State space neural networks in non-linear adaptive system identification and control

Non-linear control design methods, such as feedback linearisation and output regulation theory, are effective techniques for solving non-linear control problems. However, they assume the exact knowledge of the true model and that the states are completely accessible, which is not always true in practice. The starting point for this paper is to develop a viable practical control strategy by combining the modelling capabilities of state space neural networks with the effectiveness of the output regulation theory. By using input and output measurements and based on Lyapunov stability and non-linear observation theories a stable on-line learning methodology for the network parameters is proposed. A practical implementation for controlling a multivariable process is included to illustrate the effectiveness of the proposed adaptive control methodology. Copyright (C) 2001 IFAC.