Adaptive Emergency Control of Power Systems Based on Deep Belief Network

Abstract

Emergency control is an essential means to help system maintain synchronism after fault clearance. Traditional “offline calculation, online matching” scheme faces significant challenges on adaptiveness and robustness problems. To address these challenges, this paper proposes a novel closed-loop framework of transient stability prediction (TSP) and emergency control based on Deep Belief Network (DBN). First, a hierarchical real-time anti-jitter TSP method using sliding time windows is adopted, which takes into account accuracy and rapidity at the same time. Next, a sensitivity regression model is established to mine the implicit relationship between power angles and sensitivity. When impending instability of the system is foreseen, optimal emergency control strategy can be determined in time. Lastly, responses after emergency control are fed back to the TSP model. If prediction result is still unstable, an additional control strategy will be implemented. Comprehensive numerical case studies are conducted on New England IEEE 39-bus system and Northeast Power Coordinated Council (NPCC) 140-bus system. Results show the proposed method can detect instability of system as soon as possible and assist in maintaining reliable system synchronism.

Keywords

• Deep learning
• emergency control
• power system
• sensitivity
• transient stability prediction