Artificial Neural Networks (NN) is a well-known tool among artificial intelligence techniques that are able to reproduce arbitrary nonlinear relationships existing between input and output variables. Model based Predictive Control (MPC), or simply predictive control, is a family of control schemes that uses a model from the plant as a predictor of the future plant outputs and hence optimizes the future control inputs for the minimum future errors and minimum control energy. Among this family Generalized Predictive control (GPC) is one of the most famous. In another part of this work [5],, a neural network representation is shown to be suitable for modeling a small gas turbine engine (SR-30). In the present paper, this model is used in a model-based predictive control scheme. The results of this controller are compared with a classical Proportional-Integral-Derivative (PID) controller tuned offline with a genetic optimization technique. Both are tested on the SR-30 turbojet engine model. PID controller cannot cope with model changes in the whole operating range of the engine and therefore a predictive control scheme is then proposed as a solution to this problem. A neural model is used as a predictor for the calculation of GPC parameters. The nonlinear system free response is obtained by recursive future predictions while the dynamic response matrix is obtained by instantaneous linearization of the input /output relation.
The results illustrate the improvements in control performance that could be achieved with a neural predicative scheme compared to that of a classical PID controller.