Holistic Fault Detection, Classification, and Localization in ADNs With Optimally Placed D-PMUs

Abstract

Effective fault detection allows distribution system operators (DSOs) to provide reliable power to customers in active distribution networks (ADNs). Advantages include faster maintenance, less downtime, fewer critical system events, economic benefits, and prevention of network-wide cascading faults. In this paper, a new fault detection, classification, and localization (FDCL) method is developed that uses optimally placed distribution-level phasor measurement units (D-PMUs) to track positive, negative, and zero-sequence components considering unique distribution network attributes (UDNA). A simple feed-forward neural network (FFNN) detects and classifies faults, while an ensemble bagged trees (EBT) classifier approximates line fault location at a high fault resolution. A coordinated localization strategy using nearby D-PMUs reduces fault localization error further. The proposed voltage-induced transient method also correlates fault impedance and ground resistance. The IEEE-13, IEEE-34, and IEEE-123 test feeders verify the method's fidelity, including topology changes, spontaneous renewable integration effects alongside various D-PMU placement methods. Computational optimization through cloud computing and ADN splitting workflow is also established.