Cyber-Physical Systems Design: Electricity Markets and Network Security

摘要

This thesis presents Cyber-Physical Systems Design (CPS Design). Design of CPS is challenging and requires interdisciplinary studies of engineering and economics because of the distinguishing features of CPS: strategic (self profit-maximizing) decision makers, complex physical constraints, and large-scale networked systems. We study these features by focusing on designing markets with complex constraints including both policy and physical constraints, and decomposing large-scale CPS within the context of electricity markets and network security.;We first study market design for implementation of complex electricity policy targets, i.e. sustainability, reliability, and price efficiency, by efficient design of spot, carbon, and capacity markets that correct the deficiencies of the current electricity markets; this design does not take into account the network constraints due to the Kirchhoff's laws. To address this problem, we develop a framework based on the design of efficient auctions with constraints. Our market design sheds light on major debates in electricity policy including capacity-and-energy vs energy-only markets, carbon market vs carbon tax, and use of price or offer caps.;Second, we add network constraints due to Kirchhoff's laws of current and voltage, which are unique to electricity networks, to the design of electricity spot markets with complex physical constraints. To address this problem, we develop a framework for the design of networked markets based on the ideas from local public goods.;Finally, we study the design of defense policies for large-scale network security. Our approach is to design approximately optimal defense policies that are computable. We develop a framework based on the notion of influence graph, which captures the connectivity of the security states of the system elements, to decompose the system into subsystems. We then design approximately optimal defense policies for each sub-system. We consider non-Bayesian uncertainty and even though we do not model the attacker as a strategic decision maker, we compensate (in part) for the lack of this feature by adopting a minmax performance criterion.