Portal de Periódicos da CAPES

High impedance fault modelling and application of detection techniques with EMTP‐RV

2018; Institution of Engineering and Technology; Volume: 2018; Issue: 15 Linguagem: Inglês

10.1049/joe.2018.0217

2051-3305

Vassilis C. Nikolaidis, Angelos D. Patsidis, Aristotelis M. Tsimtsios,

Power Transformer Diagnostics and Insulation

The Journal of EngineeringVolume 2018, Issue 15 p. 1120-1124 The 14th International Conference on Developments in Power System Protection (DPSP 2018)Open Access High impedance fault modelling and application of detection techniques with EMTP-RV Vassilis C. Nikolaidis, Corresponding Author Vassilis C. Nikolaidis vnikolai@ee.duth.gr Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi, GreeceSearch for more papers by this authorAngelos D. Patsidis, Angelos D. Patsidis University of Strathclyde, Glasgow, UKSearch for more papers by this authorAristotelis M. Tsimtsios, Aristotelis M. Tsimtsios Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi, GreeceSearch for more papers by this author Vassilis C. Nikolaidis, Corresponding Author Vassilis C. Nikolaidis vnikolai@ee.duth.gr Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi, GreeceSearch for more papers by this authorAngelos D. Patsidis, Angelos D. Patsidis University of Strathclyde, Glasgow, UKSearch for more papers by this authorAristotelis M. Tsimtsios, Aristotelis M. Tsimtsios Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi, GreeceSearch for more papers by this author First published: 24 August 2018 https://doi.org/10.1049/joe.2018.0217Citations: 3AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract The aim of this study is to develop a software-oriented framework that will help simulate and study high impedance faults (HIFs) in power distribution systems, as well as design adequate HIF detection techniques. After a thorough literature review, three well-established HIF models have been developed in EMTP-RV, as part of this work. Systematic (black box) modelling is applied so that each developed HIF model can be used in any power system under investigation in the future. An overview of the mostly available HIF detection methods is also included in this study, which subsequently concludes with the development of two proper HIF detection techniques in EMTP-RV/MATLAB environment. A real 20 kV radial overhead distribution line located in Xanthi, Greece is modelled in EMTP-RV and several HIFs at various time and locations are simulated. The simulation results are presented in order to evaluate the accuracy of the developed HIF models and the effectiveness of the developed HIF detection techniques. 1 Introduction A high impedance fault (HIF) emanates from the electrical contact between a primary conductor and a high impedance object or material, such as asphalt, sand, gravel, concrete, grass, soil trees or adjacent tree branches [1]. Often this is seen as a fallen conductor left energised on the ground surface. The result of a HIF is a low magnitude fault current, which is usually lower than the normal load current of the line. Hence, HIFs cannot be sensed and cleared by conventional protection means, such as overcurrent relays and fuses that are extensively used in distribution systems. As a result, the distribution system continues to operate whilst concurrently feeding the fault. The inability to clear the fault can lead to negative outcomes such as equipment damage, fire enkindling and electrocution [2]. In order to find adequate methods to detect a HIF in distribution systems, deep understanding of the complex nature of a HIF is required. This nature is usually related with low magnitude fault current, arcing, time-varying harmonic content [2], non-linear current–voltage characteristic [3], asymmetry of fault current waveform and, provided certain conditions, gradually growing of the HIF current (build-up) until a maximum value (shoulder) [4]. Various HIF models have been proposed over the course of time to represent one or more of the aforementioned characteristics of the complex HIF nature accurately. An early model, originally proposed in [5], suggests the usage of diode-voltage sources based circuit alongside an impedance in order to simulate a HIF. A similar circuit, in which the impedance is replaced by non-linear resistances, is proposed in [6]. The authors in [4, 7] introduced a transient analysis of control systems (TACS) controlled circuit, which includes time-varying voltage sources. In [4], a model consisting of two time-varying non-linear resistances in series is developed, based on experimental data. Slight alterations in the circuit of the abovementioned models have been proposed in [8, 9]. Hochrainer's arc equations are applied in [10], where a feedback-based model is introduced. Eventually, Elkalashy et al. [11] designed a dynamic arc model in order to represent a high impedance arcing fault based on Cassie and Mayr equations. A variety of HIF detection methods has been developed. These approaches can be categorised primarily based on the measurements they use in order to detect a HIF. Mostly, current, voltage, combined current–voltage, as well as magnetic field measurements have been used in the implementation of HIF detection algorithms [12]. Moreover, detection methods vary depending on the selected domain of analysis: time versus frequency domain. Finally, signal processing techniques [e.g. wavelets, Fourier analysis, mathematical morphology (MM) etc.] are found in the literature to be applied for HIF detection in distribution systems [12]. 2 HIF modelling In this paper, three different HIF models (namely Model-1, Model-2 and Model-3) are developed in EMTP-RV. 2.1 Model-1 Model-1 [9] consists of a non-linear, time-varying resistance R 1 in series with two branches, paralleled inversely (Fig. 1). Each branch includes a diode in series with a dc voltage source (Vp and Vn) and a resistance (Rp and Rn). The current sources in parallel with the resistors are not used and set to zero. Fig. 1Open in figure viewerPowerPoint Model-1 implementation of HIFs in EMTP-RV The time-varying resistor R 1 is used to reflect the build-up and shoulder characteristics of the HIF current waveform (Fig. 2). Based on experimental data, R 1 (in ) can be expressed by [3] (1) where t is the time (in seconds) from the beginning of the fault. Fig. 2Open in figure viewerPowerPoint HIF current build-up effect The voltage sources Vp and Vn express the breakdown voltage (arcing voltage threshold) between the conductor and the surface in the positive and negative half-cycle period, respectively. Both voltage sources are set to 6.15 kV, in order to obtain the desired non-linear waveform of [4]. The diodes alongside the voltage sources guarantee the limitation of the current fault to zero when the voltage of the conductor is between Vp and Vn. The branch resistances Rp and Rn represent the fault resistance appearing in the positive and negative half-cycle period of the voltage waveform, respectively. The different values of branch resistances are due to the asymmetry in fault resistance appearing actually in the two half-cycle periods. Moreover, the magnitude of the branch resistances varies stochastically in the range of around the assigned value. In this work, the assigned (mean) value of Rp and Rn has been calculated for a typical 20 kV Greek overhead distribution line by scaling the respective magnitudes, documented in [13] for several contacted materials, appropriately. 2.2 Model-2 Model-2 [4] employs two non-linear time-varying resistors in series to represent a HIF. The first resistor R 1 (t) is that described in Section 2.1 and is used again in order to represent the build-up and shoulder of the fault current waveform. The second time-varying resistor, namely R 2 (t), is used to represent the non-linearity and the asymmetry of the HIF current in the steady state. Based on experiments [4] performed by Korea Electric Power Corporation (KEPCO), R 2 (t) can be determined by dividing the current–voltage magnitudes, as retrieved by a measured periodic characteristic, which has the form shown in Fig. 3. In this work, KEPCO's particular current–voltage characteristic is scaled down in order to adapt to the Greek distribution system operating voltage level of 20 kV. Fig. 3Open in figure viewerPowerPoint Current–voltage characteristic of a steady-state HIF resistance [14] As can be seen from Fig. 3, each voltage magnitude corresponds to two different current magnitudes. Therefore, in the implementation of Model-2 in EMTP-RV (Fig. 4), two non-linear resistances Rn 1 and Rn 2 are applied instead of the previously described R 2 (t). Rn 1 corresponds to the voltage-increasing period (lower part of the characteristic of Fig. 3), whereas Rn 2 corresponds to the voltage-decreasing period (higher part of the characteristic of Fig. 3). The derivative dv /dt is used as a criterion to distinguish between an increasing and a decreasing voltage magnitude. Fig. 4Open in figure viewerPowerPoint Model-2 implementation of HIFs in EMTP-RV 2.3 Model-3 Model-3 [11] represents a time-varying arc conductance using the well-known differential equations of Cassie-Mayr (2) (3) where g is the time-varying arc conductance, is the arc time constant, G is the stationary arc conductance, |i | is the absolute value of the arc current and V arc is the voltage clipping level of the arc. The time constant can be approximated [11] by (4) where A, B are constants determined through experiments. V arc can be determined as the arc voltage value when dg /dt = 0 and, if it is synchronised with the instant of maximum current occurrence, and it is given by (5) The implementation of Model-3 in EMTP-RV is quite complex and it is schematically presented in Fig. 5. Fig. 5Open in figure viewerPowerPoint Model-3 implementation of HIFs in EMTP-RV 3 HIF detection techniques Two HIF detection techniques are developed in this paper. The first applies MM and the second applies spectral analysis (SA). Although applying such relatively advanced detection techniques seems incompatible with current protection philosophy, modern numerical relays are capable of including complex algorithms. Therefore, these or similar detection methods could be foreseen as future integrations into commercial numerical relays. 3.1 Computation framework Both HIF detection techniques are developed in MATLAB, which cooperates with EMTP-RV offline. That means that initially, a HIF scenario is simulated in EMTP-RV. Then, the resulted waveforms are visualised in ScopeView, whereas the corresponding raw data are exported in appropriate data files. These data files are pre-processed and imported as.mat files in MATLAB in order to run the desired HIF detection algorithm for the simulated scenario. At this stage, further data pre-processing may be needed, mainly in order to match the samples taken from the simulation with the sampling frequency of the applied detection algorithm. Finally, a HIF detection flag is asserted or not. The whole procedure is illustrated in Fig. 6. Fig. 6Open in figure viewerPowerPoint Interface between EMTP-RV and MATLAB 3.2 MM-based technique In power system applications, MM-based techniques show superior behaviour as processors of non-linear signals in time domain. They have been used in studies regarding power quality, power system transients and power system protection. The MM-based HIF detection method developed in this paper is that presented in [15]. This method utilises the voltage waveform in order to acquire the necessary information about a HIF detection. The basis behind this selection is that since in distribution systems the overcurrent relay is installed at the substation, any distortion in the current measurement (waveform) due to a HIF would be suppressed by the load current fed along the line. A great advantage of the method is that it can distinguish HIFs between other transient phenomena such as load or capacitor switching, which produce similar disturbances in the voltage waveform at the substation level. The main MM transformations are dilation and erosion. Every other transformation, such as opening and closing emanates from those two transformations. The execution of those transformations heavily relies on the proper decision of the structuring element (SE), which is a signal-processing tool, designed to transform and divide a signal into subsignals. According to [15], the closing opening difference operation (CODO) function has been proven highly efficient in HIF detection. For a given signal f (n) and a structure element g (m), where n and m are integers, and n > m, CODO function is defined as follows: (6) A linear SE with a length of 2 and height of 0.01 is ideal for detecting HIFs in distribution systems based on a normalised measured voltage waveform [16]. The time delay for the CODO is , where m is the SE length and is the sampling rate. The algorithm is implemented in MATLAB using the built-in functions of dilation, erosion, opening and closing. In particular: The voltage waveform of interest (substation voltage), retrieved from EMTP-RV simulation, is normalised and used as input for the algorithm. Proper sampling of the voltage waveform is performed for the entire simulation time. The SE is created. The CODO function output is calculated and normalised. CODO function outputs a sequence of spikes after a significant disturbance occurring in the distribution system distorts the voltage waveform at the substation. These spikes may appear for a short or a longer period after the disturbance and may differ significantly in their magnitude depending on the root cause. Disturbances that produce significant spikes may be a load/capacitor switching or a HIF. In fact, it has been proved [15] that in case of a load or capacitor switching operation, CODO output consists of one low magnitude spike and a successive series of two high magnitude spikes. On the other hand, even under normal network operation spikes may be caused due to small disturbances. In order to filter out spurious spikes appearing during normal network operation, as well as to discriminate between normal situations and actual disturbances, an adequate threshold (CODOthr) in the CODO output magnitude is determined. Hence, the algorithm monitors CODO output continuously and it is accessed only if a spike exceeds the predefined threshold CODOthr. In this work, CODOthr is set at 115% of the maximum CODO output observed during normal network operation, as proposed by Gautam and Brahma [15]. Since, however, disturbances like capacitor or load switching can produce spikes of sufficient magnitude to trigger the algorithm, additional criteria are applied in order to discriminate between HIFs and load/capacitor switching operations. In particular, a wait time delay T w, as well as a reset time delay T r, are determined. Actually, a timer starts counting when the CODO output exceeds the threshold CODOthr. A HIF is identified if spikes continue exceeding the CODOthr after time delay T w passes and before the reset time delay T r is exceeded. Consequently, at least a second spike detected between T w and T r is required to produce a HIF alarm. Spikes during T w are not activating the HIF alarm. The absence of a spike until the end of T r indicates that there is no HIF in the network, and therefore the algorithm resets itself. 3.3 SA-based technique The second method developed as part of this work, based on [17], monitors the one-sided amplitude spectrum of the apparent impedance seen by an observer at the departure of the distribution line. Obviously, in this case, both voltage and current measurements from the line at the substation level are required. The transformation of the measured waveforms from the time domain into the frequency domain is achieved by applying the discrete Fourier transform (DFT). DFT transforms a sequence of discrete-time complex samples from the time domain into another sequence of complex samples in the frequency domain through the following equation: (7) where n is the time index, x (n) is a sample in the discrete time domain, N is the number of samples, k is the frequency index and X (k) is the k th coefficient of the DFT. The DFT frequency resolution is given by (8) Additionally, as X (k) is a complex number, the absolute value is calculated as seen below: (9) The one-sided amplitude spectrum is calculated through (10) (11) In order to avoid the aliasing phenomenon, the sampling frequency f s must be at least twice as the maximum frequency of the analysed signal. Furthermore, in order to avoid the spectral leakage phenomenon, the total number of samples should be determined by an integer multiplication of the signal period. The algorithm is implemented in MATLAB, following five discrete steps: The line current i (t) and voltage v (t) signals are retrieved from the EMTP-RV simulation. DFT is applied to i (t) and v (t). The one-sided amplitude spectrum of the current (Ik, k = 0,…, N /2) and the voltage (Vk, k = 0,…, N /2) is calculated. The one-sided amplitude spectrum of the impedance Zk is calculated by dividing Vk with Ik for each particular sample k = 0,…, N /2. If Zk is monotonously decreasing as k increases, the network operates in a normal state. Otherwise, the monotony of Zk is investigated for k > 4. If Zk is monotonously increasing as k increases beyond 4, a capacitor switching action is identified. If this condition does not hold, the monotony of Zk is investigated in the range . If Zk is monotonously decreasing in this range, a load switching action has occurred. Otherwise, a HIF is identified. 4 Simulation analysis 4.1 Test system description The real 20 kV radial overhead line numbered 22 (hereafter called Line-22), departing from the 150/20 kV substation in Xanthi, Greece, is modelled in EMTP-RV. The maximum recorded current of this line is 150 A with a lagging power factor equal to 0.9. To simplify a bit the line representation, 0.4 kV loads fed by 20/0.4 kV substations are modelled as 20 kV constant power loads directly connected to the medium voltage line. Moreover, loads that are in a close distance are combined into a single medium voltage composite load. Finally, very short laterals with small conductor cross-sections are assumed directly connected to the main feeder. The Line-22 representation in EMTP-RV is illustrated in Fig. 7. Fig. 7Open in figure viewerPowerPoint Line-22 representation in EMTP-RV (with indicators of the fault positions) 4.2 Simulation procedure and results All three HIF models (Model-1, Model-2, and Model-3) developed in this work are used in the simulation analysis that follows. Three different fault positions are assumed, namely the departure of the line, the electrical centre of the line and the remotest end of the line. These positions are indicated with the stroke symbols in Fig. 7. Furthermore, two different fault initiation instances are assumed: at the zero crossing and at the maximum of phase-a voltage. MM- and SA-based HIF detection techniques are applied, considering a sampling rate equal to 4 kHz for both, in order to fulfil the methods' requirements. For properly applying the MM-based HIF detection technique, a considerable number of CODO outputs is taken initially. Based on the analysis of those outputs, T w and T r are taken equal to 0.2 and 0.5 s, respectively. CODOthr is set at 0.003 pu. The MM-based technique detected efficiently all the HIF cases, which were represented by Model-1 and Model-2, occurring on the three different network positions and at the two different time instances (phase-a voltage maximum and zero magnitude) considered. The average detection time was 0.68115 s. The MM-based technique was not able to detect the same HIF cases, in terms of fault position and time instance, when Model-3 was used. This is due to the extremely low fault current magnitude (0.075% of load current) because of the 200 kΩ resistance R _object considered in Model-3 [11]. By varying the magnitude of the resistance R _object, it was found that the minimum required fault current in order to detect a HIF with the MM-based technique is 5% of the load current. No other disturbance apart from HIF was mistakenly detected by the MM-based technique. Fig. 8 illustrates the CODO output for a HIF at the remotest end of the line, initiated at zero crossing of phase-a voltage. Model-3 is used in this case with a reduced R _object. As can be seen, a series of spikes is produced after the fault instance at t = 0.5 s, which clearly exceeds the CODOthr (0.003 pu), even after Tw expires. Hence, the HIF is efficiently detected and an alarm signal is produced. Fig. 8Open in figure viewerPowerPoint CODO output Regarding the SA-based HIF detection technique, the sampling period is set equal to 2.0 s, in order to avoid any spectral leakage. Eighteen (18) different scenarios involving the three HIF models developed as part of this work, the two different fault initiation times, and the three different fault positions in the network are simulated. SA-based technique successfully distinguished and detected all HIFs. Fig. 9 illustrates the one-sided amplitude spectrum of the impedance Zk for a HIF at the remotest end of the line, initiated at zero crossing of phase-a voltage. Model-3 is used in this case with R _object equal to 200 kΩ. The green solid line connects the calculated apparent impedance magnitude corresponding to an integer harmonic. Since Zk oscillates as k increases, the existence of a HIF in the network is confirmed. Several load/capacitor switching actions were simulated and no false alarm was given in any case for both methods. Fig. 9Open in figure viewerPowerPoint One-sided amplitude spectrum of Zk 5 Acknowledgments The authors thank POWERSYS, for providing them the license to use EMTP-RV. In addition, they thank HEDNO SA, for providing them the actual data of the distribution line under investigation. 6 References 1Aucoin B., Jones R.: 'High impedance fault detection implementation issues', IEEE Trans. Power Deliv., 1996, 11, (1), pp. 139 – 148 2Hou D.: 'High-impedance fault detection–field tests and dependability analysis'. 36th Annual Western Protective Relay Conf. Proc., Spokane, Washington, USA, 2009 3Chang T.: ' Impact of distributed generation on distribution feeder protection'. Master of Science Thesis, University of Toronto, Canada, 2010 4Nam S.R., Park J.K., Kang Y.C. et al.: 'A modeling method of a high impedance fault in a distribution system using two series time-varying resistances in EMTP'. IEEE Power Eng. Soc. Summer Meeting Proc., Columbus, Ohio, USA, 2001, vol. 2, pp. 1175 – 1180 5Emanuel A., Cyganski D., Orr J. et al.: 'High impedance fault arcing on sandy soil in 15 kv distribution feeders: contributions to the evaluation of the low frequency spectrum', IEEE Trans. Power Deliv., 1990, 5, (2), pp. 676 – 686 6Sharaf A.M., Snider L.A., Debnath K.: 'A neural network based back error propagation relay algorithm for distribution system high impedance fault detection'. 2nd Int. Conf. on Advances in Power System Control, Operation and Management Proc., Hong Kong, Honk Kong, 1993, pp. 613 – 620 7Chan D., Wai T., Yibin X.: 'A novel technique for high impedance fault identification', IEEE Trans. Power Deliv., 1998, 13, (3), pp. 738 – 744 8Lai T.M., Snider L.A., Lo E.: 'Wavelet transform based relay algorithm for the detection of stochastic high impedance faults'. Int. Conf. on Power System Transient Proc., New Orleans, USA, 2003, pp. 1 – 6 9Sheng Y., Rovnyak S.M.: 'Decision tree-based methodology for high impedance fault detection', IEEE Trans. Power Deliv., 2004, 19, (2), pp. 533 – 536 10Michalik M., Rebizant W., Lukowicz M. et al.: 'Wavelet transform approach to high impedance fault detection in MV networks'. IEEE PowerTech Conf. Proc., 2005, pp. 1 – 7, St. Petersburg, Russia 11Elkalashy N., Lehtonen M., Darwish H. et al.: 'Modeling and experimental verification of a high impedance arcing fault in MV networks', IEEE Trans. Dielectr. Electr. Insul., 2007, 14, (2), pp. 375 – 383 12Ghaderi A., Ginn H.L., Mohammadpour H.A.: 'High impedance fault detection: a review', Int. J. Electr. Power Energy Syst., 2017, 143, pp. 376 – 438 13'High Impedance Fault Detection Technology', Report of PSRC Working Group D15, 1996 14Uriarte F.B.: ' Modeling, detection, and localization of high-impedance faults in low-voltage distribution feeders'. Master of Science Thesis, Virginia Tech Polytechnic Institute and State University, USA, 2003 15Gautam S., Brahma S.M.: 'Detection of high impedance fault in power distribution systems using mathematical morphology', IEEE Trans. Power Syst., 2013, 28, (2), pp. 1226 – 1234 16Gautam S., Brahma S.M.: 'Guidelines for selection of an optimal structuring element for mathematical morphology based tools to detect power system disturbances'. IEEE Power & Energy Society General Meeting Proc., San Diego, USA, 2012, pp. 1 – 6 17Liang Z.: ' High impedance fault detection in power distribution systems with impedance-based methods in frequency domain'. Master of Science Thesis, University of British Columbia, Canada, 2016 Citing Literature Volume2018, Issue15October 2018Pages 1120-1124 FiguresReferencesRelatedInformation