Abstract – This paper proposes an intelligent method to detect High Impedance Fault (HIF) in distribution networks using Wavelet Transform (WT) and data mining based Decision Tree (DT) model. The proposed method uses WT to decompose the current signal and extracts significant features of the signal. A data mining model reduces the features of the signal and also frames a DT model for the classification of HIF or non-HIF cases. The current signal data for HIF and non-HIF events (Capacitance switching, Linear and Non-Linear load switching,) have been acquired by an accurate model of an actual distribution system using MATLAB / SIMULINK. The simulation results show that the proposed method can provide a consistent and powerful protection for HIF.
Index Terms -High impedance fault, Wavelet transform, Data mining, Decision tree, Non-linear load.
The electric power distribution system is usually overhead lines. These are more vulnerable to breakdown of power supply due to the fact that is exposed to dissimilar climatic circumstances. Some of these failures may be detected and located easily. However, there are few failures which cannot be detected by conventional protective means. For example, when an energized broken or unbroken line connected to high impedance objects or surfaces draws the least amount of current 1 with no evidence of defect, such kind of fault classified as High Impedance Fault (HIFs). If The distribution system running with unidentified HIF for hours or days leads to damaging the equipment connected to the supply. Moreover, from the investigation, it is observed that, followed by HIF, the electric arcs turn out an arbitrary, erratic and asymmetric current 2. The distribution systems are close to populated areas, where the electrical arcs lead to fatal for the public.
Detection of HIFs are matter of interests from early1970, nevertheless the enlightening on detection process yet to be completed. A method based on the lower order harmonic ratio presented in 3. The drawback of such kind of methods needs to set one or few threshold value which is affecting the performance of detection method. Time-frequency analysis based methods 4-5 exposed good performance in the detection process. However, the percentage of false detection is shown as a major setback for practical applications.
Time domain approaches use to detect the change in current or voltage signal from pre-fault conditions. Mathematical morphology based time domain methods found in 6-8, which is an effective detection method in a balanced system, there are few issues associated with unbalanced network.
Wavelet Transform (WT) has been extensively used in signal processing because of its capability to detect the frequency component and their position in time. More than a decade such methods applied to power system protection. Although WT based methods are providing good detection rate with linear loads 9-13, no evidence of non-linear loads inclusion with the systems while detecting HIFs except in 14. The existence of non-linear loads (NLLs) in the system has continuously increased right through the power distribution grids. Whereas huge quantities of existing methods have failed to consider NLLs while modelling and designing of practical HIF detection methods. The NLLs and HIFs characteristics have closely resembled each other, which will make the existing methods less effective. Hence, In this paper, an enhanced method for detection of HIFs including huge variation of NLLs is Proposed.
The rest of the paper has been organized as follows. Section II explains the test system and also the characteristics of HIF model engaged for simulation. The process of proposed methodology has been thoroughly discussed in Section III, including the decomposition of a signal, feature extraction, feature selection and classification. Section IV has results and discussion and the paper concluded with the main highlights of the work in section V.