Machine learning application for predicting 110 kV PTL failures based on PTL parameters

This study proposes the use of machine learning algorithms to predict 110 kV power line failures based on data on the parameters of the lines themselves. Five classifiers were tested as machine learning algorithms: support vector machine, logistic regression, random forest, gradient boosting LightGBM Classifier and CatBoostClassifier. For designed model a pipeline and a compositor of heterogeneous features were used to automate the process of data conversion and eliminate the possibility of data leakage. Data were prepared using hot coding method for categorical variables and standardization method for quantitative ones. The model was trained using the cross-validation method with stratified separation. Through the use of grid search and random parameter optimization techniques, the classifiers’ hyperparameters were changed. The prediction quality of the trained models was compared using the metrics ROC-AUC, AUC-PR, Accuracy, accuracy, recall and F-1 measure. The best results in predicting outages were achieved by the logistic regression model with the class weighting method to combat class imbalance, the ROC-AUC metric of which reached 0.84 on the test sample. Thus, this study confirms the possibility of using data on power line parameters to predict 110 kV power line failures.