Commande non lineaire a modele predictif pour une machine asynchrone

Induction machines are widely used in industry because they are more rugged, reliable, compact, efficient and less expensive compared to other machines used in similar applications. They however, represent highly nonlinear, coupled, multivariable, complex control plant with unknown disturbances and time varying parameters requiring complex control algorithms. This PhD thesis deals with the development of a high performance nonlinear predictive control induction motor drive. The research work is directed towards improving trajectory tracking capability, stability guarantee, robustness to parameters variations and disturbance rejection. The main part of nonlinear predictive control is the system behavior prediction model. Two methods are used for the design of the prediction model. First, the design is based on neural networks modeling. The control law in this case is called neural predictive control. The other prediction model is obtained using mathematical tools of differential geometry for nonlinear predictive control.; In the neural predictive control, a multivariable modeling of the machine is done using a multi-layer neural network in order to design a nonlinear predictor. Optimal control is obtained by minimization of a quadratic criterion. A reference model, obtained from the inversion of the machine model, is included in the optimization criterion, which helps to improve the optimization.; In nonlinear predictive control, two types of control algorithms are proposed in this work. First, a multivariable controller is used for the system control, with the rotor speed and the rotor flux norm as outputs. Then, a cascaded controller is used for electromagnetic torque, rotor flux norm and rotor speed control. The prediction model is obtained by using the Taylor series expansion. The nonlinear predictive controller is enhanced by embedding a disturbance observer, which behaves like a PID or PI speed controller according to the relative degree of the speed output. This combination is called nonlinear PID (or PI) predictive control . Initially, control action is carried out assuming that all the states are known by measurement. Then, in the next step an observer is implemented instead of measurement. Here, the Lyapunov method is used to prove the global stability of the complete control scheme (process + control + state observer).; The trajectories tracking, robustness to parameters variations and disturbance rejection are successfully achieved using this nonlinear predictive controller. Therefore, we believe that this work constitutes a major contribution to the domain of variable speed induction motor drive.