Robust control of a supermarket refrigeration system using multi-stage NMPC

Abstract

Using Nonlinear Model Predictive Control (NMPC), real world systems that are subject to stringent constraints can be controlled efficiently when accurate plant models are employed. Model uncertainties however may lead to poor performance of the controller and to constraint violations because of the wrong predictions. As in real applications, there often is a significant plant model mismatch, the controller must be robust to plant-model mismatch without deteriorating the performance of the controller significantly. Multi-stage NMPC is a robust scheme which has been proven to be less conservative than open loop worst case solutions because of the presence of feedback information at the future time stages is explicitly accounted in the problem formulation. In this paper, we study the application of this NMPC technique to a supermarket refrigeration system under uncertainty and show that the presence of uncertainties in the model lead to constraint violations when standard NMPC scheme is applied. The robust multi-stage NMPC scheme improves the controller performance by accounting for the uncertainties in the prediction and results in a reliable operation of the system.