Robust Model Predictive Control: Implementation Issues with Comparative Analysis

Abstract

Abstract Robust Model Predictive Control (RMPC) is related to a variety of methods designed to guarantee control performance using optimization algorithms while considering systems with uncertainties. Many RMPC methods were created with different formulations, however there is lack of studies comparing their performances using pre-established metrics. Thus, the objective of this paper is to analyze and compare different RMPC methods using a Continuous Stirred-Tank Reactor (CSTR) plant with parametric uncertainties as a case of study. The methods are sorted by Min-max MPC, Tube-based MPC and LMI-based MPC, which were carefully chosen based on their relevance in the current literature. The metrics used to compare the methods are separated in off-line, related to configuration of the controller prior to its use, and on-line, which are related to the computational cost of the algorithm and difficulties that may arise during its application. Based on simulation results, it will be shown that the LMI-based methods are slower for calculating the control signal, but are easier to implement. Whereas, both Min-max and Tube-based methods require calculation of robust control invariant sets which might be difficult for more complex systems, however they are much faster when comparing to LMI.