Coalitional model predictive control for systems of systems

Abstract

An aspect so far rarely contemplated in distributed control problems is the explicit consideration of individual (local) interests of the components of a complex system. Indeed, the focus of the majority of the literature about distributed control has been the overall system performance. While on one hand this permitted to address fundamental properties of centralized control, such as system-wide optimality and stability, one the other hand it implied assuming unrestricted cooperation across local controllers. However, when dealing with multi-agent systems with a strong heterogeneous character, cooperation between the agents cannot be taken for granted (due to, for example, logistics, market competition), and selfish interests may not be neglected. Another critical point that must be kept into consideration is the diversity characterizing systems of systems (SoS), yielding very complex interactions between the agents involved (one example of such system is the smart grid). In order to tackle such inherent aspects of SoS, the research presented in this thesis has been concerned with the development of a novel framework, the coalitional control, that extends the scope of advanced control methods (in particular MPC) by drawing concepts from cooperative game theory that are suited for the inherent heterogeneity of SoS, providing as well an economical interpretation useful to explicitly take into account local selfish interests. Thus, coalitional control aims at governing the association/dissociation dynamics of the agents controlling the system, according to the expected benefits of their possible cooperation. From a control theoretical perspective, this framework is founded on the theory of switched systems and variable structure/topology networked systems, topics that are recently experiencing a renewed interest within the community. The main concepts and challenges in coalitional control, and the links with cooperative network game theory are presented in this document, tracing a path from model partitioning to the control schemes whose principles delineate the idea of coalitional control. This thesis focuses on two basic architectures: (i) a hierarchically supervised evolution of the coalitional structure, and (ii) a protocol for autonomous negotiation between the agents, with specific mechanisms for benefit redistribution, leading to the emergence of cooperating clusters.