Work

Decentralized Coordination Control for Dynamic Multiagent Systems

Year: 2015
Type: dissertation
Abstract: Engineered networks such as power grids, communication networks and logistics operations are becoming increasingly interdependent and comprised of intelligent components that are capable of autonomously processing and influencing their local environment, as well as interacting with other such components. These emerging capabilities are poised to enable unprecedented efficiency gains across previously disparate application domains. Fully harnessing this potential is contingent in part on our ability to effect coordinated local interactions among such components, ensuring their optimal collective behavior. In this thesis we formulate several Decentralized Coordination Control Problems (DCCPs) for networked multiagent systems involving either static or dynamic agents. We propose that Consensus Optimization (CO) methods constitute a viable foundation for the synthesis of coordination control strategies that address large classes of DCCPs. In order to understand both the potential and the limitations of such strategies, we develop a new framework for the convergence analysis of general CO schemes. In contrast with existing analytic approaches, our framework is based on powerful system-theoretic techniques enabling the study of DCCP scenarios in which the agent dynamics interact with those of the CO algorithm itself, thereby affecting its performance. Aside from fulfilling its intended purpose, our analytic viewpoint also leads to the relaxation of several standard assumptions imposed on numerical CO algorithms, and suggests a methodical approach to deriving conditions under which existing centralized optimization methods can be decentralized. In addition, we propose the Reduced Consensus Optimization (RCO) algorithm -- a streamlined variant of CO which is more innately suited to the DCCP context. Agents implementing RCO generally need not be aware of the network size and topology, and the overall processing and communication overhead associated with agent coordination may be substantially reduced relative to CO. On the basis of RCO, we propose a methodology for the design of decentralized content caching strategies in information-centric networks.
Language: English
Open Access status: green
OpenAlex ID: W2588923573
Cites: 166
Cited by: 5
Related to: 10