Classification of power quality events in the transmission grid: comparative evaluation of different machine learning models
In power quality event root-cause identification within transmission networks, distinguishing between ABC and ABCG short-circuit faults remains challenging due to their high similarity. Method: This study systematically evaluates multiple machine learning models for multi-class fault classification using real-world measurements from Turkey’s national transmission monitoring system. Time–frequency domain features are extracted and optimized via grid search for hyperparameter tuning; Cubic SVM and XGBoost are rigorously compared against baseline models. Results: Both Cubic SVM and XGBoost achieve significantly higher accuracy and markedly reduce ABC/ABCG misclassification compared to alternatives. Notably, this work presents the first empirical validation of Cubic SVM’s superiority in power quality event classification within a national-scale, operational transmission grid. Furthermore, both models have been integrated into Turkey’s nationwide real-time power quality monitoring system as its core event classification module.