Utilizing a Combination of Prony Analysis and Singular Value Decomposition for Intelligent Fault Locating in Bipolar High Voltage Direct Current Transmission Lines

High voltage direct current (HVDC) transmission lines can be used to transfer bulk power over long distances. Accurate estimation of fault location in these transmission lines is very essential to speed up the maintenance operations. This paper presents a new approach for intelligent fault locating in bipolar HVDC transmission lines using the pattern recognition techniques and the machine learning algorithms. In the proposed approach, using a combination of Prony analysis (PA) and the singular value decomposition (SVD), some useful features are extracted from the post-fault voltage signals measured at one the line terminals. Then, a pre-trained generalized regression neural network (GRNN) receives the extracted features and estimates the corresponding fault location. Tests conducted on a sample bipolar system show that the proposed fault locator has accurate and appropriate performance despite changes in fault location, fault resistance, and pre-fault current. The obtained average percentage fault location errors for the positive-pole-to-ground (PG), positive-pole-to-negative-pole (PN), and positive-pole-to-negative-pole-to-ground (PNG) faults in the sample system under study are 0.264%, 0.287%, and 0.225%, respectively.