Convolutional Neural Network-assisted fault detection and location using few PMUs

Detection and identification of faults in large distribution systems with limited metering continue to pose significant challenges for system operators. The prevalence of installed phasor measurement units (PMUs) in power systems, provides an invaluable resource for fault detection and location. Therefore, there is a rich literature on methods that leverage the benefits of these PMUs for the purpose of event detection. However, a notable proportion of the proposed solutions assume the presence of one PMU at every bus, or at a large percentage of buses that will make the system observable. Despite concerted efforts to substantially increase the number of installed PMUs, majority of utilities have yet to attain comprehensive PMU coverage at the terminals of every transmission line. Thus, many of the proposed methods cannot be practically applied to existing power systems with limited number of installed PMUs. Consequently, the objective of this paper is to develop a method, aided by the application of Convolutional Neural Networks, to identify and precisely locate faults within a system, even when only a minimal number of Phasor Measurement Units (PMUs) are present. [Display omitted] •ANN approaches are shown to enhance fault detection capabilities for limited number of PMUs.•CNN model is shown to outperform MLP in terms of fault detection performance.•Placement of PMUs is shown to have a significant effect on the performance of fault detection.