An orthogonal neural network for nonlinear function modelling

Abstract

Neural networks provide one of the means for identification or control of a nonlinear system. Multilayer feedforward network is often used in those areas. Using this kind of neural network may result in premature termination of learning before the network weights reach a global minimum, a small training convergence, an unsuitable weight parameters initialize and the stability for on-line implementation. In this paper, an orthogonal neural networks (ONN) is presented. The ONN’s hidden layer consists of neurons with orthogonal activation function. The ONN have a much higher learning speed than the multilayered feedforward networks.