AI-driven single-end partial discharge localization in power cables based on time domain reflectometry and transfer function analyses

Abstract

Accurate localization of partial discharge (PD) in power cables is critical for minimizing downtime and associated costs. Therefore, this paper presents a single-end localization method that simplifies implementation by avoiding the complexities of double-sided or distributed schemes. A fundamental challenge for online monitoring systems based on a single-end measurement scheme is the accurate and autonomous identification of incident pulses and their corresponding reflections, particularly in environments where impulse noise and PD-like interference are present and may resemble actual PD pulses, making it difficult to distinguish true events from interfering pulses. In this regard, this paper proposes a method based on the traveling wave characteristics and transfer function (TF) analysis to pinpoint the PD source accurately, even in challenging conditions such as multi-path propagation, impulse noise, and simultaneous PD events. To achieve this, a cable-specific attenuation characteristic is developed and incorporated within a two-step signal segmentation algorithm, and then the U-Net model is employed to estimate PD pulses’ arrival time precisely. Additionally, the proposed method provides a statistical analysis of its maximum localization capability based on the noise level and cable length. The performance of the method is assessed under both homogeneous and inhomogeneous cable configurations. The results demonstrate a localization error of less than 1% for a 1.5 km cable.