安军, 黎梓聪, 周毅博, 石岩, 毕建航. 基于改进双智能体D3QN的电网N−1新增风险防控策略[J]. 中国电机工程学报, 2025, 45(3): 858-869. DOI: 10.13334/j.0258-8013.pcsee.231282
引用本文: 安军, 黎梓聪, 周毅博, 石岩, 毕建航. 基于改进双智能体D3QN的电网N−1新增风险防控策略[J]. 中国电机工程学报, 2025, 45(3): 858-869. DOI: 10.13334/j.0258-8013.pcsee.231282
AN Jun, LI Zicong, ZHOU Yibo, SHI Yan, BI Jianhang. N−1 New Risk Prevention and Control Strategy for Power Grid Based on Improved Double Agent D3QN Algorithm[J]. Proceedings of the CSEE, 2025, 45(3): 858-869. DOI: 10.13334/j.0258-8013.pcsee.231282
Citation: AN Jun, LI Zicong, ZHOU Yibo, SHI Yan, BI Jianhang. N−1 New Risk Prevention and Control Strategy for Power Grid Based on Improved Double Agent D3QN Algorithm[J]. Proceedings of the CSEE, 2025, 45(3): 858-869. DOI: 10.13334/j.0258-8013.pcsee.231282

基于改进双智能体D3QN的电网N−1新增风险防控策略

N−1 New Risk Prevention and Control Strategy for Power Grid Based on Improved Double Agent D3QN Algorithm

  • 摘要: 城市电网在发生N-1故障后,极可能新增运行风险,导致N-1-1时出现大面积停电事故。为管控城市电网N-1后运行风险,该文提出一种改进双智能体竞争双深度Q网络(dueling double deep Q network,D3QN)的城市电网N-1风险管控转供策略。根据风险管控原则,提出一种无需额外历史数据、考虑备自投装置、单供变电站风险和单供负荷母线风险的N-1场景指标;建立计及动作次序、指标间关系的负荷转供三阶段求解模型。以含预动作-变化探索值选择策略的改进双智能体D3QN方法,将负荷转供分为多个子转供环节学习,使转供思路清晰化,对动作空间进行降维,提高训练寻优效果,得到管控N-1风险的负荷转供策略。通过城市电网多场景算例分析,验证该文模型和方法的有效性。

     

    Abstract: After the occurrence of N-1 faults in urban power grids, there is a high possibility of new operational risks, leading to large-scale power outages during N-1-1. To control the operational risks of urban power grids after N-1, the paper proposes an N-1 new risk control transfer strategy based on improved Dueling Double Deep Q Network (D3QN) method with double agent for urban power grids. Based on the principle of risk management and control, an N-1 scenario indicator that considers the risk of single supply substations is proposed, as well as single supply load buses and backup automatic switching device. And a three-stage solution model for load transfer that takes into account the sequence of actions and the relationship between indicators is established. The article proposes a D3QN method with double agent and pre-action changing ε selection strategy, which divides the load transfer into multiple sub transfer stages for learning, clarifies the transfer idea, reduces the dimensionality of the action space, improves the training optimization effect, and obtains a load transfer strategy that controls N-1 risks. The effectiveness of the model and method proposed in this paper has been verified through the analysis of multiple scenarios in urban power grids.

     

/

返回文章
返回