Demand Response Vulnerable to Adversarial Price Forecast Attacks

A new study explores the vulnerability of industrial demand response systems to cyberattacks involving adversarial data modifications. Specifically, the research focuses on how manipulated electricity price forecasts can negatively impact the decision-making processes in energy-intensive production. Adversarial attacks, which are crafted data manipulations, aim to degrade the outcomes of prediction or decision-making algorithms. The study designed adversarial attacks to distort electricity price forecasts, which were then used to solve scheduling optimization problems for industrial processes. It found that these attacks can indeed diminish the profits gained from demand response. However, when the perturbations are subtle enough to be difficult for human users to detect, demand response still retains approximately 90% of its financial benefits compared to steady-state operations. A key finding is that the impact of these attacks depends more on the *orientation* of the adversarial perturbations rather than just their magnitude, highlighting the need for attack analyses to consider the sensitivities of scheduling optimization models.