Using genetic algorithm in robust nonlinear model predictive control

Abstract

Publisher Summary This chapter discusses the simulation results of first-principle nonlinear model-based predictive control (NMPC) of a high-purity distillation column. Two different NMPC approaches are proposed to improve the robustness of the NMPC controller, in which the advantageous properties of genetic algorithm (GA) in solving successfully complex nonconvex nonlinear optimization problems are exploited. The first approach uses GA for pre-optimization in solving the on-line open-loop control problem. From the best solution obtained, the sequential-quadratic-programming (SQP) continues the solution. This approach improves both the control performance and the robustness of the NMPC. The computational burden in this case is decreased. The second approach uses GA to solve the optimization problem in parallel with the SQP improving, thus, the robustness of the NMPC, practically eliminating the controller failures due to the failure of the optimization and conferring to it a great importance for practical NMPC implementations.