Detection and Localization of Faults in Distribution System Using Frequency and Time-Frequency Domain Analysis

Abstract

The rapidly growing demand for electric power leads to interconnection in distribution power systems. With the increasing number of interconnections in power systems, it becomes more and more complicated. The most important role of all the power system engineers is to ensure utmost reliability and continuity of service. It is also essential to reduce the repair time and speed up the system restoration after a fault. The outage time of power increases due to old fault detection and localization techniques. It causes massive workforce and time wastage. Therefore, it is highly needed to identify the faults and its exact location, so that the system restoration process becomes faster. It is not feasible to avoid the natural hazards, accidents, and mal-function of power system equipment. The natural hazards can further create a situation of a broken conductor, which lead to high impedance arc fault (HIAF). The HIAF poses a significant threat to the living being as it involves arcing. The enormous amount of heat generation during arc is also a major concern in this regard. The instrument for measuring the transient ought to have a high sampling rate to give accurate precision in portraying transient conditions concerning their amplitude and frequency content. These qualities are fundamental for performing a transient investigation. Based on the features of the fault transients, time-frequency based approaches i.e. wavelet, Stockwell Transform (ST), Hilbert-Huang Transform (HHT) and etc. has been applied along with contemporary machine learning algorithms i.e. artificial neural network (ANN), support vector machine (SVM), k-nearest neighbor (KNN) etc., to detect, classify and localize different kind of faults. The results obtained using different combinations of signal processing technique and machine learning based approach successfully detects, classifies and localizes different type of faults in a distribution system. An ST- ANN based approach is also developed to detect and localize faults in hybrid distributed generation (HDG) system with complex architecture. On the other hand, the fault signals of HIAF has been validated by both simulation and experimentation. The impact of HIAF under the influence of different arcing surface including metallic and non-metallic surfaces have also been analyzed in details. Suitable detection and classification algorithms have been developed based on empirical mode decomposition (EMD), ST and appropriate machine learning technique to improve the protection measures of a distribution system. All the results related to simulation and experimental has been repeated and compared with contemporary literature to get a comparative view and towards the selection of appropriate signal processing technique and machine learning algorithm. The proposed frequency and time-frequency based methods are found to be more superior in detecting and localizing faults in distribution system.