Commande adaptative neuronale par retour de sortie des systèmes non linéaires

Abstract

In this work we have analyzed an output feedback adaptive neural control allowing direct update of neural network weights. This algorithm is based on control error minimization. The ideal control signal cannot be calculated, so the error control is estimated by a fuzzy inference system. The estimate having the correct sign, it influences only the leaning rate. A simulation study is carried out for two cases: a state feedback control and output feedback control based on a nonlinear high gain observer. We applied this approach to two nonlinear systems the inverted pendulum and a CSTR (Continuously Stirred Tank Reactor). The results of simulation showed the performances of the strategy in the two cases.