Constrained Nonlinear Predictive Control Using Neural Networks and Teaching–Learning-Based Optimization

In this work, an efficient constrained nonlinear predictive control algorithm, based on the meta-heuristic optimization strategy, is proposed. The aim is to develop a predictive control algorithm having a simple design and implementation procedure on the one hand, and that can give satisfactory performance with a large class of nonlinear systems on the other hand. To achieve this goal, a feedforward multilayer neural network is used to predict the future outputs of the system, and the optimization problem of predictive control is resolved using teaching–learning-based optimization strategy. Due to their interesting proprieties, feedforward neural networks are widely used in nonlinear systems identification and control. Indeed, they have a simple structure and can accurately approximate any nonlinear mapping. The teaching–learning-based optimization is used in a large number of applications in different fields of engineering and has gained wide acceptance among the optimization researchers community. Unlike other metaheuristic algorithms, this algorithm requires only common controlling parameters like population size and number of generations for its working; it does not require any algorithm-specific parameters. To assess the effectiveness of the proposed control algorithm, the control of the model of the continuous stirred tank rector and the 2-DOF manipulator robot model is considered. A comparative study, using the conventional PID controller, the fuzzy logic control, the computed torque control and the neural network-based model predictive control using particle swarm optimization, is carried out. To further demonstrate the effectiveness of the proposed controller, the control algorithm is numerically implemented and applied to a system with fast dynamics, namely the induction motor. The obtained results, through the simulation and the experimental study, indicate that the proposed controller presents better control performance than the other controllers. Furthermore, the experimental study shows that the developed control strategy can be effectively used to control, in real time, systems with fast dynamics.