Investigation on Transient Stability Enhancement of Multi-VSG System Incorporating Resistive SFCLs Based on Deep Reinforcement Learning
Lei Chen, Jingguang Tang, Xuefeng Qiao, Hongkun Chen, Jiahui Zhu, Yuqi Jiang, Zekai Zhao, Ruize Hu, Xinyi Deng
Abstract
Virtual synchronous generator (VSG) strategy is an effective means for renewable energy sources to be connected to power grids. Meanwhile, a greater focus should be on the multi-VSG system's (MVS) transient stability issue. Regarding the MVS with resistive superconducting fault current limiters (R-SFCLs), this article proposes a control method based on deep reinforcement learning (DRL) agent for increasing the transient stability. Firstly, the theoretical model of the MVS incorporating R-SFCLs is established, and R-SFCLs are used to limit the fault current in VSGs and keep the fault ride-through (FRT) operation. Then, the working mechanism of the proposed method is elaborated, by designing the Markov decision process (MDP) model of transient control, and applying the improved softmax deep deterministic policy gradients (SD2) algorithm to train the DRL agent. The advanced time series feature extraction network (TSFEN) based on the convolutional neural network (CNN) and gate recurrent unit (GRU) is suggested to enhance the actor network and critic network of the SD2. A detailed simulation model is created using MATLAB, and a comparison with the traditional VSG control, only R-SFCL, deep deterministic policy gradients (DDPG) algorithm, and improved power loop scheme is conducted. From multiple cases, the proposed approach can satisfactorily boost the transient stability of the MVS, and the generalization ability of the DRL agent under different untrained scenarios is validated. The proposed method's validity and suitability are well-confirmed.