Economic MPC with Infinite Horizon

There is a growing interest in the development of MPC algorithms that replace the traditional quadratic objective function with a purely economic objective. However, implementation of such algorithms can result in unexpected and sometimes pathological closed-loop behavior, including inventory creep-down and bang-bang actuation during active constraint intervals. This paper will illustrate some of these issues. While an infinite horizon formulation seems to be a natural route to avoiding the above myopic behavior, implementation with an infinite horizon presents a number of technical and conceptual challenges. To address these issues, a number of subtle changes to the original problem formulation will be introduced and thus will allow for a computationally robust implementation of the infinite-horizon controller.