Real defect partial discharge identification method for power cables joints based on integrated PJS-M and GA-SVM algorithm with multi-source fusion

Previous studies on 10 kV cable intermediate joint defects have mainly focused on typical defect types and employed single-sensor data acquisition, leading to incomplete characterization of defect features and reduced recognition accuracy. To address this limitation, three real-type partial discharge (PD) models were developed based on common defects encountered in actual manufacturing. PD signals were collected using a combination of High-Frequency Current Transformer (HFCT) and Ultra High Frequency (UHF) sensors, capturing time-domain waveforms, frequency-domain spectra, and Phase-Resolved Partial Discharge (PRPD) patterns, from which feature quantities were extracted. These features were used to train a novel Genetic Algorithm Weighted Support Vector Machine (GAW-SVM) model, which incorporates an adaptive PJS-M weighting coefficient and a correlation-analysis–based dynamic correction mechanism into the conventional GA-SVM framework. The proposed model was compared with several state-of-the-art SVM optimization algorithms, including GA-SVM, PCA-SVM, and PSO-SVM. Under multi-source feature fusion, the GAW-SVM achieved a defect recognition accuracy of 98.84%, outperforming GA-SVM by 3.49%, PCA-SVM by 2.33%, and PSO-SVM by 1.17%. These results demonstrate that the proposed method significantly improves the accuracy of identifying complex real-type defects in 10 kV cable intermediate joints under multi-source feature conditions, providing a reliable diagnostic basis and technical reference for partial discharge detection in industrial applications.