Overview of Methods for Voltage Sag Performance

Estimation
Ivan B. N. C. Cruz, Athena P. Lavega, and Jordan R. C. Orillaza
Electrical and Electronics Engineering Institute
University of the Philippines Diliman
Quezon City, Philippines
[email protected]

Abstract—Voltage sags may cause malfunction of customer

equipment and interruption of customer processes. Voltage sag
performance should be quantified to incite informed decisions
from utilities, customers, and regulators. This paper describes
existing voltage sag performance estimation methods and
classifies these as either model- or measurement-based,
depending on the data required for implementation. Model-
based performance estimation is classified either as mean
performance estimation or Monte Carlo approach. The methods
of fault positions and critical distances, as well as derivatives of
these, fall under the first group, while Monte Carlo techniques
applied either to short-circuit calculations or EMTP models fall
under the second group. Measurement-based performance
estimation includes regression based on site and system
characteristics, and conventional state estimation formulation.

Index Terms—Power quality; voltage sags; state estimation.

I. INTRODUCTION
Voltage sag is defined as a short-duration condition Figure 1. Classification of voltage sag performance estimation methods.
wherein the rms voltage is between 0.1 p.u. and 0.9 p.u lasting
for 0.5 cycles to 1 minute [1]. Typical causes of voltage sags A previous paper focused on describing and classifying
are high currents or overcurrents, such as during faults, existing methods for voltage sag profile estimation [5]. Profile
switching of heavy loads, starting of large motors, and estimation methods were classified into either model- or
temporary disconnection of supply in fault-clearing switching measurement-based, depending on the information required
operations [1-3]. for implementation. Model-based methods do not require
voltage measurements and require only information to develop
From customer viewpoint, voltage sags may cause
system models, while measurement-based methods definitely
customer equipment to malfunction and may interrupt
require voltage measurements, but may or may not require
customer processes [1-2]. Utilities may measure voltage sag
system models. Fig. 1 shows the classification of methods for
performance in one of two ways: (1) statistical analysis of
voltage performance estimation reviewed in this paper; the
historical performance, and (2) probabilistic assessment or
same classification scheme is applied to voltage sag
performance estimation [2]. The first option provides typically
performance estimation methods.
more accurate assessment, but is expensive and takes a long
time to implement [2]. It may not form a useful basis for The contents of this paper are as follows. Section II
predicting future performance, thereby making it unsuitable discusses model-based performance estimation methods while
for planning purposes. Section III deals with measurement-based methods.
Connections with the ideas in the previous paper are discussed
Voltage sag state estimation may be further classified into
in Section IV. Finally, Section V contains the conclusions.
two types [4]: (1) sag profile estimation, wherein voltages at
unmonitored buses during a single fault event are estimated,
and (2) sag performance estimation, wherein sag performance
at unmonitored buses and/or of the system are estimated.

This work was supported by the Engineering Research and Development

for Technology (ERDT) Program of the Department of Science and
Technology (DOST) of the Republic of the Philippines.

978-1-5090-3792-6/16/$31.00 ©2016 IEEE

508
 II. MODEL-BASED PERFORMANCE ESTIMATION METHODS The method of critical distances has been applied to
systems with balanced faults [17-19] and systems with single-
A. Mean Performance Estimation phase and phase-to-phase faults [18, 20]. For cases where the
The methods that fall under this group all have similar X/R ratio between source and feeder differ by a significant
framework for estimating the voltage sag performance of a amount, the exact expression for computing the critical
system. These methods all involve determining the exposed distances has been provided [20]. A comparison of the method
area for the buses of interest for a given sag depth, and using of critical distances and the method of fault positions was also
the failure rates of components within the exposed area to performed [7].
arrive at the sag frequency. Very few of the methods under
this group particularly deal with sag duration. 3) Method of Analytical Expressions
This last group of methods under model-based methods
These methods may be further subdivided into: (1) method use analytical expressions derived from sequence bus
of fault positions, (2) method of critical distances, and (3) impedance matrices [21-23] or voltage sags matrices [24-25],
method of analytical expressions, such as bus impedance to determine the exposed area.
matrices and voltage sags matrices.
Under these methods, the following steps are
1) Method of Fault Positions implemented:
Under the method of fault positions, the following steps are
1. Build the analytical expressions describing the system
implemented [2]:
under consideration.
1. Choose the fault positions.
2. Determine the critical distances from these expressions.
2. Simulate faults at the fault positions.
3. Determine the exposed area for the buses of interest.
3. Determine the exposed area for the buses of interest.
4. Determine the failure rates of components within the
4. Determine the failure rates of components within the exposed area to arrive at the sag performance estimate.
exposed area to arrive at the sag performance estimate.
Analytical expressions based on the bus impedance matrix
Ref. [6] was the first one to propose the method of fault were first used to obtain voltage sag estimates for a balanced
positions for estimating voltage sag performance. The transmission system [21]. Similarly, the method was applied
disadvantage of the method is that the result depends on the to balanced and unbalanced systems by generating bus
choice and number of fault positions. More fault positions impedance matrices for each sequence component [22], and in
usually lead to higher accuracy, but will also require greater a mixed transmission and distribution system [23].
computational effort. Different ways of choosing the fault
Instead of the bus impedance matrix, the voltage sags
positions have been implemented [7-10].
matrix was used [24-25]. The secant method for quadratic
The fault positions method was applied to compare it with interpolation was used instead of analytical expressions to
monitoring [11]. The effect of different fault distribution on determine the critical points along the line [26-27].
the results of the method was investigated [10, 12-13].
Original authors of some of the previous work involved
Protection coordination was considered [14-15]. Complete
associated their work either with the method of fault positions
matrix formulations for the method have been provided for
[22] or the method of critical distances [23]. However, these
balanced systems [16].
methods do not explicitly choose and use fault positions and
2) Method of Critical Distances do not use the voltage divider model in determining the
The method of critical distances uses the voltage divide exposed areas. It is believed that this method significantly
model to determine critical distances, the distance from the evolved from the first two methods and warrants a separate
bus of interest (that is, the point of common coupling or PCC) category.
within which faults will cause voltage to drop below a certain
level (defined the critical voltage) [17]. Under this method, the B. Monte Carlo Approach
following steps are implemented [2]: Monte Carlo simulations refer to algorithms that perform
repeated simulations, wherein parameters that are probabilistic
1. Determine the PCC for each bus and each fault.
in nature are assigned randomly generated values in every
2. Use the voltage divider model to determine the critical simulation [28]. Monte Carlo simulations have been applied to
distances from each PCC. voltage sag performance estimation using either short-circuit
calculations [29-34] or EMTP models [35-38]. A study
3. Determine the exposed area for the buses of interest. comparing the voltage sags indices calculated using short-
4. Determine the failure rates of components within the circuit calculations and using EMTP models concluded that
exposed area to arrive at the sag performance estimate. there was no significant difference between the two; the
difference in results were attributed to whether the pre-fault
The advantage of this method is that it does not suffer and loading conditions were considered or not [39].
from the requirement of the fault position and requires less
information for similar or higher accuracy, but it is applicable
mostly to radial systems or systems with simple loops [17].

509
 Under this group of methods, the probability distributions lightning ground flash density in flashes/km2-yr, and the term
of the following will have significant impact on the results correlated to the concept of critical distances [41].
[29]: (1) fault characteristics such as location, type, duration,
 A   xrural 
T
and impedance; (2) fault incidence instant, that is, when does (2)
the fault occur with respect to the voltage/current sinusoid; (3)     n ⋅ kV 2
N10 =  B   xsuburban  + D ⋅ l3 + E ⋅ N g + F ⋅ f
protective response times of the system such as fault-clearing  C   x  MVAxfmr
 urban 
times and/or reclosing times.
The first group of model-based methods estimates the This method does not require the data to build system
long-term mean performance while the Monte Carlo methods models. However, it does require large amounts of historical
provides the variability of these estimates [31]. For example, data—not just with respect to voltage measurements but also
while the first group uses discrete numbers for the failure rates with respect to events (such as faults and lightning) and
to estimate the voltage sag frequency, Monte Carlo methods locations (such as urban/rural classification and tree coverage
take into account the probability distribution of the failure characteristics). Furthermore, while the regression technique
rate. The Monte Carlo methods also more easily allow and resulting equations are valid, the coefficients will apply
analysis that considers other variables subject to variability. only to the particular utility from which the data was gathered;
The disadvantage of the methods using the Monte Carlo there is no guarantee that the coefficients obtained may be
approach is that they require large amounts of memory and applied to another utility.
computation efforts.
B. Classical State Estimation Formulation
III. MEASUREMENT-BASED PERFORMANCE ESTIMATION The classical state estimation formulation uses the
METHODS following equation [42]:
Measurement-based performance estimation methods H=MX+E (3)
require measurements in order to be used, and may require
system models as well. These methods fall under one of two where H is the measurement vector, M is the measurement
groups: (1) regression based on site characteristics, and (2) matrix, X is the state vector being estimated, and E is the
conventional state estimation formulation. The first group measurement noise vector. In their formulation, X is used to
does not require system model data but require more general represent the number of faults occurring along the line
data about the system while the second group requires system segments, H is used to represent the number of sags recorded
model data. by the meters, M is built using analytical expressions, while E
is neglected [42]. The matrix M is binary, depending on
A. Regression Based on Site Characteristics whether a fault on a line segment results in a retained voltage
below a threshold at a point of interest. The values of matrix
While the relationship between lightning and faults was
M may be computed using short-circuit calculations to
known, Ref. [40] was likely one of the first to suggest
determine the voltage at the points of interest.
correlating frequency of lightning events with voltage sag
performance. Using data obtained from an actual power Eq. (1) is initially developed analytically for several
quality survey, possible site characteristics that affect the thresholds of voltage sag. These equations collectively may be
power quality of a power system were investigated [41]. expressed as one system of equations, which is similar to (1).
Regression techniques were used to determine what factors The actual measurements from the meters are then used to
would be significant predictors. These predictors are then used compute for the measurement vector (H) before estimating the
to obtain the coefficients of a linear model of the number of number of voltage sags (X) [42].
voltage sags events.
The classical formulation of state estimation applied to
The regression equation (1) that best predicted SARFIITIC voltage sag performance estimation was solved using linear
consisted of a linear combination of total exposure length of integer programming [42], neural networks [43], genetic
the circuit (both three- and single-phase portions), the algorithm [44], and singular value decomposition [45].
lightning ground flash density in flashes/km2-yr, and a term
correlated to the concept of critical distances [41]. This last IV. PROFILE AND PERFORMANCE ESTIMATION METHODS
term was based on the total number of feeders coming off of Some voltage sag state estimation procedures stop once the
the substation bus, the base voltage, and the station voltage sag profile has been obtained. Other methods, such as
transformer MVA rating. An additional constant was added in
the method of fault positions and the methods under the
case tree coverage was moderate to heavy.
Monte Carlo approach, require the voltage sag profile to
n f ⋅ kV 2 estimate the voltage sag performance. Other methods, such as
N ITIC = A + B ⋅ l + C ⋅ N g + D ⋅ + E⋅x (1) the regression approach, the classical state estimation, the
MVAxfmr method of critical distances and analytical expressions,
directly estimated voltage sag performance without obtaining
In contrast, the regression model (2) that best predicted the voltage sag profile estimates.
number of interruption events (that is, the number of events
wherein residual voltage is at most 0.1 p.u.) consisted of a
constant dependent on whether the area was rural, suburban or
urban, the exposure length of the three-phase circuit, the

510
 V. CONCLUSIONS diagnosis,” Przeglad Elektrotechniczny (Electrical Review), vol. 89, no. 8,
pp. 143-149, 2013.
Quantifying the frequency and severity of voltage sags is [4] G. Olguin, “Voltage Dip (Sag) Estimation in Power Systems based on
important for customers, utilities and regulators. Various Stochastic Assessment and Optimal Monitoring,” Ph.D. dissertation,
methods have been presented in the literature: many require Chalmers University of Technology, Sweden, 2005.
[5] I. B. N. C. Cruz, A. P. Lavega, and J. R. C. Orillaza, “Overview of
system models, others use measurements, some methods use Voltage Sag Profile Estimation,” in 2015 IEEE Region 10 Conf.
both models and measurements. Voltage sag severity takes (TENCON 2015), Macau, 2–4 Nov. 2015.
into account voltage sag depth (which is addressed by all [6- [6] L. Conrad, K. Little, and C. Grigg, “Predicting and preventing
42]) and duration (which is significantly considered only by problems associated with fault-clearing voltage dips,” IEEE Trans. Ind.
some [6, 7, 14-15, 19, 29, 34-38]). Appl., vol. 27, no. 1, pp. 167-172, Feb. 1991.
[7] M. R. Qader, M. H. J. Bollen, and R. N. Allan, “Stochastic prediction
This paper described and classified different methods for of voltage sags in a large transmission system,” IEEE Trans. Ind. Appl.,
voltage sag performance estimation. These methods were vol. 35, no. 1, pp. 152-162, Jan./Feb. 1999.
[8] S. P. J. Rombouts and L. Driessen-Mutters, “Calculation of sag
classified into: (A) model- and (B) measurement-based
quantities including protection settings and breaker failure in voltage
methods. This classification allows the sag performance sag prediction,” in Proc. 15th Int. Conf. & Exhibition on Electricity
estimation method to be chosen based on data and Distribution (CIRED’99), Nice, France, 1–4 Jun. 1999, pp. [6].
measurements that are available. [9] V. R. C. Fonseca and M. F. Alves, “A dedicated software for voltage
sag stochastic estimate,” in Proc. 2002 10th Int. Conf. on Harmonics
Under model-based sag performance estimation methods and Quality of Power (ICHQP’02), Rio de Janeiro, Brazil, 6–9 Oct.
are the following group of methods: (1) mean performance 2002, vol. 1, pp. 99-104.
estimation and (2) Monte Carlo approach. Based on the [10] M. T. Aung, J. V. Milanovic, and C. P. Gupta, “Propagation of
starting point for the calculations, the first group may be asymmetrical sags and the influence of boundary crossing lines on
voltage sag prediction,” IEEE Trans. Power Del., vol. 19, no. 4, pp.
further subdivided into: (i) the method of fault positions, (ii) 1819–1827, Oct. 2004.
method of critical distances, and (iii) analytical expressions. [11] M. H. J. Bollen, M. R. Qader, and R. N. Allan, “Stochastical and
Based on the models used, the second group may be further statistical assessment of voltage dips,” in IEE Colloq. on Tools and
subdivided into: (i) frequency-domain short-circuit analysis Techniques for Dealing with Uncertainty (Digest No. 1998/200),
and (ii) time-domain EMTP models. London, UK, 27 Jan. 1998, pp. 5/1-5/4.
[12] J. V. Milanovic, M. T. Aung, and C. P. Gupta, “The influence of fault
It can be seen that while fault positions and critical distribution on stochastic prediction of voltage sags,” IEEE Trans.
distances have traditionally comprised model-based methods, Power Del., vol. 20, no. 1, pp. 278-285, Jan. 2005.
[13] B. Q. Khanh, D.-J. Won, S.-I. Moon, “Fault distribution modeling
there are significant developments in using analytical using stochastic bivariate models for prediction of voltage sag in
expressions as well as Monte Carlo techniques to obtain distribution systems,” IEEE Trans. Power Del., vol. 23, no. 1, pp. 347-
performance estimates. 354, Jan. 2008.
[14] M. T. Aung and J. V. Milanovic, “Stochastic prediction of voltage sags
On the other hand, under measurement-based sag by considering the probability of failure of the protection system,”
performance estimation methods are the following methods: IEEE Trans. Power Del., vol. 21, no. 1, pp. 322-329, Jan. 2006.
(1) regression based on site and system characteristics and (2) [15] L. Chen and X. Xiao, “Voltage sags frequency assessment considering
classical state estimation formulation. Based on the solution the coordination of protection system,” in Proc. 2012 15th Int. Conf. on
Harmonics and Quality of Power (ICHQP’12), Hong Kong, 17–20 Jun.
scheme used, the second group may be further subdivided into 2012, pp. 932-935.
(i) linear programming, (ii) neural networks, (iii) genetic [16] G. Carpinelli, P. Caramia, C. Di Perna, P. Varilone, and P. Verde,
algorithms, and (iv) singular value decomposition. “Complete matrix formulation of fault-position method for voltage-dip
characterisation,” IET Generation, Transmission & Distribution, vol. 1,
Model-based methods consider voltage sags primarily no. 1, pp. 56-64, Jan. 2007.
caused by faults. Voltage sags due to lightning were quantified [17] M. H. J. Bollen, “Fast assessment methods for voltage sags in
only using measurement-based methods. Voltage sags due to distribution systems,” IEEE Trans. Ind. Appl., vol. 32, no. 6, pp. 1414-
motor starting may be estimated using either type of method, 1432, Nov./Dec. 1996.
[18] M. H. J. Bollen, “Method of critical distances for stochastic assessment
but are hardly addressed in the literature, presumably because of voltage sags,” in IEE Proc. — Generation, Transmission and
effect is localized. Distribution, Jan. 1998, vol. 145, no. 1, pp. 70-76.
[19] M. H. J. Bollen, T. Tayjasanant, and G. Yalcinkaya, “Assessment of the
For actual systems, once estimates for sag profile and sag number of voltage sags experienced by a large industrial customer,”
performance have been obtained, validation of simulation IEEE Trans. Ind. Appl., vol. 33, no. 6, pp. 1465-1471, Nov./Dec. 1997.
results with actual monitored data should be performed [46] [20] M. H. J. Bollen, “Additions to the method of critical distances for
and comprehensive PQ study should take into account stochastic assessment of voltage sags,” in IEEE Power Eng. Soc. 1999
expected equipment sensitivity such as immunity levels and Winter Meeting, 31 Jan.–4 Feb. 1999, vol. 2, pp. 1241-1246.
[21] Y.S. Lim and G. Strbac, “Analytical approach to probabilistic
performance curves [47]. prediction of voltage sags on transmission networks,” IEE Proc. –
Generation, Transmission, Distribution 2002, vol. 149, no. 1, pp. 7-14,
REFERENCES Jan. 2002.
[22] E. Espiñoza and A. Hernandez, “An analytical approach for stochastic
[1] IEEE 1159-2009: Recommended Practice for Monitoring Electric assessment of balanced and unbalanced voltage sags in large systems,”
Power Quality. IEEE, 26 Jun. 2009. IEEE Trans. Power Del., vol. 21, no. 3, pp. 1493-1500, Jul. 2006.
[2] M. H. J. Bollen, Understanding Power Quality Problems—Voltage [23] M. N. Moschakis and N. D. Hatziargyriou, “Analytical calculation and
Sags and Interruptions. Piscataway, NJ: IEEE Press, 2000. stochastic assessment of voltage sags,” IEEE Trans. Power Del., vol.
[3] A. Kazemi, A. Mohamed, H. Shareef, and H. Zayandehroodi, “Review 21, no. 3, pp. 1727-1734, Jul. 2006.
of voltage sag source identification methods for power quality

511
[24] S. Quaia and F. Tosato, “A method for analytical voltage sags [36] J. A. Martinez and J. Martin-Arnedo, “Voltage sag studies in
prediction,” in 2003 IEEE Bologna PowerTech Conf., Bologna, Italy, distribution networks part I: system modeling,” IEEE Trans. Power
23–26 June 2003, pp. [6]. Del., vol. 21, no. 3, pp. 1670-1678, Jul. 2006.
[25] S. Quaia and F. Tosato, “The voltage sags matrix: a useful tool for [37] J. A. Martinez and J. Martin-Arnedo, “Voltage sag studies in
power quality analysis,” Elektrotehniski Vestnik (Electrotechnical distribution networks part II: voltage sag assessment,” IEEE Trans.
Review), vol. 69, pp. 171-174, 2002. Power Del., vol. 21, no. 3, pp. 1679-1688, Jul. 2006.
[26] C.-H. Park and G. Jang, “Fast method to determine an area of [38] J. A. Martinez and J. Martin-Arnedo, “Voltage sag studies in
vulnerability for stochastic prediction of voltage sags,” in Proc. Inst. distribution networks part III: voltage sag index calculation,” IEEE
Elect. Eng. – Gen. Trans. Dist., vol. 152, no. 6, pp. 819-827, Nov. Trans. Power Del., vol. 21, no. 3, pp. 1689-1697, Jul. 2006.
2005. [39] J. M. C. Filho, R. C. Leborgne, P. M. Silveira, and M. H. J. Bollen,
[27] C.-H. Park and G. Jang, “Stochastic estimation of voltage sags in a “Voltage sag index calculation: comparison between time-domain
large meshed network,” IEEE Trans. Power Del., vol. 22, no. 3, pp. simulation and short-circuit calculation,” Electric Power Systems
1655-1664, Jul. 2007. Research, vol. 78, pp. 676-682, 2007.
[28] W. Li, Risk Assessment of Power Systems: Models, Methods, and [40] M. F. McGranaghan, D. R. Mueller, and M. J. Samotyj, “Voltage sags
Applications, 2nd ed., IEEE: John Wiley & Sons, Inc., 2014. in industrial systems,” IEEE Trans. Ind. Appl., vol. 29, no. 2, pp. 397-
[29] S. O. Faried and S. Aboreshaid, “Stochastic evaluation of voltage sags 403, Mar./Apr. 1993.
in series capacitor compensated radial distribution systems,” IEEE [41] T. A. Short, A. Mansoor, W. Sunderman, and A. Sundaram, “Site
Trans. Power Del., vol. 18, no. 3, pp. 744-750, Jul. 2003. variation and prediction of power quality,” IEEE Trans. Power Del.,
[30] S. O. Faried, R. Billinton, and S. Aboreshaid, “Stochastic evaluation of vol. 18, no. 4, pp. 1369-1375, Oct. 2003.
voltage sag and unbalance in transmission systems,” IEEE Trans. [42] E. Espiñosa-Juarez and A. Hernandez, “A method for voltage sag state
Power Del., vol. 20, no. 4, pp. 2631-2637, Oct. 2005. estimation in power systems,” IEEE Trans. Power Del., vol. 22, no. 4
[31] G. Olguin, M. Aedo, M. Arias, and A. Ortiz, “A Monte Carlo pp. 2517-2526, Oct. 2007.
simulation approach to the method of fault positions for stochastic [43] E. Espiñosa-Juarez, J. R. Espiñoza-Tinoco, and A. Hernandez, “Neural
assessment of voltage dips (sags),” in 2005 IEEE/PES Transmission networks applied to solve the voltage sag state estimation problem: an
and Distribution Conf. and Exhibition: Asia and Pacific, Dalian, China, approach based on the fault positions concept,” in the 2009 Electronics,
2005, pp. 1-6. Robotics, and Automotive Mechanics Conf. (CERMA 2009), Morelos,
[32] G. Olguin, D. Karlsson, and R. Leborgne, “Stochastic assessment of Mexico, 2009, pp. 88–93.
voltage dips (sags): the method of fault positions versus a Monte Carlo [44] J. Lucio, E. Espiñosa-Juarez, and A. Hernandez, “Voltage sag state
simulation approach,” in Proc. 2005 IEEE Russia Power Tech, 27-30 estimation in power systems applying genetic algorithms,” IET
Jun. 2005, St. Petersburg, Russia, pp. 1-7. Generation, Transmission & Distribution, vol. 5, no. 2, pp. 223-230,
[33] X. Zambrano, A. Hernandez, R. M. de Castro, and M. Izzeddine, 2011.
“Comparison of voltage sags prediction methods in power networks,” [45] A. Hernandez, E. Espiñosa-Juarez, R. M. de Castro, and M. Izzeddine,
in 2015 IEEE 15th Int. Conf. on Environment and Elec. Eng., Rome, “SVD applied to voltage sag state estimation,” IEEE Trans. Power
Italy, 10–13 Jun. 2015, pp. 89-94. Del., vol. 28, no. 2, pp. 866-874, Apr. 2013.
[34] M. Wamundsson and M. H. J. Bollen, “Predicting the number of [46] T. C. Oliveira, J. M. C. Filho, R. C. Leborgne, and M. H. J. Bollen,
voltage dips as a function of both voltage and duration,” in Proc. 2008 “Voltage sags: validating short-term monitoring by using long-term
13th Int. Conf. on Harmonics and Quality of Power (ICHQP’08), stochastic simulation,” IEEE Trans. Power Del., vol. 24, no. 3, pp.
Wollongong, New South Wales, 28 Sep.–1 Oct. 2008, pp. 1-6. 1345-1351, Jul. 2009.
[35] J. A. Martinez and J. Martin-Arnedo, “Voltage sag stochastic prediction [47] IEEE Std 1668-2014: IEEE Trial-Use Recommended Practice for
using an electromagnetic transients program,” IEEE Trans. Power Del., Voltage Sag and Short Interruption Ride-Through Testing for End-Use
vol. 19, no. 4, pp. 1975-1982, Oct. 2004. Electrical Equipment Rated Less than 1000 V. IEEE, 22 Sep. 2014.

512