A Novel Machine Learning Based Commutation Failure Prediction Method

Commutation failure is one of the most common faults in HVDC transmission system. The effective prediction of commutation failure is conducive to the security and stability of the AC and DC system. Since the HVDC system is complex and affected by multiple factors, traditional commutation failure prediction methods are hardly to meet the requirement of accuracy and rapidity simultaneously. Therefore, this paper proposes a novel machine learning (ML) based commutation failure prediction methodology that is capable of learning the relations between the field measurements and the commutation failure results directly. Since the ML models are black-box models, several existing or revised indexes that combined with measured variables are used as input features to improve the interpretability of the proposed methodology. In addition, the prediction accuracy is further improved through feature screening. The CIGRE HVDC standard test model is used to verify the effectiveness of the proposed method.