Distributed model predictive control as a game with coupled constraints

This paper formally analyses the benefits of cooperation in distributed MPC. The algorithm guarantees feasibility for dynamic subsystems coupled through the constraints, while cooperative behaviour is promoted by local agents considering the objectives of others. Under mild assumptions, state convergence is guaranteed to state limit sets. By relating game-theoretical concepts to the algorithm, it is shown that the set of Nash solutions does not grow with increasing cooperation. Subsequently, the set of possible state limit sets also does not grow. Examples show that an improvement in the convergence outcome can be seen with only partial cooperation.