A Machine Learning-Assisted Framework for Commutation Failure Classification in High-Voltage Direct-Current (HVDC) Transmission Networks

Abstract

This paper presents a machine-learning based strategy for accurate identification of commutation failures instances. A commutation-failure identification (CFI) block has been proposed and implemented, which utilizes the DC current and the valve-side AC currents for accurate detection. The well-established CIGRE benchmark model, implemented in PSCAD/EMTDC, is employed as the test system. Extensive simulations are performed under diverse fault scenarios, and a comprehensive dataset is generated under both commutation failure and normal operating conditions. Six different machine-learning classifier models are trained using a sufficient portion of the generated dataset and are subsequently evaluated against the remaining test dataset. The performance of the classification is assessed though confusion matrices and several statistical metrics including Accuracy, Precision, Recall and F1-Score for the base case study. Several other studies including the reduced feature cases, noisy validation scenario, five-fold cross validation, and hyperparameter sensitivity analysis have been performed to validate the effectiveness and reliability of the proposed ML-assisted approach. The results demonstrate that the proposed framework can reliably distinguish commutation failure events from normal operating conditions with high accuracy. Among all these classifier models, the Random Forest classifier has been identified as the most suitable model to classify the commutation failure instances with an Accuracy of 99.53% and Recall score of 99.63% for the base case. This accurate classification of commutation failure will ensure enhanced reliability of an HVDC system.