Prediction of lightning-induced faults in high voltage transmission lines using machine learning methods

Power transmission lines are exposed to severe faults and power outages due to lightning-induced electromagnetic pulses. Although traditional methods have achieved certain success in preventing lightning-related faults, their prediction performance remains limited due to complex meteorological variables and geographical factors. In this study, a hybrid model is proposed for predicting lightning-induced faults in power transmission lines by combining the Random Forest Algorithm and the XGBoost algorithm. The model was trained using 360,000 lightning records and 713 lightning-induced fault events collected in the Antalya region between 2015 and 2022. During the data preprocessing stage, techniques such as missing data handling, outlier analysis, and feature engineering were applied. The model's performance was evaluated using metrics such as accuracy, F1-score, ROC-AUC, and confusion matrix. The results showed that the hybrid model achieved a 93% accuracy rate, significantly outperforming traditional methods. Furthermore, the proposed model demonstrated reduced false positive and false negative rates and showed high success particularly in critical fault scenarios. The findings of this study suggest that hybrid machine learning models offer a powerful alternative for predicting lightning-induced faults in power transmission lines. The proposed model, which integrates comprehensive engineering parameters for prediction, is expected to make a novel contribution to the literature. Future studies are recommended to test the model in different geographical regions, incorporate additional meteorological variables, and adapt it for real-time applications.