Attacking Learning-based Models in Smart Grids: Current Challenges and New Frontiers

Learning-based components applied to a plethora of use cases within smart grids are already a reality. These methods will undoubtedly play a key role in future energy systems.
This paper addresses challenges in the field of adversarial attacks against learning-based models in the context of smart grids. We identify unexplored areas and potential improvements in current methodologies by categorizing attacks, and assessing their ability to be reproduced. Our survey showed a noticeable resistance to distributing experimental code. Additionally, we propose the integration of explainable artificial intelligence techniques into adversarial models. We carry out an initial experiment to showcase the possible effects of this integration, offering fresh perspectives on the behavior and vulnerabilities of learning-based models within smart grids. Our initial findings provide a basis for further investigation into adversarial attacks, with a special focus on use cases that affect electrical substation security. Finally, we outline the next steps of our research in this critical area.