Fast Nonlinear Systems Modelling Via Neural Networks: A Unified Framework and its Applications

Abstract

This paper presents a unified framework for the use of neural networks for the modelling of unknown nonlinear systems. The systems are assumed to be expressed by a unknown nonlinear ARMA (i.e., NARMA) model including a set of unknown parameters. At first, it is assumed that the nominal (initial) values of these parameters are known during an initial operation period of the system. Incorporating the nominal parameter set into the structure of neural network allows one to establish a neural network model for the system via online training of the weights during the initial operation period. Using the trained neural network, the estimation of the parameters is achieved by incorporating the estimated parameters into the neural network model and by constructing a gradient descent-based estimation algorithm. Both associative memory and multilayer perceptron neural networks were used to formulate the algorithm. The applications of the unified scheme to both the fault diagnosis and the control of unknown nonlinear systems have been discussed, and desired results have been obtained via comparing existing neural network-based modelling approaches.

Key Words:

• Nonlinear systems
• neural networks
• training algorithm
• fault detection and diagnosis
• control systems
• adaptive control
• parameter estimation

Additional information

Notes on contributors

Hong Wang

Hong Wang received the M.Sc. and Ph.D. degrees in control engineering from Huazhong University of Science and Technology, P.R. China, in 1984 and 1987, respectively. He then worked as a post-doctoral research fellow at Salford, Brunei, and Southampton Universities, U.K., between 1998 and 1992. He joined the University of Manchester Institute of Science and Technology (UMIST) as a lecturer in 1992 and was a senior lecturer between 1997 and 1999. At present, he is a reader in process control. His main research interests are advanced modelling and control, neural networks, and bounded stochastic distribution control. Dr. Wang is the author of three books published by Pergamon Press, Pira International, and Springer-Verlag, and is the author or co-author of 70 published technical papers in his research areas.

Aiping Wang

Aiping Wang received the B.Sc. in mathematics from Huaibei Normal College and an M.Sc. degree in computing science from Chan Sa University, P.R. China, in 1978 and 1987, respectively. She then worked at Huaibei Normal College as lecturer, assistant professor, and professor in mathematics. Her research areas are advanced computing algorithms, neural networks, and process automation. She has co-authored one book and has also published widely in the above-mentioned research areas.