A Method for Enhancement and Identification of Typical Partial Discharge Signals in Distribution Cable Accessories Considering Signal Propagation Characteristics

Partial discharge (PD) is the main cause of insulation degradation in cables. Therefore, collecting and identifying PD pulse signals can provide an accurate assessment of the cable’s insulation condition. Traditional cable PD identification methods, based on single PD pulses, typically only consider signal characteristics along specific propagation paths. When the PD location changes, the signal characteristics also change, which may result in the identification model failing to accurately recognize the same type of discharge signal. To address this issue, this paper proposes an enhancement and recognition method for partial discharge signals of typical defects in cable accessories, which takes into account the signal propagation characteristics. The method enhances the PD signal data for typical defects using the cable's transfer function, thereby creating a dataset containing discharge signal features at various propagation distances. A classical ResNet18 model is then used to train the PD classification model on this enhanced dataset. Validation using real PD data shows that, compared to classifiers trained on signal features from a single propagation path, the classifier enhanced by this method significantly improves the recognition accuracy for PD signals from different discharge locations, achieving an accuracy of 96.23%.