Scholarlink

To thoroughly investigate the multi-physics field response mechanisms of 220 kV cable joints with scratch defects under various operating conditions, this study first establishes an integrated modeling system of the cable and joint. A 2D axisymmetric electro-thermal coupled simulation model is developed on the COMSOL platform to systematically analyze the coupled effects of different scratch morphologies on field distributions. Based on this, considering the intuitiveness and accessibility of temperature responses for defect identification, a Stacking-XGBoost temperature field prediction model is proposed, taking current load, convective heat transfer coefficient, and defect type as inputs to rapidly and accurately estimate temperature distribution along the scratch path. The model achieves R2 values of 0.9990 and 0.9983 on the training and testing sets, respectively, with MAE and R2 significantly outperforming the traditional GBR model, and a single prediction time of only 0.42 s. These findings provide a theoretical foundation and an efficient technical approach for quantitative assessment of cable joint scratch defects, digital twin modeling, and online condition monitoring.

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 Stacking-XGBoost Integrated Model for Rapid Temperature Field Prediction of Cable Joints with Scratch Defects

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fanbo Wei,
  • Bowen Luo,
  • Guoyuan Lu,
  • Jiahui Mei,
  • Peng Zhou,
  • Guanyan Chen,
  • Xiaomeng Shi,
  • Haibo Feng,
  • Zhengshen Zhu

摘要

To thoroughly investigate the multi-physics field response mechanisms of 220 kV cable joints with scratch defects under various operating conditions, this study first establishes an integrated modeling system of the cable and joint. A 2D axisymmetric electro-thermal coupled simulation model is developed on the COMSOL platform to systematically analyze the coupled effects of different scratch morphologies on field distributions. Based on this, considering the intuitiveness and accessibility of temperature responses for defect identification, a Stacking-XGBoost temperature field prediction model is proposed, taking current load, convective heat transfer coefficient, and defect type as inputs to rapidly and accurately estimate temperature distribution along the scratch path. The model achieves R2 values of 0.9990 and 0.9983 on the training and testing sets, respectively, with MAE and R2 significantly outperforming the traditional GBR model, and a single prediction time of only 0.42 s. These findings provide a theoretical foundation and an efficient technical approach for quantitative assessment of cable joint scratch defects, digital twin modeling, and online condition monitoring.