Modelling.pdfVIP

1992 ACC/WA2 Modelling, Identification and Stable Adaptive Control of Continuous-Time Nonlinear Dynamical Systems Using Neural Networks Marios M Polycarpou and Petros A. Ioannou Department of Electrical Engineering-Systems University of Southern California Los Angeles. CA 90089-2563, U.S.A Abstract The approximation capabilities of static sigmoidal type networks and of radial basis function networks have been studied by several Several empirical studies have demonstrated the feasibility of em- research groups (see for e.g. [6J-7]). In Section 2 we use these results ploying neural networks as models of nonlinear dynamical systems. to show that a proposed general network configuration composed of This paper develops the appropriate mathematical tools for synthe- static neural networks and dynamical components (such as stable fil- sizing and analyzing stable neural network baed identification and ters) forms a type of recurrent network capable of approximating a control schemes. Feedforward netwok architectures are combined large clam of dynamical systems. More precisely, it is shown that with dynamical elements, in the form of stable filters, to construct a there exist a set of weights such that for a given input, the out- general recurrent network configuration which is shown to be capable puts of the real system and the proposed recurrent neural network of approximating a large clams of dynamical systems. Adaptive iden-