Active Distribution Network Scheduling Based on Safe Deep Reinforcement Learning
ID：2
Submission ID：11 View Protection：ATTENDEE
Updated Time：2025-07-30 20:15:11 Hits：242
Poster Presentation
Start Time：2025-11-09 09:00 （Asia/Shanghai）
Duration：1min
Session：[P] Poster presentation » [P3] 3.Power system and automation
No files
Abstract
Against the backdrop of high proportion penetration of new energy, the difficulty of scheduling optimization for active distribution networks is gradually increasing. This paper proposes a safe deep reinforcement learning framework integrating a data-driven power flow model to achieve the scheduling optimization of new energy. To address the challenge of active voltage control, a safe deep reinforcement learning strategy combined with a data-driven power flow model is designed, which maps reactive power to voltage amplitude through the Q2V strategy. Simulation results on the modified IEEE 33-bus system show that the optimization effect of this framework is significantly improved compared with the traditional Q strategy and V strategy. It achieves 46.2% and 64.9% reduction in line loss respectively, while strictly controlling the node voltage deviation within the range of ±5%.
Keywords
Active distribution network,power flow model,safe deep reinforcement learning,scheduling optimization
Speaker
Comment submit