Abstract—This paper introduces a Python framework built upon open-source software for electric grids (OpenDSS) and deep learning (Open AI) for the study of further applications of reinforcement learning on distribution grid networks. This paper describes the framework and applies it to a 13-bus, grid-tied microgrid system, training and using a reinforcement learning agent to optimally control capacitor banks to maintain system voltage under changing loads. The performance of the agent is then compared to known optimal control, as well as to the performance of a capacitor controller built-in to OpenDSS, and a supervised learning-trained neural network.
Index Terms—reinforcement learning, electric grid, volt/VAR regulation, voltage control, distribution grid, microgrid
Full paper here.
Github code here.
Paper has been submitted and accepted to the 2021 IEEE PES General Meeting, Washington, D.C.
Funding graciously provided by the Stanford Bits & Watts Program.