Overlapping-Horizon MPC: A Novel Approach to Computational Constraints in Real-Time Predictive Control

Abstract

Model predictive control (MPC) represents the state of the art technology for multivariable systems subject to hard signal constraints. Nonetheless, in many real-time applications MPC cannot be employed as the minimum acceptable sampling frequency is not compatible with the computational limits of the available hardware, i.e., the optimisation task cannot be accomplished in one sampling period. In this paper we generalise the classical receding-horizon MPC rationale to the case where n > 1 sampling intervals are required to compute the control trajectory. We call our scheme Overlapping-horizon MPC – OH-MPC for short – and we numerically show its attitude at providing a tunable trade-off between optimisation quality and real-time capabilities. 2012 ACM Subject Classification Computer systems organization → Real-time systems; Information systems → Process control systems