Renewable energy sources have brought more uncertainties to the operation of energy systems. Two challenges emerge. Firstly, we generally do not know the exact model and parameters of these energy resources. Secondly, robustly incorporating the uncertainties in operations is intractable using traditional control methods. Machine learning, especially reinforcement learning techniques, can potentially overcome these challenges by interacting with the environment to find good control strategies.
Despite its potential, reinforcement learning does not readily apply to critical physical systems with hard constraints. My research focuses on closing these gaps to allow for the robust control of inverter-interfaced devices in energy systems. I will analyze problems with non-Gaussian noises and nonlinear transition models, and design reinforcement learning algorithms for continuous systems. The target applications would be the frequency and voltage regulation of both transmission networks and microgrids.
Advisor: Daniel Kirschen – Electrical & Computer Engineering