A Learning-Based Framework for Locating Faults on Power Grid Lines Based on Distortion of Traveling Waves

Accurate measurement of traveling waves (TWs) and processing their data in the time and/or frequency domains can provide valuable information about faults in the system. On this basis, this article presents a learning-based framework (LBF) for locating faults on overhead lines (OHLs) and underground cables (UGCs) using TWs. The proposed method is single-ended, i.e., it does not require communication and time synchronization, and locates faults based on distortion of the first TW received at the terminal of the line and attenuation of its frequency components. Therefore, in contrast to the majority of existing methods in the literature, differentiating between the reflected waves from the fault point and the remote end is no longer mandatory if only single-ended measurements are available. To locate faults, the proposed method trains a multilayer perceptron (MLP) model based on attenuation of TWs' frequency components when they travel from their corresponding fault locations to the terminal of the line. Since this attenuation is a function of frequency and traveling distance, a well-trained MLP model can estimate the fault's location by extracting the frequency components of its first TW using the fast Fourier transform (FFT), and by inferring their attenuation pattern. Additionally, the proposed method minimizes the effects of noise and the limited bandwidth of measurement equipment, e.g., current transformers (CTs), by considering these factors in the learning process of the MLP model. Simulation results obtained from PSCAD/EMTDC for an OHL and a UGC in CIGRE transmission and distribution benchmark systems corroborate the effectiveness and accuracy of the proposed LBF in locating faults occurred on power grid lines. The results also confirm that the proposed method works independently of fault parameters and system operating conditions.