- Repository URL:
- Cyberphysical Systems; Stochastic Control
thesis / dissertation description
In this thesis we address challenging issues that are faced in the operation of important cyber-physical systems of great current interest. The two particular systems that we address are communication networks and the smart grid. Both systems feature distributed agents making decisions in dynamic uncertain environments. In communication networks, nodes need to decide which packets to transmit, while in the power grid individual generators and loads need to decide how much to pro-duce or consume in a dynamic uncertain environment. The goal in both systems, which also holds for other cyber-physical systems, is to develop distributed policies that perform ef?ciently in uncertain dynamically changing environments. This thesis proposes an approach of employing duality theory on dynamic stochastic systems in such a way as to develop such distributed operating policies for cyber-physical systems. In the ?rst half of the thesis we examine communication networks. Many cyber-physical systems, e.g., sensor networks, mobile ad-hoc networks, or networked control systems, involve transmitting data over multiple-hops of a communication network. These networks can be unreliable, for example due to the unreliability of the wireless medium. However, real-time applications in cyber-physical systems often require that requisite amounts of data be delivered in a timely manner so that it can be utilized for safely controlling physical processes. Data packets may need to be delivered within their deadlines or at regular intervals without large gaps in packet deliveries when carrying sensor readings. How such packets with deadlines can be scheduled over networks is a major challenge for cyber-physical systems. We develop a framework for routing and scheduling such data packets in a multi-hop network. This framework employs duality theory in such a way that actions of nodes get decoupled, and results in ef?cient decentralized policies for routing and scheduling such multi-hop communication networks. A key feature of the scheduling policy derived in this work is that the scheduling decisions regarding packets can be made in a fully distributed fashion. A decision regarding the scheduling of an individual packet depend only on the age and location of the packet, and does not require sharing of the queue lengths at various nodes. We examine in more detail a network in which multiple clients stream video packets over shared wireless networks. We are able to derive simple policies of threshold type which maximize the combined QoE of the users. We turn to another important cyber-physical system of great current interest ? the emerging smarter grid for electrical power. We address some fundamental problems that arise when attempting to increase the utilization of renewable energy sources. A major challenge is that renewable energy sources are unpredictable in their availability. Utilizing them requires adaptation of demand to their uncertain availability. We address the problem faced by the system operator of coordinating sources of power and loads to balance stochastically time varying supply and demand while maximizing the total utilities of all agents in the system. We develop policies for the system operator that is charged with coordinating such distributed entities through a notion of price. We analyze some models for such systems and employ a combination of duality theory and analysis of stochastic dynamic systems to develop policies that maximize the total utility function of all the agents. We also address the issue of how the size of energy storage facilities should scale with respect to the stochastic behavior of renewables in order to mitigate the unreliability of renewable energy sources.