Deep Reinforcement Learning-Based Operation of Distribution Systems Using Surrogate Model

The rapid growth of distributed energy resources (DERs) causes many difficulties in the real-time operation of power systems. Model-based optimization methods may not be effective because of the slow response to the system uncertainty. Therefore, this study proposes a deep reinforcement learning (DRL)-based optimization model for the operation of distribution systems. The proposed method consists of two main models (i) deep neural network (DNN)-based surrogate model and (ii) deep Q learning model. First, the surrogate model is trained to map the input/output from the simulation environment. After the training process, surrogate model can replace the simulation model and therefore accelerate the learning process. Then, deep Q learning-based optimization model determines the set-points for all generators to ensure optimal power flow in the entire distribution network. Finally, an IEEE 33-bus radial distribution test system is used to evaluate the effectiveness of the proposed model.

Automated summary

This study proposes a deep reinforcement learning-based optimization model for the operation of distribution systems. The model consists of a surrogate model and an optimization model, where the surrogate model is trained to accelerate the learning process and the optimization model utilizes deep Q learning to determine the set-points for generators, ensuring optimal power flow in the distribution network.