Non-Constant Prediction-Step MPC for Processes with Multi-Scale Dynamics

Abstract

Motivated by applications in minerals processing, a novel model predictive control scheme is presented with non-constant prediction step size. In the proposed scheme, the sampling rate of the prediction and control horizons changes, with nearer steps having a shorter sampling period than ones in the more distant future. This approach allows for fine tuning of the control trajectory over the near horizon, whilst still allowing for long prediction and control horizons to account for slow dynamics. This may find application in many chemical and minerals processing plants where process units have multiple time-scale dynamics. Extensions to decentralized control of process networks are presented. The approach is illustrated using an industrial case study.