Scholarlink

<p>Modern unbalanced distribution systems are more vulnerable to False Data Injection (FDI) attacks, which simulate measurement data to destabilize system state estimation and stability. Critical branches—whose compromise can lead to serious operating issues—should be identified for enhanced resilience. This paper presents a tactic drawing on deep learning using Graph Neural Networks (GNNs) to identify critical branches for unbalanced distribution systems under FDI attacks. The technique replicates phase-dependent and spatial dependency relationships and is evaluated with IEEE 13-bus and 37-bus test feeders. Simulation results indicate the proposed GNN model correctly detects influential branches at high accuracy rates. For the 13-bus system, the scheme attained an accuracy level of 93%, and on the 37-bus feeder system, the accuracy was at 89%. The scheme accurately indicates phase-dependent weakness and scales reasonably to more complex systems. The proposed approach presents a robust solution for real-time criticality assessment of power distribution systems exposed to cyber-attacks. It serves as a foundation for more robust distribution systems. Future work will expand the scheme to dynamic conditions and implement it in real-time control systems.</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

Deep learning-based critical branch identification in unbalanced distribution systems under false data injection attacks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月6日

查看全文
  • Dangli Wang

摘要

Modern unbalanced distribution systems are more vulnerable to False Data Injection (FDI) attacks, which simulate measurement data to destabilize system state estimation and stability. Critical branches—whose compromise can lead to serious operating issues—should be identified for enhanced resilience. This paper presents a tactic drawing on deep learning using Graph Neural Networks (GNNs) to identify critical branches for unbalanced distribution systems under FDI attacks. The technique replicates phase-dependent and spatial dependency relationships and is evaluated with IEEE 13-bus and 37-bus test feeders. Simulation results indicate the proposed GNN model correctly detects influential branches at high accuracy rates. For the 13-bus system, the scheme attained an accuracy level of 93%, and on the 37-bus feeder system, the accuracy was at 89%. The scheme accurately indicates phase-dependent weakness and scales reasonably to more complex systems. The proposed approach presents a robust solution for real-time criticality assessment of power distribution systems exposed to cyber-attacks. It serves as a foundation for more robust distribution systems. Future work will expand the scheme to dynamic conditions and implement it in real-time control systems.