Commande Prédictive Non Linéaire en Utilisant Les Systèmes Neuro-Flous et les Algorithmes Génétiques

Abstract

With its performance and ease of implementation, the predictive control, has a big success in industry, unfortunately it is well suited for linear systems, and often there are no perfectly linear model. With refer to advances in modeling theory of nonlinear systems; the choice of system is oriented to the use of fuzzy-neural models that have shown its efficiencies in many applications. In general, The NMPC controller minimizes a loss function at every sampling instant. A number of future control moves (the control horizon) is calculated each time, and the first control move of this control horizon is implemented. The calculations are then repeated at the next sampling instant from the new measure of output. The major problem that arises when using nonlinear models is that the optimization problem to be solved online is generally no convex, and the computation time can be very big and convergence to a global minimum is not sure, unfortunately the classical numerical methods used requires a considerable computation time with no guarantee of convergence to a global minimum. Genetic algorithms are powerful methods of resolution, can provide an effective solution to this problem. In this work, we will study the use of models for the construction of fuzzyneural predictor of the controlled system, using genetic algorithms for the optimization phase, all in a structure of a predictive control of nonlinear systems.