Voltage Regulation Based on Deep Reinforcement Learning Algorithm in Distribution Network with Energy Storage System

Authors:

Abstract
The energy storage system connected at the end of distribution network is widely considered to be an effective means of achieving voltage regulation and ensuring the more stable voltage operation level. This paper proposes a voltage regulation method based on deep reinforcement learning to reduce the voltage fluctuation in the distribution network with battery energy storage system (BESS). The operation of battery energy storage system is modelled as a Markov Decision Making process. Then, by embedding a deep Q-network to obtain the optimal charge and discharge strategy, the calculation problem caused by too large state space could be efficaciously solved. Furthermore, the deep Q-network will be trained by replay strategy to achieve iterative convergence. And an intelligent BESS control method that provides auxiliary service for voltage regulation is ultimately obtained. Finally, the modified IEEE33 node system with BESS is analyzed by using Matpower in the framework of TensorFlow, and the effectiveness of the proposed method is proved.