Model-Free Decentralized Reinforcement Learning Control of Distributed Energy Resources
Sayak Mukherjee, He Bai, Aranya Chakrabortty · 2020
This paper presents a model-free decentralized control on distributed energy resources (DERs) to improve the dynamic performance of the power grid. The optimal control algorithm has been developed using the ideas from reinforcement learning (RL). The decentralized control design does not need the model information of the DERs, and the optimal control gains can be computed using the measurements of the DER states and its terminal voltage variables. The DER terminal voltages act as interaction variables with the rest of the grid, and the algorithm computes an optimal gain taking these interaction variables into consideration. Assuming sufficient coupling of DERs to the synchronous machine states, the controller improves the dynamics of DER states, which in turn enhances the oscillation characteristics of the power grid. We demonstrate various details of the learning design and the performance of the controller by simulations on the IEEE benchmark 68-bus model with two wind farms.