Context: Intelligent fault locating in transmission lines consists of three main steps: feature extraction, feature selection, and utilizing a learning tool. Objective: The main objective of this paper is to propose a systematic approach for intelligent fault locating in transmission lines. Method: This paper extracts a group of candidate features by applying a combination of the Wavelet Packet Decomposition (WPD) and Improved Prony Analysis (IPA) methods on single-ended voltage measurements. To have an accurate fault location estimate, useful and efficient features are selected among the candidate features using the regression relief algorithm. In this paper, performances of three regression learning tools including the Generalized Regression Neural Network (GRNN), k-Nearest Neighbor (k-NN) and the Random Forests (RF) in the fault location problem are evaluated and compared, and the best tool is introduced. Results: Numerous training and test patterns are generated through simulation of various fault types in an untransposed transmission line based on different values of fault location, fault resistance, fault inception angle, and magnitude and direction of load current. The results of evaluation using theses patterns show the high efficiency and accuracy of the proposed approach. For various fault types in the test cases, the average values of fault location estimation errors are in the range of 0.153–0.202%. Conclusion: Besides accuracy, the proposed fault locating method is immune against current signal measurement errors and it does not face the problems and costs related to the transmitting and synchronizing data of both line ends.

Keywords