Fault Diagnosis for Terminal of 10-kV XLPE Cable Based on the Improved M-Training Algorithm

The 10-kV cross-linked polyethylene (XLPE) cables are widely used in urban distribution networks, of which the cable terminals are the weak links. In this article, the fault diagnosis for terminal of 10-kV XLPE cable was studied based on the improved M-training algorithm. First, the cable terminal samples with typical defects were prepared, and partial discharge (PD) data were obtained by the experiment platform. Then, the features of PD data were extracted by using the fusion of the texture features of gray gradient co-occurrence matrix (GGCM) and Goes–Markov random field (GMRF). Afterward, the M-training algorithm was improved from two aspects: 1) four different classifiers were used instead of the four identical classifiers in the conventional algorithm and 2) to overcome the effects of the error rates of unlabeled samples on classification error, the classification error in the M-training algorithm was optimized. Based on the improved M-training algorithm, the diagnosis method for 10-kV XLPE cable terminal was proposed. Finally, the proposed method was verified: the confusion matrix was used to express the recognition results, and the recognition effects of different schemes were analyzed based on four indicators (i.e., accuracy, sensitivity, specificity, and
${F}_{{1}-\text {sc}\text {or}{e}}{)}$
. The results show that the recognition accuracy rate was 97.8%, with sensitivities of 98.1%, 97.9%, and 97.4%, and specificities exceeding 97%, and
${F}_{{1}-\text {sc}\text {or}\textit {es}}$
being 97.6%, 97.9%, and 97.9%, respectively.