Scholarlink

<p>This paper proposes a robust topology identification framework for distribution networks based on counterfactual samples and generative adversarial networks (GAN), specifically addressing challenges of data missingness, imbalanced data distribution, and dynamic topological complexity. By integrating a multi-level graph attention mechanism with a feature pyramid architecture as the base feature extractor, along with counterfactual sample generation and GAN-based data augmentation, this study establishes an efficient model to address node data deficiency and imbalanced data distributions in distribution network topology identification. Experimental results demonstrate that the proposed model achieves an identification precision of 97.12% on the IEEE 69-bus system and maintains precision of 78.12% even with 40% node data missing. Compared to the current existing models, our framework significantly enhances identification performance for low-frequency topologies and dynamically complex scenarios under imperfect data conditions through robustness improvement strategies. The system's modular and scalable architectural design enables optimized deployment in high-performance computing environments, thereby achieving real-time topology identification for large-scale distribution networks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A robust topology identification method of distribution network based on counterfactual samples and generative adversarial networks

  • 期刊论文

  • 发表日期：

    2025年9月23日

查看全文
  • Yuting Jiang,
  • Tie Chen,
  • Jiaxin Yuan,
  • Mingrui Zhao,
  • Yue Liu

摘要

This paper proposes a robust topology identification framework for distribution networks based on counterfactual samples and generative adversarial networks (GAN), specifically addressing challenges of data missingness, imbalanced data distribution, and dynamic topological complexity. By integrating a multi-level graph attention mechanism with a feature pyramid architecture as the base feature extractor, along with counterfactual sample generation and GAN-based data augmentation, this study establishes an efficient model to address node data deficiency and imbalanced data distributions in distribution network topology identification. Experimental results demonstrate that the proposed model achieves an identification precision of 97.12% on the IEEE 69-bus system and maintains precision of 78.12% even with 40% node data missing. Compared to the current existing models, our framework significantly enhances identification performance for low-frequency topologies and dynamically complex scenarios under imperfect data conditions through robustness improvement strategies. The system's modular and scalable architectural design enables optimized deployment in high-performance computing environments, thereby achieving real-time topology identification for large-scale distribution networks.