Fault location in distribution networks based on SVM and impedance-based method using online databank generation

Fault location methods help to reduce outage time and improve reliability indices and therefore are important in practice. However, the performance of traditional fault location methods which are mainly developed for transmission grid is challenged by the specification and complexities of the distribution grid. Furthermore, the errors in measurement devices compromise the accuracy of the fault localization. This paper addresses these issues through an integrated methodology. In the proposed methodology, current transformer (CT) and potential transformer (PT) errors are first applied to current and voltage data recorded at the starting point of the feeder. Then, the impedance-based fault location method (IBFLM) is used to locate possible fault locations using the recorded voltage and current. Then, at the section of possible points, some locations are selected, the same fault is simulated, and an online databank is generated. After this, using a combination of the wavelet transform, Fourier transform and minimum redundancy maximum relevance (mRMR) algorithm, some features are selected and they can be separated using support vector machine (SVM) classifier. They are utilized to select one point as the final fault location among possible locations. A real feeder is considered as the sample distribution network to assess the performance of the proposed method. Instrument errors are modeled using the Gaussian stochastic process which is added to recorded signals at the starting point of the feeder. The accuracy of the proposed method is investigated under different fault locations, fault resistances, and fault inception angles. Simulation results confirm that the proposed method is highly accurate. The proposed method is tested in a distribution network in a power system simulator in the power system laboratory of Persian Gulf University. The experimental results confirm that the accuracy and precision of the proposed method are high. The method is also compared with other state-of-the-art methods, and the results show a clear improvement.