Health Assessment of Underground Power Cables: A Data-Driven Approach Based on One-Sample Maximum Mean Discrepancy

Abstract

Effective health assessment and management of underground power cable system is essential for ensuring the cost-efficient operation of power grids. Existing practices usually monitor condition of the overall health of the power cable circuits but overlook its intricate structural complexity. This letter proposes a data-driven health assessment method considering both failure-related structural characteristics of each cable circuit and condition measurement data. Firstly, distinguishing failure indicators are identified by integrating in-depth component analysis with circuit condition measurements. Based on these features, a novel integrated health index is proposed, named as One-sample Maximum Mean Discrepancy (O-MMD), which evaluates the health condition of a circuit by quantifying the disparity between un-faulted assets and typical faulted groups in a high-dimensional feature space. This approach is able to assess whether new or modified circuits might exhibit similar fault characteristics as previously observed cases. Based on the O-MMD index, the probability of failure can be estimated for each cable circuit, and the maintenance plan can be prioritized. The proposed method is demonstrated on both seen and unseen real-world underground cable system data in Singapore.