Data-driven stochastic optimization for cyber-physical system risk management : smart power grids with renewable energy

Abstract

Unit commitment decision is the fundamental operational decision for smart grids. Yet, the inherent unpredictable nature on both supply and demand sides, and our limited prior information of underlying input models lead to both stochastic and input uncertainty. To provide reliable and cost-efficient operational unit commitment decisions, we propose a new data-driven stochastic unit commitment model to hedge against the input and stochastic uncertainties simultaneously. Built on that, we develop a novel parallel optimization-based framework that further controls the finite sampling error caused by sample average approximation.

The cyber-physical nature of smart grids means that the uncertainty from the cyber side also has an impact on the smart grid real-time operation. Thus, to ensure a reliable system operation and power production, the impact of cyber-attacks should be considered. We focus on Distributed Denial of Service (DDoS) attacks, which overwhelm the communication network of the smart grid by jamming data and propose a simulation-based stochastic unit commitment model to hedge against both stochastic uncertainty of wind power and DDoS threats. The proposed unit commitment model can bring reliable and cost-efficient decisions when smart grids are under DDoS attacks.

Specifically, simulation is often used when making the strategic decisions. Yet, when we use input model estimates to drive simulation, the input uncertainty can lead to incorrect assessment of system performances and further cause the non-optimal design selection. To address this issue, we develop a budget allocation strategy that efficiently quantifies the impact of the input while controlling the impact of simulation estimation uncertainty. Meanwhile, when we make strategic decisions, we should consider its impact on the operational cost as well, which means both the investment cost and the subsequent expected operational cost should be factored in. Yet, the current optimization model used in the literature is over-simplified. We then propose an innovative simulation-based optimization framework that is capable of integrating strategic and operational decisions. A metamodel-assisted two-stage optimization framework is developed. It can efficiently use the computational resource to iteratively search for the optimal first-stage strategic decisions and second-stage operational decisions.

In this thesis, we focus on risk management for smart grids with renewable energy. Smart grids are power grids that integrate renewable energies and advanced modern technologies, which are interactive, stochastic, multi-locational, and cyber-physical in nature. There exist many sources of uncertainty in smart grids, coming from both cyber and physical sides. To improve the reliability, efficiency and resilience of smart grids, we develop a simulation and stochastic optimization framework that hedges against various sources of risk and delivers coherent strategic and operational decisions.