Scholarlink

<p>This paper tackles the challenges of generating and managing power grid anomalies caused by operational variability and unexpected disruptions. Traditional models struggle with adaptability across domains, especially without labeled data or when operational conditions deviate from historical trends. To improve grid stability and reliability, this research introduces key innovations for anomaly generation and simulation under diverse conditions. First, a risk-based methodology is introduced to identify critical transmission lines, integrating a hidden failure model and system-level risk assessments. This method uniquely combines the probability of line failures with their system-wide impact, accurately identifying critical lines during cascading failures. Second, the augmented domain-adversarial network model is proposed, utilizing domain-adversarial learning to address power flow forecasting challenges across varying operational environments. By leveraging graph neural networks for realistic power flow simulations and graph attention networks for feature transformation, the model significantly improves prediction accuracy in scenarios with limited labeled data. Lastly, the integration of the maximum mean discrepancy criterion within a reproducing kernel Hilbert space minimizes feature distribution discrepancies between domains, enhancing the model’s adaptability and resilience to dynamic changes in power systems. This research facilitates the generation and simulation of power grid anomalies, enabling timely detection of potential issues. By creating realistic anomalous events, grid operators can better test and refine detection systems, improving fault identification and response in complex environments. This approach strengthens grid stability and reliability, supporting the development of more resilient power 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

The generation of power grid anomalous events based on domain-adversarial neural networks

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Yuyin Qiu,
  • Jianguo Qian,
  • Jiachuan You,
  • Min Lu,
  • Xiaonuo Zhang,
  • Jing Zhang

摘要

This paper tackles the challenges of generating and managing power grid anomalies caused by operational variability and unexpected disruptions. Traditional models struggle with adaptability across domains, especially without labeled data or when operational conditions deviate from historical trends. To improve grid stability and reliability, this research introduces key innovations for anomaly generation and simulation under diverse conditions. First, a risk-based methodology is introduced to identify critical transmission lines, integrating a hidden failure model and system-level risk assessments. This method uniquely combines the probability of line failures with their system-wide impact, accurately identifying critical lines during cascading failures. Second, the augmented domain-adversarial network model is proposed, utilizing domain-adversarial learning to address power flow forecasting challenges across varying operational environments. By leveraging graph neural networks for realistic power flow simulations and graph attention networks for feature transformation, the model significantly improves prediction accuracy in scenarios with limited labeled data. Lastly, the integration of the maximum mean discrepancy criterion within a reproducing kernel Hilbert space minimizes feature distribution discrepancies between domains, enhancing the model’s adaptability and resilience to dynamic changes in power systems. This research facilitates the generation and simulation of power grid anomalies, enabling timely detection of potential issues. By creating realistic anomalous events, grid operators can better test and refine detection systems, improving fault identification and response in complex environments. This approach strengthens grid stability and reliability, supporting the development of more resilient power systems.