Improved output constraint-handling for MPC with disturbance uncertainty

Many of the robust model-predictive controllers (MPC) developed to-date suffer from excessively conservative control because they rely upon open-loop predictions of future system uncertainty. Open-loop predictions overestimate the uncertainty in future process outputs, because they do not consider that future controller actions that will reduce this uncertainty. In this paper, we present a model-predictive controller that uses a closed-loop model to estimate the uncertainty in future process inputs and outputs due to stationary or non-stationary, stochastic disturbances. The new controller solves a stochastic program at each execution in order to determine the set of control moves that will optimize the expected performance of the system while maintaining the uncertain process output within its allowable bounds. As demonstrated by simulation studies, the proposed controller provides improved dynamic and constraint-handling performance when compared with nominal-model MPC and with robust MPC that rely upon open-loop uncertainty descriptions. Extensions for the input-constrained case are discussed.