Evaluation of Nonlinear Model-Based Predictive Control Approaches Using Derivative-Free Optimization and FCC Neural Networks

Nonlinear control methods have been researched with the objective of improving performance of control loop systems. Among such control methods, nonlinear model-based predictive control (NMPC) strategies present significant importance, mainly due to explicit performance optimization and constraint handling. NMPC depends on a representative nonlinear model of the process to be controlled and an adequate optimization method. This work focuses on these two aspects. Simulation tests with a wastewater treatment process model are presented, to evaluate the use of two optimization methods, differential evolution and bound optimization by quadratic approximation (BOBYQA), under different conditions. Experimental results using BOBYQA and a fully connected cascade artificial neural network in a pressure process are presented, showing a performance improvement comparing to a linear model predictive controller.