Construction and Application of Fault Prediction Model for Distribution Network Based on LightGBM

The frequent occurrence of distribution network failures has had a serious impact on the economy, commerce, and people’s livelihood in Guizhou region. Therefore, building accurate and efficient fault prediction models has become an urgent need to ensure power supply stability and improve operation and maintenance efficiency. This paper constructed multiple machine learning models (including Random Forest - RF, eXtreme Gradient Boosting - XGBoost, Light Gradient Boost Machine - LGBM) for fault prediction in distribution networks. On the basis of comparing and analyzing the performance of various models, the LGBM model shows significant advantages and better prediction performance. In specific applications, the prediction model based on LGBM integrates more than 19 key factors such as meteorological information, operating hours, regional coding, holiday electricity consumption characteristics, equipment aging, etc. After optimization training, it outperforms other models in accuracy, recall, F1 value, AUC and other indicators, significantly improving its ability to predict distribution network faults under complex multi factor influences. In contrast, the random forest model only achieved an accuracy of 0.6744 and a lower accuracy (0.4052), indicating its limited ability to model nonlinear feature interactions.