Power System Transient Stability Control with Grid-Connected Virtual Synchronous Machines: A Deep Reinforcement Learning Approach

Résumé fourni par la source

The integration of virtual synchronous machine (VSM) constitutes both a promising approach and a new challenge for maintaining power system stability, especially given the increasing penetration of renewable energy. This study addresses the transient stability control issues concerning the coordination of synchronous generators and VSMs with a novel control scheme driven by deep reinforcement learning (DRL). Initially, the system model involving SGs and VSMs are developed, highlighting the divergent control strategies and objectives of the grid-following and grid-forming VSMs. Afterwards, a Partially Observable Markov Decision Process (POMDP) is constructed to model the transient stability control problem into a reinforcement learning framework. The hierarchical attention long short-term memory (HALSTM) network in combination with the soft actor critic (SAC) algorithm is utilized to capture the time-series feature and enhance the robust exploration. Subsequently, comprehensive experiments have demonstrated the effectiveness of the HALSTM-SAC framework through comparisons with the droop control, indicating evident improvements in transient stability with the DRL framework.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.