Scholarlink

<p>The reliability of urban power distribution networks is increasingly threatened by rising demand and environmental stressors. This study proposes a novel, data-centric framework for Seasonally Adaptive maintenance, analyzing fault patterns in a tropical 33&#xa0;kV distribution network in Lagos, Nigeria. Utilizing a dataset comprising 5955 fault transaction entries and 7058 specific fault occurrences collected over seven years (2017–2024), statistical diagnostics were integrated with advanced ensemble learning to characterize failure modes. The analysis revealed that fault frequency peaks during the wet season, yet dry-season faults exhibit higher severity scores. Three recurring fault types, Line Trips, Jumper Cuts, and Vegetation Interference, were found to account for 80% of disruptions. By applying Association Rule Mining, actionable if–then rules linking environmental triggers to mechanical failures were extracted. To predict high-severity incidents, tuned Gradient Boosting models were developed and benchmarked. The optimized LightGBM and XGBoost models achieved superior performance, recording the Receiver Operating Characteristic—Area Under the Curve (ROC-AUC) scores 0.88, statistically outperforming the standard Gradient Boosting baseline (<i>p</i> &lt; 0.05), while achieving comparable performance to other ensemble variants. Feature importance analysis confirmed that seasonality is the primary driver of fault severity, followed closely by jumper and cable vulnerabilities. These findings validate a shift from reactive repairs to a proactive strategy where maintenance resources are dynamically allocated based on seasonal risk profiles. This work provides utility operators with a validated, quantitative toolset to reduce the Mean Time To Restore (MTTR) and enhance grid resilience against climatic variability.</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

Seasonally adaptive fault forecasting for proactive and resilient maintenance in urban power distribution networks

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Xueli Yang,
  • Yujin Xiang,
  • Qin Wang

摘要

The reliability of urban power distribution networks is increasingly threatened by rising demand and environmental stressors. This study proposes a novel, data-centric framework for Seasonally Adaptive maintenance, analyzing fault patterns in a tropical 33 kV distribution network in Lagos, Nigeria. Utilizing a dataset comprising 5955 fault transaction entries and 7058 specific fault occurrences collected over seven years (2017–2024), statistical diagnostics were integrated with advanced ensemble learning to characterize failure modes. The analysis revealed that fault frequency peaks during the wet season, yet dry-season faults exhibit higher severity scores. Three recurring fault types, Line Trips, Jumper Cuts, and Vegetation Interference, were found to account for 80% of disruptions. By applying Association Rule Mining, actionable if–then rules linking environmental triggers to mechanical failures were extracted. To predict high-severity incidents, tuned Gradient Boosting models were developed and benchmarked. The optimized LightGBM and XGBoost models achieved superior performance, recording the Receiver Operating Characteristic—Area Under the Curve (ROC-AUC) scores 0.88, statistically outperforming the standard Gradient Boosting baseline (p < 0.05), while achieving comparable performance to other ensemble variants. Feature importance analysis confirmed that seasonality is the primary driver of fault severity, followed closely by jumper and cable vulnerabilities. These findings validate a shift from reactive repairs to a proactive strategy where maintenance resources are dynamically allocated based on seasonal risk profiles. This work provides utility operators with a validated, quantitative toolset to reduce the Mean Time To Restore (MTTR) and enhance grid resilience against climatic variability.