Exploiting parallelization in explicit model predictive control

Abstract

Traditionally Model Predictive Control (MPC) has been mainly restricted to processes with rather slow dynamics and with sampling times ranging from a few minutes to hours, such as the ones encountered in the areas of (petro)chemicals, minerals and metals. However, recent algorithmic advances (such as the explicit approach for MPC) allowed the application of MPC to problems arising in the automotive or power electronics industry where the time scales are in the milli-or even the microsecond area. In this study we aim to push the limit of explicit MPC even further by exploiting the computational power offered by parallel CPU architectures. We present the parallelisation of three different algorithms and we report experimental results showing how for certain problems, the parallelisation offers performances that top state-of-the-art approaches.