A Hidden Markov disturbance model for Offset-free linear model predictive control

Abstract

Model predictive controllers are often designed with integral action to impart robustness. For this, disturbance models are usually employed. It is customary to append integrated white-noises to either the input or output channels. However, neither by themselves may be adequate representations in the presence of switching disturbance patterns that are typically witnessed in process industries. In order to handle such scenarios, we first propose a differenced state-space formulation that can incorporate both input and output disturbances while retaining detectability. Then, we couple it with Hidden Markov Model (HMM) to express the switching characteristics of the disturbances. This bypasses the need to add artificial noises into state variables to consider both the input and output disturbances, as previously suggested. Simulation examples are provided to highlight closed-loop performance improvement as a result of the proposed formulation.