Towards Understanding the Challenges of Applying Reinforcement Learning to the Power Grid

Matthew Kyle Schlegel, Martha White, Matthew E. Taylor, Mostafa Farrokhabadi

Keywords: Power Systems, Reinforcement Learning, HiTL, Real Systems

TL;DR: This paper outlines many of the challenges that reinforcement learning faces in power system control with respect to Dulac-Arnold et al. (2021) and introduces two new challenges that arise from power system control.

Abstract: This paper presents a high-level overview of the challenges facing the application of reinforcement learning (RL) to power system control. As power systems evolve with increasing complexity and uncertainty, RL offers promising approaches for adaptive control. Building on the framework of Dulac-Arnold et al. (2021), we explore how power systems embody and expand on these challenges. We also introduce two additional challenges that araise from the power systems domain: (1) learning policies over multi-timescale action spaces, and (2) fostering effective collaboration between RL agents and human operators. By outlining these challenges for the power systems domain, this work aims to enable future research and collaborative efforts between the power systems and RL communities.

Submission Number: 13