Constrained Controller and Observer Design by Inverse Optimality

Abstract

Model predictive control (MPC) is often tuned by trial and error. When a baseline linear controller exists that is already well tuned in the absence of constraints and MPC is introduced to enforce them, one would like to avoid altering the original linear feedback law whenever they are not active. We formulate this problem as a controller matching similar to the works of Di Cairano and Bemporad (2009), Di Cairano and Bemporad (2010), and Tran et al. (2015), which we extend to a more general framework. We prove that a positive-definite stage-cost matrix yielding this matching property can be computed for all stabilizing linear controllers. In addition, we prove that the constrained estimation problem can also be solved similarly, by matching a linear observer with a moving horizon estimator. Finally, we discuss various aspects of the practical implementation of the proposed technique in some examples.