Physics Informed Reinforcement Learning for Power Grid Control using Augmented Random Search

Abstract

Wide adoption of deep reinforcement learning in energy system domain needs to overcome several challenges , including scalability, learning from limited samples, and high-dimensional continuous state and action spaces. In this paper, we integrated physics-based information from the generator operation state formula, also known as Swing Equation, into the reinforcement learning agent's neural network loss function, and applied an augmented random search agent to optimize the generator control under dynamic contingency. Simulation results demonstrated the reliability performance improvements in training speed, reward convergence, and future potentials in its transferability and scalability.