1

Fault-Induced Transient Analysis for Real-Time Fault Detection and Location in Transmission Lines
F. B. Costa and B. A. Souza
AbstractThe real-time detection of the fault-induced transients on both transmission line terminals can allow the directional protection system to detect and locate faults in agreement with a high-speed fault clearance. This paper presents a waveletbased method for real-time detection of the fault-induced transients. The method can detect whether the fault is within the protected line and estimate the fault location through a communication link at both line terminals. The wavelet coefcients were computed with both the Discrete Wavelet Transform (DWT) and the Maximal Overlap Discrete Wavelet Transform (MODWT). However, the MODWT provided faster fault detection and better fault location. The method was implemented and evaluated by means of the Real Time Digital Simulator (RTDS). Index TermsFault-induced transients, fault location, transmission line protection, real-time wavelet transform.

I. I NTRODUCTION

IGH-SPEED fault detection and location are important issues in power system engineering in order to clear the faults quickly, restoring power supply as soon as possible with minimum interruption. In this context, the traveling wavebased protection can provide very fast relay operation in transmission lines. However, the high-frequency components of the traveling waves generated by the system faults are not within conventional transducer bandwidths. On the other hand, the traveling wave detection requires a high sampling frequency, limiting its application. To overcome these drawbacks, the fault-induced transients with a typical frequency spectrum from a few hundred Hz to various kHz can provide extensive information about the fault type, detection, location, direction, and sustained time in satisfactory agreement with real application in high-speed protective relays [1], [2], [3]. Fault-induced transients are non-stationary in both time and frequency domains. In this way, voltages and currents with fault-induced transients can be properly analyzed by using the DWT. This transformation is a well-known powerful tool to detect transients in power system disturbances [4], [5], such as fault-induced transients in three-phase overhead transmission lines. Much research has been focused on wavelet-based techniques applied on analyzing power system transients [6], [7], [8], detecting and classifying faults [9], [10], and estimating the fault location [11], [12].
This work was supported by the Brazilian National Research Council (CNPq) and PROPESQ-UFRN. The author is with Federal University of Rio Grande do Norte - School of Science and Technology, Campus Univesit ario Lagoa Nova, Natal - RN, CEP:59.078-970, Brazil. E-mail: [email protected]. The author is with the Department of Electrical Engineering at Federal University of Campina Grande, 882 Apr gio Veloso Av, Bodocong o, Campina Grande - PB, CEP:58.109-970, Brazil. E-mail: [email protected].
Presented at the International Conference on Power Systems Transients (IPST11) in Delft, Netherlands on June 14-17, 2011.

The fault-induced transients can be detected in real-time by means of the wavelet coefcient analysis [13]. The energies of the wavelet coefcients have also been used to detect and classify faults in real-time [14]. The real-time analysis of the wavelet coefcients can lead to the development of worthy fault detection and location methods to yield results in satisfactory agreement with real protective applications. Notwithstanding, the number of published papers regarding the real-time fault detection and location methods based on wavelet transforms has been small. The MODWT is a variant of the DWT which does not use the down-sampling process. In real-time applications, the wavelet coefcients of the MODWT are computed in each simulation time step or soon after each sampling process. As a consequence, the fault-induced transients can be detected faster by using the MODWT [13]. This paper addresses a fault detection and location method based on the real-time analysis of the wavelet coefcients. The proposed method is composed by three modules: 1) real-time detection of the fault-induced transients by using the wavelet coefcients of both the phase voltages and currents just at local terminal; 2) Identication of internal or external faults upon the protected line by using wavelet information from both terminals; 3) real-time fault location estimation by using wavelet information from both terminals. The RTDS was used in order to evaluate the performance of the proposed fault detection and location method. The wavelet coefcients at rst scale of both the MODWT and DWT were used for fault-induced transient detection, and good results were obtained. However, as expected, the fault-induced transients were detected faster and the estimation of the fault location was better with the MODWT-based method. II. P REPROCESSING OF THE R EAL -T IME FAULT D ETECTION AND L OCATION Fig. 1 depicts a simplied owchart of the proposed realtime fault location method. The voltages and currents are obtained from the system through potential transformers (PTs) and current transformers (CTs), respectively. Anti-aliasing lters with an appropriate cut frequency followed by analog to digital converters are required to convert the signals into digital signals. In this paper, the wavelet coefcients of the sampled voltages and currents are computed in real-time. These coefcients are used by the proposed method in order to detect and locate faults in real-time. The process to compute the realtime wavelet coefcients are described in the remainder of this section, and the real-time fault detection and location is properly addressed in the next section.

Fig. 1.

Flowcharting of the preprocessing module. Fig. 2. Power system model.

The simulation time step of the RTDS is t=50 s. In this case, the sampling frequency used in this paper is fs =20 kHz. Therefore, the frequency spectrum of the wavelet coefcients at the rst scale falls into the range [5, 10] kHz [15]. As a consequence, the wavelet coefcients at the rst scale are well-suitable for a fast detection of the highest frequency components of the fault-induced transients [13]. In addition, the real-time computation of the wavelet coefcients by using only the rst scale reduces the simulator efforts by comparing with applications in which various wavelet decompositions are required. According to [16], the effectiveness of the wavelet analysis is largely inuenced by the choice of the mother wavelet. However, the wavelet Daubechies 4 (db4) provides an accurate detection of the fast transients in power systems [5]. The four coefcients of the db4 lters provide a fast computation of the wavelet coefcients, which is an excellent feature for real-time applications [13]. The MODWT is a variant of the DWT. Both of them use low- and high-pass lters to divide the frequency-band of the input signal into low- and high-frequency components at various scales. However, in contrast to the DWT, there is no down-sampling process in MODWT [15]. As a consequence, the wavelet coefcient computation of the MODWT requires more computational effort. However, the MODWT provides twice the amount of coefcients to be analyzed in real-time (faster real-time analysis). The computationally efcient pyramid algorithm of the MODWT can be easily adapted for real-time computation of the wavelet coefcients [13]. If the samples of the signal are obtained in real-time, the wavelet coefcient soon after each sampling process is given by
L

The real-time fault detection and location (Fig. 3) is composed of three modules: Fault-induced transient detection: The real-time detection of the fault-induced transients is independently accomplished in each line end. Internal Fault detection: After the fault-induced transient detection, the method in a specic bus needs information from another one in order to identify either internal or external faults. Fault location: taking into account the traveling wave velocity, the fault is located according to the fault-inception time at each line end. When the proposed method is used at just one line end, only the fault-induced transients are detected.

Fig. 3.

Flowchart of the real-time fault detection and location.

w(k ) =
l=1

h(l)x(k + l L),

(1)

since {x(k -L+1),...,x(k -1),x(k )}; w are the wavelet coefcients of a signal x; h are the coefcients of the wavelet lter. The real-time simulator must be robust enough to solve the network equations and also compute the wavelet coefcients. III. R EAL -T IME FAULT D ETECTION AND L OCATION In order to develop a wavelet-based method for fault detection and location, the transmission power system shown in Fig. 2 was modeled and simulated by using the RTDS. The system is composed of two 500 kV, 400 km long transmission lines: a protected line from bus 1 to bus 2 and an unprotected line between buses 2 and 3. The line parameters were obtained from actual transmission lines belonging to the Chesf utility, a Brazilian power system.

A. Fault-Induced Transient Detection in Real-Time During the steady-state system operation the voltages and currents are usually composed of a fundamental power frequency component, some harmonics, and noises with low and high frequency components. It is well known that power system noises present normal probability distribution [17]. At steady-state operation, the wavelet coefcients of the MODWT, at the rst scale, are mainly inuenced by the high frequency noises of the signals (from 5 to 10 kHz). After an extensive statistical analysis of actual oscillographic data from Chesf power system, it was veried that the wavelet coefcients can be approximated by a function with normal probability distribution with parameters given by the magnitude average and standard deviation . Similar features of the wavelet coefcients of the DWT were observed by [18]. Fig. 4 depicts the phase current of an actual oscillographic record without transient disturbances. The wavelet coefcients at the rst scale and the related relative frequency histogram are also shown in Fig. 4. A normal probability density function with parameters given by the wavelet coefcient average and standard deviation was also obtained. The estimated density function and the density histogram match up quite well. In this case, the wavelet coefcients at the rst scale are mainly due to the high frequency noises of the current and can be approximated by a function with normal probability distribution.

Fig. 4. High frequency noise analysis during steady-state operation: (a) phase current; (b) wavelet coefcients at the rst scale; (c) density histogram of the wavelet coefcients.

According to the normal probability distribution function, about 99.73% of the wavelet coefcient values fall into the range [-3 , +3 ]. During steady-state operation, all wavelet coefcients of a signal (voltage or current) are expected to be inside the range [W1 , W2 ]=[-4 , +4 ] (steady-state thresholds). For each phase voltage and current, both W1 and W2 are computed in real-time considering the wavelet coefcients in the last cycle. The wavelet coefcient at the current simulation time step must present a value within the steady-state thresholds, i.e.: w(k )[W1 , W2 ]. When a line fault occurs, such as a fault due to an insulator ashover or fallen conductor, the abrupt change in voltage at the point of the fault generates a high frequency electromagnetic impulse called traveling wave which propagates along the line in both directions away from the fault point at speeds close to light velocity. As a consequence, fault-induced transients with low and high frequency components can be observed at each line end at a specic time after the fault inception. These transients can be properly detected by using the wavelet coefcients of voltages and currents: the wavelet coefcients regarding the fault-induced transients are higher than the ones related to steady-state. Fig. 5 depicts the phase A voltage and current (vA and iA ) at fault location, bus 1, and bus 2 (half cycle), regarding an AG fault with rf =1 , f =90o , and df =45 km from bus 1, simulated in real-time at the protected line of the power system shown in Fig. 2. The wavelet coefcients of the voltages and currents at buses 1 and 2 were also computed in real-time (Fig. 5). According to Fig. 5, kf is the sample related to the fault inception time. However, due to the traveling wave propagating time, the fault-induced transients at buses 1 and 2 were detected at samples kf 1 and kf 2 , respectively. The wavelet coefcients regarding these transients usually present very high values, mainly at kf 1 and kf 2 samples. Taking the wavelet coefcients at steady-state as reference, a fast increasing (or decreasing) of the wavelet coefcients at least in one of the voltages and currents allows the fault-induced transient

detection in real-time. In this case, the wavelet coefcient of a specic signal at the current simulation time step must present a value outside the steady-state thresholds. B. Fault Detection Fig. 6 depicts a protected line with both internal and external faults. The monitored currents at each line end are taken as reference to be owing into the protected line. In case of an internal fault (Fig. 6(a)), the traveling waves before any reection and refraction propagate against the monitored current directions. As a consequence, the rst wavelet coefcient peak of a current at both line ends presents the same polarities. With regard to external faults (Figs. 6(b) and (c)), the rst wavelet coefcient peak of a current at both line ends presents opposite polarities to each other.

Fig. 5. Fault-induced transients of an internal fault detected by using the wavelet coefcients of the MODWT.

Fig. 6. Traveling waves and monitored current polarities: (a) inside fault; (b) left outside fault; (c) right outside fault.

When a fault is within the protected line and it is not in the middle of the line, fault-induced transients will be detected at distinct times at line terminals (Fig. 5). When these transients are detected at one of the line ends, a communication link between the buses is used in order to send information to another one. In this paper, both the traveling wave propagation time to cross the line and the communication time between the buses are taken as the same (k samples). After both the fault-induced transient detection and communication between buses, the fault is taken as within the protected line whether the rst wavelet coefcient peak of a specic current at both line ends present the same polarities. For instance, in Fig. 5, the wavelet coefcient peak of phase A current decreases at fault inception time at each line end. Fig. 7 depicts the phase A voltage and current (vA and iA ) at bus 1, bus 2, and fault location (half cycle), regarding an AG fault with rf =14.9 , f =98.9o , and df =70 km from bus 3, simulated in real-time on unprotected line of the power system shown in Fig. 2. The wavelet coefcients of these signals at buses 1 and 2 were also computed in real-time and shown in Fig. 7. According to Fig. 7, the rst wavelet coefcient peak of iA at both line ends presented opposite polarities. In addition, the difference between the samples related to the fault-inception time at each line end is equivalent to the traveling wave propagation time to cross the line (|kf 1 -kf 2 |k ). C. Real-Time Fault Location How to dene the velocity of the travelling wave correctly is critical to the accuracy of fault location methods based on fault-induced transient or travelling wave analysis. Most of fault location schemes consider the propagation velocity of travelling waves as the velocity of light. However, it is well-known that on multi-conductor transmission lines the travelling waves propagate with different velocities in different modes [19].

Fig. 7. Fault-induced transients of an external fault detected by using the wavelet coefcients of the MODWT.

The propagation velocity of the travelling wave was computed by using the distributed parameter model of a long transmission line. In this paper, by using the distributed parameters shown in Fig. 2, the propagation velocity of the travelling wave was taken as 98% of the velocity of light. By using this velocity, the waves propagate L km in one time step of 50 s. Therefore, each sample from the fault inception time (sample kf ) to the fault inception time at monitored bus (samples kf 1 or kf 2 ) represents a fault away L15 km. According to Fig. 5, at bus 1, the fault-induced transients could be detected 3 samples after the fault inception time (kf 1 -kf =3: fault located about 45 km away from bus 1). On the other hand, at bus 2, the fault-induced transients could be detected 24 samples after the fault inception time (kf 2 -kf =24: fault located about 360 km away from bus 1). This fault was simulated on the protected line, 45 and 355 km away from buses 1 and 2, respectively. Taking into account the simulation time step t=50 s, the sample amount equivalent to the traveling wave propagation time to cross a 400 km transmission line is 27 samples (k =27 samples). According to Fig. 7, at bus 2, the faultinduced transients could be detected 22 samples after the fault inception time (kf 2 -kf =22: fault located about 330 km away from bus 2). On the other hand, at bus 1, the faultinduced transients could be detected 49 samples after the fault inception time (kf 1 -kf =49: fault located about 735 km away from bus 1). This fault was simulated upon an unprotected line, 330 and 730 km away from buses 2 and 1, respectively. Beyond the polarities of the rst wavelet coefcient peak of the currents in both buses, the comparison between the fault location estimation of both buses can also indicate whether a fault is internal the protected line.

According to [14], [13], the sample regarding the faultinduced transients at monitored buses (kf 1 and kf 2 ) can be faster and accurately detected by means of the wavelet coefcient analysis of the MODWT. However, the sample regarding the fault inception upon the transmission line kf can not be identied by means of voltage and current analysis at a specic bus. Fortunately, the fault location can be estimated through the samples kf 1 and kf 2 , as follows L t(kf j kf i ) v , (2) 2 where dij is the fault location estimation at bus i by using additional information from bus j ; L is the line length; v is the propagation velocity of the traveling wave; kf i and kf j are the samples regarding the fault-induced transient detection at buses i and j , respectively. In this paper, v =0.98c, where c is the velocity of the light. By using Eq. 2 with v =0.98c, the fault location estimation regarding the internal fault shown in Fig. 5 would be d12 =45.65 km from bus 1 and d21 =354.35 km from bus 2. In this case, an internal fault is identied. With regard to the external fault shown in Fig. 7, the fault location estimation would be d12 =398.45 km from bus 1 and d21 =1.55 km from bus 2. In this case, an external fault is identied. In this paper, an external fault is identied when dij L-L and dji L, or vice-versa. In addition, transients due to a switching operation either on protected line terminals or other power system equipment away from the protected line can be detected as external faults. In this paper, a communication link between the buses is used in order to detect an internal fault and estimate the fault location in each bus. The communication delay between the buses was taken as k =27 samples. When the fault-induced transients are detected at bus i, information regarding the polarities of the wavelet coefcients of the currents and the samples in which these transients were detected (kf i ) are sent to bus j . Therefore, the fault detection and location at bus j is carried out at sample klj =kf i +k . As analogy, the fault detection and location at bus i is carried out when information from bus j were received, at sample kli =kf j +k . Therefore, the bus further from the fault will detect and locate the fault faster. dij = IV. P ROPOSED M ETHOD P ERFORMANCE E VALUATION The transmission power system shown in Fig. 2 was modeled and simulated by using the RTDS. The proposed method was also modeled by using the RTDS at two distinctive power system points (buses 1 and 2) in order to detect and locate internal faults upon a protected line (Fig. 8). The transducers were modeled with typical parameters.

In order to evaluate the performance of the wavelet-based method for real-time fault detection and location, a total of 1000 faults on protected line was simulated in real-time, each one with random values of fault resistance rf {0, 50} , fault inception angle f {0o , 180o }, and fault location df {15, 385} km from bus 1. With regard to the fault type, 100 AG, BG, CG, AB, BC, AC, ABG, BCG, ACG, and ABC faults were simulated. The fault-clearing time in each simulation was about 75 ms (4.5 cycles of 60 Hz). The fault detection and location method was performed with both MODWT and DWT. The transients regarding the fault inception time at each line end were identied with a mean error of one sample (50 s) by means of the real-time MODWT and about two samples with the real-time DWT, due to the down-sampling process to compute the wavelet coefcients. Therefore, the faults were detected faster by using the MODWT. With regard to the fault detection, in both MODWT- and DWT-based methods, all faults were detected as internal faults after the communication of the fault locators. With regard to the fault location, the MODWT-based method provided a mean error of 0.65% and 0.63% at buses 1 and 2, respectively. By using the DWT-based method, mean errors of 1.05 and 1.02% were obtained at buses 1 and 2, respectively. The mean error at fault location for all types of faults is summarized in Tab. I. The fault location is not inuenced by the type of fault and provides better performance with the MODWT. A total of 1000 faults on unprotected line was also simulated in real-time, each one with random values of fault resistance rf {0, 50} , fault inception angle f {0o , 180o }, and fault location df {15, 385} km from bus 2. With regard to the fault type, 100 AG, BG, CG, AB, BC, AC, ABG, BCG, ACG, and ABC faults were simulated. The fault-clearing time in each simulation was about 75 ms (4.5 cycles of 60 Hz). In all cases, the fault-induced transients were successfully detected and the faults were detected as external faults by using both MODWT and DWT. In order to improve the obtained results, a sensitivity analysis of the effectiveness of the proposed wavelet-based fault detection and location scheme with reference to different mother wavelets have been performing and will be addressed in future works.
TABLE I P ERFORMANCE EVALUATION OF BOTH MODWT- AND DWT- BASED METHODS . Type of fault AG BG CG AB BC AC ABG BCG ACG ABC Amount of simulations 100 100 100 100 100 100 100 100 100 100 Mean error (%) Bus 1 Bus 2 MODWT DWT MODWT DWT 0.6452 0.6535 0.6047 0.6691 0.7002 0.7439 0.7147 0.5976 0.6032 0.6098 1.0656 1.0539 1.0470 1.0730 1.1040 1.0722 1.0802 1.0581 1.9443 1.0004 0.6143 0.6336 0.6570 0.6388 0.6590 0.5866 0.6836 0.5509 0.6139 0.6221 1.0556 1.0399 1.0453 0.9686 1.0451 0.9264 1.0098 1.0605 1.0094 1.0373

Fig. 8. Power system model with the proposed fault detection and location method.

The proposed fault location scheme can provide good performance with high sampling frequency. However, good results were obtained on real-time simulation of the RTDS by using a sampling frequency of 20 kHz (simulation time step of 50 s). Nowadays, this sampling frequency can be used by high-speed protective relays. The proposed MODWT-based analysis will be further implemented into a new directional relaying scheme in order to design a reliable method for highspeed fault clearance in protection systems. V. C ONCLUSIONS The real-time fault detection and location based on the wavelet coefcient analysis was very fast, performing both the high-speed fault detection and good fault location. this can make it a promising candidate for a protective relaying algorithm. In addition, the method seems to be little dependent on the conditions that would affect most of the other algorithms, such as line parameters and fault parameters (resistance, location, inception angle, and type). The real-time fault detection and location method was performed with both the discrete wavelet transform and the maximal overlap discrete wavelet transform. The fault-induced transients were detected faster and the faults were located accurately by using the maximal overlap discrete wavelet transform. Taking into account the sampling frequency of 20 kHz (value near the typical digital fault recorders), the obtained fault location accuracy (mean error of about 0.63% with the maximal overlap discrete wavelet transform) is quite remarkable. R EFERENCES
[1] G. W. Swift, The spectra of fault-induced transients, IEEE Transactions on Power Apparatus and Systems, vol. PAS-98, no. 3, pp. 940947, May 1979. [2] A. T. Johns, R. K. Aggarwal, and Z. Q. Bo, Non-unit protection technique for ehv transmission systems based on fault-generated noise, part 1: Signal measurement, IEE Proc. Gener. Transm. Distrib., vol. 141, no. 2, pp. 133140, March 1994. [3] Z. Bo, M. Redfern, and G. Weller, Positional protection of transmission line using fault generated high frequency transient signals, IEEE Transactions on Power Delivery, vol. 15, no. 3, pp. 888 894, Jul. 2000. [4] S. Santoso, E. J. Powers, W. M. Grady, and P. Hofmann, Power quality assessment via wavelet transform analysis, IEEE Transactions on Power Delivery, vol. 11, no. 2, pp. 924930, April 1996. [5] M. H. J. Bollen and I. Y.-H. Gu, Signal Processing of Power Quality Disturbances. New York, USA: IEEE, 2006. [6] W. A. Wikinson and M. D. Cox, Discrete wavelet analysis of power system transients, IEEE Transactions on Power Systems, vol. 11, no. 4, pp. 20382044, Nov 1996. [7] S. A. Probert and Y. H. Song, Detection and classication of high frequency transients using wavelet analysis, Proc. 2002 IEEE Power Engineering Society Summer Meeting, vol. 2, pp. 801806, Chicago, USA, July 2002. [8] F. B. Costa, B. A. Souza, and N. S. D. Brito, Detection and classication of transient disturbances in power systems, IEEJ Trans. PE, pp. 910916, Oct 2010. [9] O. A. S. Youssef, Fault classication based on wavelet transforms, Transmission and Distribution Conference and Exposition, vol. 1, pp. 531536, Nov 2001. [10] K. M. Silva, B. A. Souza, and N. S. D. Brito, Fault detection and classication in transmission lines based on wavelet transform and ann, IEEE Transactions on Power Delivery, vol. 21, no. 4, pp. 20582063, Oct 2006.

[11] D.-J. Zhang, Q. H. Wu, Z. Q. Bo, and B. Caunce, Transient positional protection of transmission lines using complex wavelets analysis, IEEE TRANSACTIONS ON POWER DELIVERY, vol. 18, no. 3, pp. 705710, jul. 2003. [12] S. P. Valsan and K. Swarup, Wavelet transform based digital protection for transmission lines, International Journal of Electrical Power & Energy Systems, vol. 31, no. 7-8, pp. 379388, 2009. [13] F. B. Costa, B. A. Souza, and N. S. D. Brito, Real-time detection of fault-induced transients in transmission lines, IET Electronics Letters, pp. 753755, May 2010. [14] F. B. Costa, Uma t ecnica de diagn ostico em tempo real de dist urbios transit orios baseada na transformada wavelet para uso em registradores o, Ph.D. dissertation, Federal University of Campdigitais de perturbac a ina Grande, Campina Grande, Brazil, 2010. [15] D. B. Percival and A. T. Walden, Wevelet Methods for Time Series Analysis. New York, USA: Cambridge University Press., 2000. [16] L. Peretto, R. Sasdelli, and R. Tinarelli, On uncertainty in waveletbased signal analysis, Instrumentation and Measurement, IEEE Transactions on, vol. 54, no. 4, pp. 1593 1599, 2005. [17] W. D. Stevenson JR. and J. J. Grainger, Power System Analysis. New York, USA: McGraw-Hill, 1994. [18] A. C. Parsons, W. M. grady, and E. J. Powers, A wavelet-based procedure for automatically determining the beginning and end of transmission system voltage sags, IEEE Power Engineering Society 1999 Winter Meeting, vol. 2, pp. 13101315, Feb. 1999. [19] J. D. Glover and M. S. Sarma, Power System Analysis and Design. New York: Thomson-Engineering publisher, 2002.

F. B. Costa received his BS, MS, and PhD in Electrical Engineering from Federal University of Campina Grande (UFCG), Brazil, in 2005, 2006, and 2010, respectively. He works currently as a professor at Federal University of Rio Grande do Norte - School of Science and Technology, Brazil. He received an honorable mention from the 2010 Brazilian Award of Engineering for his Ph.D. dissertation on fault and power quality problems. His research interest are electromagnetic transients, power quality, fault diagnosis, and power system protection.

B. A. Souza received his PhD in Electrical Engineering from Federal University of Para ba, Brazil, in 1995. He currently works as a professor at the Department of Electrical Engineering of Federal University of Campina Grande, Brazil. His main research activities are on optimization methods applied to power systems, electromagnetic transients, power quality and fault diagnostic.