Distributed satisficing MPC

Abstract

Abstract : To obtain a Pareto-optimal solution, the classical cooperative MPC implementsa categorical altruism imposed by a fixed global cost sharedby all the local controllers. Instead, this thesis implements a situationalaltruism where a global cost, neither imposed nor fixed, emerges fromconvex local costs and local specifications. The satisficing controllersemploy a distributed algorithm to find a solution that lies in a convexregion that is satisfactory and sufficient for all controllers (satisficing= satisfy + suffice), while optimizing in the direction of the analyticcenter of such a region. The system is modeled as being a network oflinear subsystems, coupled by their inputs, and the algorithm uses adistributed interior-point method to avoid fixed points when the constraintsare also coupled. The optimal solution of the satisficing MPC,besides Pareto-optimal, gives more importance to the controllers witha worst performance at the moment. Situational altruism permits amore balanced division of resources, avoiding the exploitation of onecontroller by the others. The satisficing MPC is shown to be stabilizingeven if suboptimal, provided that it is satisficing. To this end,stabilizing constraints are added to the basic formulation.<br>