Dynamic Neural Networks for Modeling and Control of Nonlinear Systems

Abstract

Abstract This paper considers the design of a dynamic neural network (DNN) for modeling of a class of nonlinear systems for the purpose of real-time control. The primary contribution of the paper is in developing a DNN estimator with a stable training technique for on-line modeling of unknown (black box) dynamic nonlinear systems. The DNN acts as a generic model of the system, which can be trained on-line and, hence, can be utilized for the implementation of an adaptive model-based control strategy. The training of the network is based on a novel scheme that arranges the outputs of the hidden layer of the DNN into a set of basis functions. This allows for the derivation of a stable rule for the training of the DNN's weights and does not require random initialization of the weights.