# Cross-Bonded Cable Circuits Identification Based on Deep Embedded Clustering of Sheath Current Sensing

## TL;DR

This paper introduces a new method for identifying high-voltage cable circuits online using sheath current data and deep learning clustering.

## Contribution

The novel approach combines temporal convolutional networks and clustering to enable online identification of cross-bonded cable circuits.

## Key findings

- The proposed method achieved 95.37% identification accuracy on a 110 kV cable test platform.
- The method demonstrates robustness against domain gaps and varying operating conditions.
- Sheath current temporal similarity was verified using Dynamic Time Warping simulations.

## Abstract

Online identification of HV cable circuits is vital for routine inspection and maintenance, yet existing passive electromagnetic wave injection methods are limited to offline operations. To fill the gap and achieve the online identification of HV cable circuits, an online circuit identification methodology based on sheath current temporal characteristics and deep embedded clustering is proposed. First, an equivalent circuit model of the multi-circuit cross-bonded cable sheath was built to deduce the temporal similarity of sheath currents within the same circuit, establishing the identification criterion. Second, the robustness of the temporal similarity under various operating conditions was verified via simulation based on the Dynamic Time Warping (DTW) distance. Then, a combined model of Temporal Convolutional Network Autoencoder (TCN-AE) and K-medoids was established to transform circuit identification into a temporal clustering problem of sheath currents, realizing circuit determination by synchronously monitoring the time-series sheath current data of multi-circuit HV cross-bonded cables. The method was verified on a full-scale 110 kV cable test platform. The results show that the identification accuracy reached 95.37%, and the proposed method can effectively identify the circuits of cross-bonded cables with high robustness against the domain gap, having significant engineering application value.