Contribution to improve the Performance of neural network Model predictive control

Abstract

This doctoral thesis proposes different control approaches to enhance the performance of the neural network model predictive control. For this purpose, a feed forward neural network model is suggested to build the prediction model for the model predictive control. Various metaheuristic algorithms are integrated to solve the nonlinear and non-convex optimization problem of the neural network model predictive control, such as particle swarm optimization, teaching learning-based optimization, and driving training-based optimization algorithms. Four algorithms are introduced and tested on different systems. In the first algorithm, the teaching learning-based optimization algorithm is used to solve the neural network model predictive control optimization problem, with the implementation of a reduction weighting coefficient in the cost function.This algorithm is applied to control the permanent magnet synchronous motor speed. As for the second algorithm, a driving training-based optimization algorithm is proposed to solve the optimization problem of the neural network model predictive control. This algorithm is used to control the speed of an electric vehicle. Furthermore, a sequence of reduction weighting coefficients is integrated into the cost function of the previous algorithm. This algorithm is applied for the tracking control of a two degrees-offreedom robot manipulator.To ensure a closed-loop stability of the controlled system a Lyapunov function is incorporated as a constraint in the cost function of the neural network model predictive control based on the driving training optimization approach. The re- sulting control algorithm is applied to the speed control of an induction motor, then compared against a set of control approaches including optimized PID based on particle swarm optimization algorithm, fuzzy PID, neural network model pre- dictive control based on different optimization algorithms, such as particle swarm optimization, teaching learning-based optimization algorithm, and driven training- based optimization algorithm.