Scholarlink

Real-time monitoring and fault prediction for equipment are crucial for stable grid operation, especially as global warming intensifies extreme weather impacts. Traditional methods like XGBoost and Random Forest struggle with complex feature engineering and fail to capture comprehensive spatial correlations, hindering accurate grid health assessments. This study introduces an innovative deep learning architecture combining Graph Neural Networks (GNN) and Long Short-Term Memory (LSTM) networks to address these limitations. By integrating distribution network topology, our model captures multidimensional space-time correlation between lines, enhancing global situational awareness in both time and space. Specifically, it employs multivariate temporal analysis and graph convolution modules to fuse heterogeneous data, including meteorological and sensor information, ensuring reliable power supply. Experimental results demonstrate significant improvements over traditional algorithms, with a 26.6% increase in precision and a 24.2% increase in recall. This approach effectively resolves issues of local perceptual bias and lack of dynamic correlation in complex grid environments through end-to-end learning.

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

Channel Independent GNN-LSTM Model for Efficient Power Outage Prediction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jinzhi Guo,
  • Zhengmao Tan,
  • Nan Liu,
  • Yinggang Sun,
  • Jinbo Yang

摘要

Real-time monitoring and fault prediction for equipment are crucial for stable grid operation, especially as global warming intensifies extreme weather impacts. Traditional methods like XGBoost and Random Forest struggle with complex feature engineering and fail to capture comprehensive spatial correlations, hindering accurate grid health assessments. This study introduces an innovative deep learning architecture combining Graph Neural Networks (GNN) and Long Short-Term Memory (LSTM) networks to address these limitations. By integrating distribution network topology, our model captures multidimensional space-time correlation between lines, enhancing global situational awareness in both time and space. Specifically, it employs multivariate temporal analysis and graph convolution modules to fuse heterogeneous data, including meteorological and sensor information, ensuring reliable power supply. Experimental results demonstrate significant improvements over traditional algorithms, with a 26.6% increase in precision and a 24.2% increase in recall. This approach effectively resolves issues of local perceptual bias and lack of dynamic correlation in complex grid environments through end-to-end learning.