Tracking Explicit Model Predictive Controllers for Low-Level Control Applications

Abstract

In explicit model predictive control (eMPC), the bulk of the computational load of classic MPC is performed during the off-line design stage, which enables the controller to be implemented on standard industrial automation equipment and for processes with fast dynamics, however, the computational complexity restricts applicability to small-scale control problems. The approach is appealing for a niche of control applications, but practical applications have been scarce so far. We describe one possible offset-free tracking setup that allows implementation of eMPC with relatively fast sampling and reasonably long horizons for practical applications. The applicability of eMPC is illustrated by an experimental case study of cooling-water temperature control in a biogas-fuelled combined-heat-and-power production unit, where eMPC replaces a pre-existing single-loop PID controller, with the aim of reducing unnecessary excursions of the cooling water temperature away from its set-point in the critical range near output constraints. eMPC controllers were designed using local linear analysis and tested both on a simplified simulation model and experimentally on the CHP unit. The performance improvements due to tuning of eMPC feedback action and due to the constraints-handling ability were examined, and several implementation issues related to the practical implementation of eMPC are discussed.