Adversarial attacks in demand-side electricity markets

Detecting and characterizing malicious operations in electricity markets presents substantial challenges due to the innate complexity of power networks. This paper explores adversarial attacks in electricity markets, i.e., we consider the scenario in which malicious market participants make undetectable changes to the input data of other participants, resulting in significant negative impacts on their operations. We propose a novel methodology to generate adversarial attacks, which leverages linear optimization theory and statistical learning to generate suboptimal, feasible solutions or super-optimal, infeasible solutions that convincingly resemble optimal solutions. The methodology is general and applicable for generating adversarial attacks on optimization-based decision systems. We use it to analyze the potential economic losses, policy making, and social consequences resulting from compromised demand-side operations. In particular, we focus on scenarios where certain demand side-players, i.e., microgrids, or prosumers in a general sense, engage in malicious activities by targeting other microgrids and the independent system operator. We model the electricity market dynamics using an equilibrium problem with equilibrium constraints (EPEC). To make the model easier to solve, the EPEC is reformulated using a combination of the Fisher–Burmeister function and strong duality theorem. Results show that microgrids can artificially decrease its consumption by up to 12% without triggering alarms. Additionally, this amount can be increased by an additional 14% to 22%, contingent on the strategy employed to monitor the observable variables. We also study the implications of the increasing utilization of solar generation, showing its potential to mitigate the impact of adversarial attacks while also introducing new risks to the system.