CNN-Based Detection of Sheath Incorrect Connection in the Cross-Bonded HV Cable Systems Using the Sheath Current Phasor Difference

Incorrect connection of metallic sheath is a typical defect in long HV(High-Voltage) cross-boned cable systems. However, the previous detection methods based on sheath current rely on three criteria: the amplitude of the sheath current, the ratio of the sheath current to the load current, and the relative values of the sheath current across the three sheath loops. These criteria, however, are insufficient for accurately identifying the defect. In this paper, a CNN (Convolutional Neural Network) based methodology using the sheath current phasor difference at both ends of the same sheath loop is proposed to distinguish the 34 types of incorrect connection with 2 types of correct connection. Taking an 110 kV cable as the simulation case, the results show that the sheath current phasor difference is significantly affected by the unequal lengths of the three small segments of the cable. Besides, the sheath current phasor difference at both ends of the same loop can not only distinguish the correct and incorrect connection states of cross-bonded high-voltage cable sheaths but also differentiate different connection error modes. The classification model for high-voltage cable cross-bonding errors based on CNN can effectively recognize 36 different cross-bonding configurations of high-voltage cables, with a model accuracy of over 98%. Through comparison with other algorithms, it is found that the CNN exhibits the best performance in identifying various cross-bonding error scenarios.