Plant-wide Optimal Control with Decentralized Mpc

Most standard MPC implementations partition the plant into several units and apply MPC individually to these units. It is known that this strategy can lead to sub-optimal plant-wide control performance, especially if the units interact strongly. This paper tackles the problem of achieving optimal control performance in plants with such an MPC structure. A modeling framework, geared for use in MPC, that incorporates the interactions between the subsystems is employed. One may think that modeling the interactions and communicating the control actions between the controllers is sufficient to improve controller performance. We show that this idea is incorrect and can lead to closed-loop instability. A cooperation based MPC algorithm that converges to the plantwide optimum is developed. In practical implementations, the cooperation based MPC scheme may have to be terminated before convergence is reached. To permit such flexibility, we propose a feasible cooperation based MPC algorithm. All cooperative iterates in this algorithm are feasible and the resulting MPC controller is closed-loop stable. Two examples comparing the performance of optimal and sub-optimal MPC controllers are presented.