applied to the nonlinear system identification. In this feedback GMDH-type neural network algorithm, the optimum neural network architecture is automatically selected from three types of neural network architectures such as the sigmoid function type neural network, the radial basis function (RBF) type neural network and the polynomial type neural network. Furthermore, the structural parameters such as the number of feedback loops, the number of neurons in the hidden layers and the relevant input variables are automatically selected so as to minimize the prediction error criterion defined as Prediction Sum of Squares (PSS). The identification results show that the feedback GMDH-type neural network algorithm is useful for the nonlinear system identification and is ideal for practical complex problems since the optimum neural network architecture is automatically organized.