On feasibility, stability and performance in distributed model predictive control

We present a stopping condition to the duality based distributed optimization algorithm presented in [1] when used in a distributed model predictive control (DMPC) context. To enable distributed implementation, the optimization problem has neither terminal constraints nor terminal cost that has become standard in model predictive control (MPC). The developed stopping condition guarantees a prespecified performance, stability, and feasibility with finite number of algorithm iterations. Feasibility is guaranteed using a novel adaptive constraint tightening approach that gives the same feasible set as when no constraint tightening is used. Stability and performance of the proposed DMPC controller without terminal cost or terminal constraints is shown based on a controllability parameter for the stage costs. To enable quantification of the control horizon necessary to ensure stability and the prespecified performance, we show how the controllability parameter can be computed by solving a mixed integer linear program (MILP)