This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

Distribution Systems High Impedance Fault

Location: a Parameter Estimation Approach
Leonardo U. Iurinic Student Member IEEE, A. R. Herrera-Orozco, Renato G. Ferraz, Member IEEE,
Arturo S. Bretas, Senior Member IEEE
Abstract—High impedance fault location has always been a arcing faults detection and location must be done in a short
challenge for protection engineering. On the other hand, if this window of time, a limitation that is not applied to HIF
task is successfully realized maintenance action could be detection and location.
performed in order to avoid potential injuries. For an effective
The detection of HIF and arcing faults presents great
protection scheme the high impedance fault location should be
performed, but a lack of research on this area is noted. This challenges. Many works have been realized in this subject, as
paper proposes a new analytical formulation for high impedance [3], [6]–[11]. For this purpose, the estimation of the fault
fault location in power systems. The approach is developed in location is the next natural step after detection. Conversely,
time domain, considering a high impedance fault model comparatively less research studies have been conducted to
composed by two antiparallel diodes. Using this model, the fault estimate the HIF location.
distance and parameters are estimated as a minimization
In a broad sense, some basic research lines can be identified
problem. Firstly, a linear least square based estimator is applied
without consideration of line capacitance. Secondly, a steepest in the technical literature. The first consists on simultaneous
descent based estimator is proposed in order to consider the line HIF detection and faulted point direction estimation [8]–[10].
capacitance. Studies were carried out with the IEEE 13 bus This approach requires for accurate fault location directional-
modeled in ATP. Encouraging test results are found indicating detection devices installed and synchronized on the power
the method’s potential for real life applications. network. Other approaches try to estimate the fault location
using remote power system measurements. These works
Index Terms— High Impedance Fault, Fault Location,
Parameter Estimation.
proposed to use knowledge-based techniques to solve the HIF
location problem [12], [13]. However, these techniques need a
I. INTRODUCTION great reliable database of simulation and real cases. A third
research line can be also mentioned. This consists on a hybrid
A particular and worrisome case of power systems faults
are the High Impedance Faults (HIF). These faults are
usually caused by an unwanted contact of a bare-energized
impedance and high frequency based fault location approach
[14]. This approach shown good results but has not been tested
for HIF cases.
conductor with a surface that significantly restricts the flow of Analyzing the technical literature, none of the approaches
fault current to a level below the detectable by conventional presented for HIF location can be defined as best, rather
overcurrent devices [1]. The most common causes of such which, each one complements the other and their
faults in Power Distribution Systems (PDS) are broken implementation depends on the available resources. On the
conductors contact with poorly grounded objects, like trees, other hand, currently a lack of consistent mathematical
vehicles and wood fences. A HIF does not behave as a linear formulations for HIF location is perceived, mainly because a
resistance, as classical fault location methods assume, instead HIF model is not included in the equation derivation. In this
it tends to present a nonlinear behavior of voltage as a function context, the present work develops a formulation for fault
of current due to existence of electric arcs and modification in location that includes a HIF model [2]. The final result can be
surface conditions [2]. classified as a contribution for the directional-detection
At this point, it is important to refer to the non-permanent approach with emphasis in primary distribution systems,
arcing faults, whose features are close to those of HIF [3], [4]. providing fault distance estimation for detected HIF. For this
Arcing faults also maintain a nonlinear characteristic between purpose, studies as [3]–[5], [7] have been taken as primordial
voltage and current, but producing detectable fault current references. Still, in comparison with previous works, the
levels. Thus, solutions given for the detection and location of proposed approach considers a more complete HIF model, line
arcing faults could contribute in solving the issues concerned capacitance and equation derivation is made in phase-
with the HIF. Contrary to HIF, arcing faults tends to be self- components, allowing the consideration of unbalanced
cleared by the automatic switchgear action [4], [5]. Hence, operation.
The remainder of this paper is as follows. Section II
L. U. Iurinic and A. R. Herrera-Orozco are with the Universidade Federal
do Rio Grande do Sul, Porto Alegre RS 90035-190, Brazil (e-mail: describes the proposed analytical methodology. Section III
[email protected], [email protected]). describes methodology implementation aspects. A case study
R. G. Ferraz is with the Universidade de Caxias do Sul, Caxias do Sul RS is presented on Section IV. Section V presents the conclusions
95070-560, Brazil (e-mail: [email protected]).
A. S. Bretas is with the University of Florida, Gainesville, FL 32611-6200, of this work.
USA (e-mail: [email protected]).

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

II. ESTIMATION OF FAULT PARAMETERS TOGETHER WITH THE and line parameters can be derived:
FAULT DISTANCE v1  x R i1  xC v1   x L i1  xC v1  vF . (3)
The HIF model proposed by [2] or the one presented in [15] Line parameters can be calculated from the line topology or
can be represented by a mathematical model similar to the can be estimated by means of measurements [16], but the
arcing fault model of [5] or [7]. In this context, the use of a voltage at point F is also needed and it is not straightforward
Least Square Estimation (LSE) approach to estimate the HIF to obtain. If both voltage and current measurements at
parameters together with the fault distance is proposed. terminal 2 are available, vF can be estimated by an equation
Consider the single-phase line system illustrated in Fig. 1. similar to (3) [17]. On the other hand, the estimation of vF can
This is a radial system fed at node 1. Upstream system from be made indirectly by means of a fault model. In traditional
node 1 is modeled by a Thévenin source and its equivalent one-terminal fault location methods this model consists on a
impedance. An equivalent load at node 2 represents the linear resistor connected to ground or between faulted phases
downstream system. Still, a phase to ground fault at point F is [18]. Although this assumption is widely accepted in the
considered. The line model considered between node 1 and the research community, it will produce unacceptable results if a
faulted point is illustrated in the same figure, where R, L and C HIF case is analyzed. For this reason, a HIF model must
represent the per-unit-length line resistance, inductance and replace the classical linear resistor model in order to develop
shunt capacitance. To simplify the analysis it is assumed that an adequate mathematical approach for fault location.
all the capacitance of the line is concentrated at node 1. All Therefore, the anti-parallel HIF model illustrated in Fig. 2,
equations are initially developed for the single-phase line proposed by [2] and analysed in [15] is considered. Such
model. After, they will be extended for a three-phase system, model yields currents and voltage-current characteristics that
considering single-phase to ground faults. In this work, the are similar to field test results. This can be seen when voltage
phase domain representation is chosen instead of the and current waveforms generated by the model of Fig. 2 are
symmetrical components one. This selection aims formulation compared with those presented in [17] or [18]. This model
generalization, as it allows unbalanced system operation consideration provides a path to solve equation (3), aiming
consideration. fault distance and parameters estimation.
Bearing in mind the faulted line model of Fig. 1 and that Considering the HIF model illustrated in Fig. 2, the fault
voltage and current samples are measured at terminal 1, an voltage can be expressed as a function of the fault current:
vF  RF iF  LF iF  VFp sgp iF   VFn sgn iF  ,
expression that relates the measured quantities as a function of (4)
the fault distance at any time instant can be developed.
Consider initially the Kirchhoff's voltage law applied to the where RF and LF are the resistive and the inductive parts of the
system represented on Fig. 1: HIF model. The parameter VFp is the positive arc voltage,
v1  x  Ri1F  Li1F
   vF . (1) present when iF is in its positive semi-cycle and, VFn is the
negative arc voltage, present when iF is in its negative semi-
In (1), x is the fault distance, i1F is the current that flows to the cycle. These two arcing voltages sources cannot exist at the
fault through the series branch of the line model and vF is the same moment, and they are turned on and turned off using the
fault voltage. An apostrophe is used to indicate the first following functions:
derivative in function of time and two apostrophes indicate the
1, iF  0
second derivative. In order to express (1) as a function of the sgp  iF    . (5)
measured current at terminal 1, the effect of the shunt 0, iF  0
capacitance must be considered:  0, iF  0
sgn  iF    . (6)
i1F  i1  xC v1. (2)
1, iF  0
Finally, replacing (2) into (1), an equation that relates the By the replacement of (4) into (3), the fault voltage is
measured voltage and current at node 1 with the fault distance embedded in the HIF model, and (3) becomes:
v1  x  R i1  Li1   RF iF  LF iF 
VFp sgp  iF   VFn sgn  iF   x 2  R C v1  LC v1 .
(7)
At this point it is essential to highlight that parameters x, RF,
LF, VFp and VFn are considered to remain constant during the
analyzed period. In turn, this assumption is not exactly true in
the first period after the fault begins, when the buildup

Fig. 1. Single-phase two-terminal line model and a fault at point F. Fig. 2 High Impedance Fault model proposed in [2]

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

 
phenomenon occurs [20]. However, the shoulder phenomenon 1 (14)
θˆ  XT X XT y.
is characterized by temporary constant values of the HIF
parameters and, after some cycles the fault tends to stabilize. The solution of (14) is a simple task to achieve with modern
For this reason, the fault parameters are assumed constant in digital computation systems. The unique requirement is an
the formulation. adequate selection of samples.
Still, it is possible to rewrite (7) in matrix notation: All equations were developed for a single-phase system
v1   x RF LF VFp VFn x 2   consideration. However, their extension to a three-phase
system is simply done by the replacement of the parameters R,
sgp  iF  sgn  iF  vsh  ,
T
vs iF iF (8)
L, C, as well as x, RF, LF, VFp and VFn by 3x3 matrices and,
signals v1, i1 and iF by 3x1 signal vectors. Then, after some
where:
straightforward algebra, equation (11) can be extended for
vs   Ri1  Li1  ; (9)
three-phase lines as:
vsh    RCv1  LCv1 . (10)  x 
 v1,1
T  T
  v s ,1
T
iF,1 T
iF,1  
T
sgp iF,1  
T
sgn iF,1 T 
v sh ,1 
  RF 

Expression (8) is very elegant but shows two main difficulties  T   T
 v1,2   v s ,2
T
iF,2 T
iF,2 sgp  i T
sgn  i  T T  
v sh LF 
to be overcome. The first is the appearance of a quadratic term F,2 F,2

  (15)
,2
   VFp 
of the fault distance that leads to a nonlinear relation and, the     V 
second drawback is the remaining of the fault current iF and its v1,TN  v T
 s, N
iF,T N iF,TN  
sgp iF,T N  
sgn iF,T N T
v sh 
,N   2 
Fn

derivative. If the line capacitance is neglected, the signal vsh  x 

vanishes and the first difficulty is avoided. The disregard of Note that all variables in (15) are in bold indicating that they
the line capacitance is then considered as a first are matrices.
approximation, leaving to a linear relation in (8) and
v1  v1a v1c  ,
T
simplifying its solution. Regarding to the second drawback, v1b (16)

i1  i1a i1b i1c  ,

there is no simply way to get around this problem, meaning T
(17)
that the fault current must be estimated. In turn, different ways
iF  iFa iFb iFc  ,
T
can be conceived in order to estimate iF and the approach used (18)
in this work is explained in section III.
vs  vsa vsc  ,
T
In order to estimate the unknown parameters of (8), the line vsb (19)
voltages and currents can be uniformly sampled with a
vsh  vsha vshc  ,
T
preselected sampling frequency. Then, equation (8) can be vshb (20)
written for a set of N samples and constitute the following  RFa 0 0 
R F   0 0  ,
system of equations:
RFb (21)
 x 
 v1,1   vs ,1 iF,1 iF,1  sgp  iF,1  sgn  iF,1      0 0 RFc 
v  v   RF 
 1,2    s ,2 iF,2 iF,2 sgp  iF,2  sgn  iF,2     LF  .(11)
  LFa 0 0 
     
L F   0 LFb 0  ,
    VFp  (22)
v1, N  vs , N iF, N iF, N sgp  iF, N  sgn  iF, N    
VFn   0 0 LFc 
In order to obtain (11) the transpose of (8) was considered as VFpa 0 0 
well as the line capacitance was neglected. In equation (11) 
VFp   0 VFpb 0  , (23)
numbers after coma indicates the sample number. Still, this
 0 0 VFpc 
relation has a linear system of equations form:
y  Xθˆ  ξ, (12) VFna 0 0 
where y is the vector of sampled v1, X is the first matrix of 
VFn   0 VFnb 0  (24)
(11), called regressors matrix and θ̂ is the estimated vector of  0 0 VFnc 
parameters. The vector ξ has the same dimension of y and and
was added in order to represent the noise.
Clearly (12) has a structure that allows the application of  x 0 0
the classical LSE approach for parameter estimation, which x   0 x 0  . (25)
consists on the minimization of the quadratic norm of the  0 0 x 
vector ξ expressed as:
Signals (19) and (20) are calculated respectively with (9)
J θ  ξ T ξ. (13)
and (10), considering the line series resistance and inductance
The final solution of (13) is a simple algebraic relation given matrices and the shunt capacitance matrix.
by: The equations (11) or (15) can also be valid for a section
downstream node 2 of Fig. 1. Thus, if a section has a mixed

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

distribution line type, as a mixed overhead line and cable, the consisted on the polynomial approximation approach [21].
section should be divided into the necessary number of single This method models the signal through a polynomial
distribution line types. Downstream line sections are analyzed formulation. Estimation of polynomial coefficients are
if estimated fault location is higher than section length. In realized considering a small sample window and applying the
order to apply the formulation on downstream line sections, least square estimation method [21]. After, the analytical
initial line section node voltages and currents are necessary. derivative of the estimated function is used to approximate the
Current and voltage signals should be propagated to the next analyzed signal derivative.
node using equation (3). Thus, the digital signal is approximated using a short time
window by:
III. IMPLEMENTATION OF THE PROPOSED APPROACH y  n    0  1n  
 k 1nk 1   k nk  e  n  , ni  n  n f .
In the previous section the theoretical development of the
formulation was presented without considering computational (26)
implementation issues. Aspects related with fault current Where k is the polynomial degree, e(n) is the regression error,
estimation, numerical signal derivatives calculation and the and ni, nf the index corresponding to the beginning and the end
choice of samples to apply the LSE are described in this of the interval of samples where the approximation is being
section. calculated. Therefore, in this window the following matrix
relation can be written:
A. Fault current estimation
 0 
The primordial difficulty with one-terminal fault location
 y  ni   1 ni nik 1 nik   1 
formulations is the need to estimate the fault current. In order     
to provide a mathematical path for fault current estimation, the    e (27)
 y  n f   1 n2 k  
 
superposition theorem is considered. Consider the circuit
 n kf 1 n f   k 1 
represented in the Fig. 1. At the fault period, the current   k 
leaving node 1 is composed by the load current, line
capacitance current and the fault current. As a HIF produces a y  Xθˆ  e (28)
low current in comparison with the load current, one can
Therefore, parameters θ̂ of (28) can be estimated using
assume that the load current is not significantly modified by
(14), resulting in the polynomial’s coefficients that best
the fault occurrence. Hence, in this work the consideration that
approximates the signal y, in the window of selected samples.
the pre-fault current at node 1 is very similar to the fault
Finally, defining a sample of interest within the interval,
period load current is assumed. As HIF are mainly phase-to-
named n0, the estimation of the first and second derivatives are
ground faults, it makes sense to use the difference between the
obtained by:
fault and pre-fault residual currents as an estimation of the
fault current, similar to the proposed by [4]. However, as yˆ   n0   ˆ1    k  1  k 1n0k 2  k  k n0k 1 , (29)
considered signals in this work are in time domain, the yˆ   n0   ˆ 2  
subtraction must be made from the corresponding samples of
different cycles. We propose the implementation of such by a  k  2  k  1 k 1n0k 3   k  1 k n0k 2 .
(30)
cycle-by-cycle difference digital filter as illustrated in Fig. 3. The maximum order of the derivative that can be estimated
Samples subtraction using three cycles before the actual fault using this methodology is limited to the degree k of the
instant shown to provide an accurate estimation and was in estimated polynomial. In addition, the minimum number of
this work applied. samples that must be considered is k+1. In this work, the
B. Numerical derivatives signals were approximated with third-degree polynomials
using eleven samples per window [21]. The reference sample
The need of the first and second signal derivatives
is located at the middle of the estimation windows.
knowledge in the proposed formulation makes necessary the
selection of an estimation method. Thus, several numerical C. Implementation of the Least Square Estimation
derivative approaches were studied. The selected method Equation (11) shows that positive and negative parts of the
fault current must be used in order to apply the LSE approach.
If only a negative or positive semi-cycle of iF is considered,
the fourth or fifth column of the regressors matrix X will be
zero and then the matrix becomes singular. Therefore, in this
case one of these columns must be excluded from the equation
and, the same can said about equation (15). If the selected
window contains both positive and negative semi-cycles of the
fault current, equations (11) and (15) can be directly used.
In order to avoid an extensive study of best set of samples
to solve the LSE problem selection, equation (14) was
Fig. 3. Illustration of the fault current estimation.
implemented in a moving window of 1.5 cycle length.

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

Therefore, the LSE is performed for each window as a filter

and the results are automatically calculated when an HIF
occurs.
Simulation tests presented unsatisfactory results when
estimation of the fault current is near its zero crossing.
Therefore, a function was programed in order to select the
appropriate samples of voltages and currents to apply the LSE
algorithm. This function selects the samples corresponding
with an estimated fault current that exceeds a predefined Fig. 5. Diagram of the IEEE 13 nodes test feeder [23]. Principal section is
threshold. The procedure is initiated with the obtaining of the drawn in thick line and is composed by sections 1, 2 and 3.
maximum absolute value of the estimated fault current into a also be installed in different branches of the distribution
moving window of 1.5 cycle. This value is finally multiplied system, establishing a distributed measurement configuration
by an empirical factor of 0.15 in order to get the threshold. that can be used to improve the fault location. It should be
The overall process of the LSE implementation is illustrated clear however that the proposed methodology does not require
in the Fig. 4. J1 is the window where the LSE is applied and such remote measurements. Modifications made on the IEEE
J2 is the window where the threshold is obtained. J2 is shifted 13 nodes test feeder were: the substitution of the voltage
to the pair of J1 and is applied on the estimated fault current. regulator at node 650 by a transformer; replacement of the
D. Solution Considering Line Capacitance underground cables by overhead lines and all loads were
modeled as constant impedance. In order to get an idea of the
If the line capacitance is considered, the nonlinear relation
simulated system load, the RMS currents as phase a, b and c
shown in (15) reflects closer the behavior of the system
are 484 A, 563.6 A and 515.5 A, respectively.
illustrated in Fig. 1. The assumption of the line capacitance
Only faults on the main section 1 were simulated using the
implies that the proposed linear LSE is no longer the correct
model presented in Fig. 2. Table I presents the parameters
approach to estimate the HIF parameters and the fault
used to simulate the HIF. Each combination of parameters was
distance. In order to take into account the line capacitance, a
randomly chosen and identified with the number shown in the
suitable algorithm must be used to solve the nonlinear relation
first column of the Table I. In this table one can see that fault
presented in (15). The selected approach used in this work was
currents are very low with relation to the actual load current.
initially to use the solution of the LSE based method as a first
Therefore, a conventional overcurrent protection device would
estimation, and the final value of x is estimated using the
not operate. The line sections were dividend in ten points
nonlinear relation presented in (15). Still, solution of (15) is
where the considered HIF were simulated with three different
obtained with the residue function (13) and the steepest
inception instants. The time step used in the ATP simulation
descent Newton’s method [22]. This is a nonlinear LSE
was of 1 µs and all resulting signals were resampled at a
approach (NLLSE), applied with the samples between 0.5 and
sampling rate of 256 samples per cycle of 60 Hz.
2.5 cycles after fault inception. The threshold used is an
average of the linear LSE based method threshold. TABLE I
HIGH IMPEDANCE FAULT MODEL PARAMETERS
Id iF (%)* RF (Ω) XLF (Ω) VFp (V) VFn (V)
IV. PROPOSED VALIDATION TEST
1 40.9 10 0 0 0
In order to validate the proposed formulation, several HIF 2 39.1 10 1 80 100
were simulated on the modified IEEE 13 nodes test feeder 3 6.2 50 10 500 700
illustrated in the Fig. 5 [23]. This figure shows a Disturbance 4 2.4 100 23 1000 1200
Monitoring Equipment (DME) at the substation, indicating 5 1.6 150 30 1000 1200
that voltage and current waveforms are acquired at this 6 1 200 45 1200 1400
location. Dashed line arrows indicate that others DME could 7 1.5 160 26 1110 1100
8 2.4 115 15 800 900
9 2.4 134 21 360 670
10 4.1 80 10 410 600
* Percentage in relation of 563.6 A, the RMS load current at phase b.

In the subsequent analyses, the perceptual estimation errors

were calculated as:
 estimated fault distance    fault distance
error  100% (31)
 line length 
for the fault distance and:
 estimated parameter    parameter 
error  100% (32)
Fig. 4. Implementation of the linear least squares estimation approach for fault
 parameter 
location. for the others parameters. Equations (31) and (32) are not

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

presented as an absolute value error. Therefore a negative

error means parameter under-estimation and a positive error
means over-estimation.
A. HIF at section 1: a particular case of the linear LSE
The first analysis consisted on applying HIF, characteristics
presented in TABLE I, in the first section of the test system
represented in Fig. 5. One of the results of the application of
the proposed LSE using voltages and currents signals recorded
at node 650 are shown in Fig. 6. One can notice that all
estimated parameters pass through an abrupt change when the
HIF occurs. Before the inception of HIF, equation (14) has not
a defined solution because matrix X is almost always singular.
This characteristic shows that the proposed formulation can
also be used to detect the HIF occurrence. However, it should Fig. 7. Distance estimation perceptual error of the linear least squares
estimator approach. Mean value.
be clear that no further studies on the distinguishing between
such faults and other switching phenomena where made.
B. HIF at section 1: global behavior of the linear LSE
The proposed approach produces various HIF distances and
parameters estimations, as explained in subsection C of the
section III and exemplified in Fig. 6. In order to obtain unique
parameters estimation, an average calculation operation was
performed using an estimation set. As observed in Fig. 6, all
estimations tend to stabilize when the moving windows (see J1
and J2 in Fig. 4) are completely filled with samples of the fault
period. In view of this characteristic, the average was
calculated using results within one cycle after 1.5 cycles of
fault inception. This average is the final result considered in
the linear LSE proposed approach.
As previously mentioned the test set consisted of 10 HIF,
presented in Table I, applied in ten distances along the
Fig. 8. HIF parameters estimation perceptual error of the linear least squares
considered section I. Each fault occurrence was selected at estimator approach. Mean value.
three time instants corresponding with the maximum positive, relation to the fault inception angle. The same can be said
zero crossing and maximum negative of the faulted-phase regarding to the HIF parameters estimation, as presented in
voltage. These inception angles were respectively named as Fig. 8.
0°, 90° and 180°. Results of the distance estimations are The average of the distance estimation error presented in
presented in Fig. 7 and results of the HIF parameters Fig. 7 maintains a negative trend, showing that the proposed
estimation are presented in Fig. 8. These figures present the method tends to under-estimate the fault distance. The error
average value of the estimation errors. also tends to increase with the fault distance.
Fig. 7 presents an important characteristic, which is that the As shown in Fig. 8, the average estimations of RF, VFp and
proposed approach has not shown an appreciable effect in VFn are significantly better than the average estimation of
distance. In addition, Fig. 8 also shows that the average
estimation errors present a slight increase with the fault
distance. The average estimation error of LF is not as good as
the others HIF parameters and the increase of errors in
function of the fault distance are more significant.
One should note that the considered faults present a
nonlinear behavior. Some produce fault currents less than
10 % of the actual load current. A classical fault location
method that considers the fault as a pure resistance will not
produce results at all. Despite of the obtained errors on fault
distance estimation, the proposed approach generates most
useful estimates.

Fig. 6. Results of the proposed implementation of the LSE to estimate the HIF
C. HIF at section 1: consideration of line capacitance
parameters. The simulated HIF is the number 4 (Table I) at 360 m from the Results obtained with the linear LSE and presented in the
substation (node 650). True values are shown in bold horizontal lines.

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

previous subsection indicate that the system of equations the bus number 650 of Fig. 5. Test results of the fault distance
shown in (15) presents a poorly-conditioned behavior estimation are presented in Fig. 11. In this case, both methods
regarding to the estimation of x and LF. In the linear approach presented a reasonable estimation of the fault distance. As the
the line capacitance was neglected. Therefore, the proposal of lines of the test system are unbalanced, the method proposed
the present subsection is to evaluate if the consideration of the in [3] presents a slightly worse estimative. This is true because
line capacitance can produce a significant improvement of [3] is conceived using an symmetrical components approach,
results. This is done by applying the approach presented in the assuming a balanced system.
subsection D of the section III. In the second case, the fault was set with VFp = 700 V,
Results are presented in Fig. 9 and Fig. 10, in the same way VFn = 1000 V, LF = 0.0265 H and RF = 20 Ω. The comparative
as for the linear LSE. The behavior of the NLLSE was very test results of the fault distance estimation are presented in
similar to those obtained with the linear approach. This results Fig. 12. In this case, the method presented in [3] does not
shows that the consideration of the line capacitance does not present an accurate estimation of the fault distance. On the
affect significantly the results in the considered test system. other hand, the proposed approach presents a good estimation
in this case. Comparative tests considering fault parameters
estimation were not included since [3] does not provide such.

Fig. 9. Distance estimation perceptual error of the nonlinear least squares Fig. 11. Results of fault distance estimation using: (a) the proposed approach
estimator approach. Mean value. and; (b) approach presented in [3]. The fault is at 300 m from the substation
(node 650), value showed in bold horizontal lines. The HIF parameters are
RF = 20 Ω, LF = 0 H, VFp = VFn = 1000 V.
.

Fig. 12. Results of fault distance estimation using: (a) the proposed approach
and; (b) approach presented in [3]. The fault is at 300 m from the substation
(node 650), value showed in bold horizontal lines. The HIF parameters are
RF = 20 Ω, LF = 0.0265 H, VFp = 700 V, VFn = 1000 V.
Fig. 10. HIF parameters estimation perceptual error of the nonlinear least
squares estimator approach. Mean value.
V. CONCLUSIONS
D. Comparative analysis
In this paper, a HIF location analytical methodology is
This subsection presents a simple comparative analysis in proposed. The method uses a parameter estimation based
order to highlight the contribution that the present work makes approach for fault distance and parameters calculation. In the
to the state of the art. Hence, the proposed approach is proposed formulation, only one terminal voltage and current
compared with [3]. Two fault scenarios are analyzed. signals are used and the fault location and parameters can be
In the first case, a fault composed only by an arc and an arc estimated within the first cycles after the fault inception. The
resistance is simulated. This is equivalent to the fault model technique allows consideration or not of line capacitance. If
considered in [3] and means that VFp = VFn and LF = 0. The RF line capacitance is not considered, a linear set of equations is
parameter was set 20 Ω and the fault distance at 300 m from solved with a LSE based approach. If line capacitance is

U.S. Government work not protected by U.S. copyright.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2015.2507541, IEEE
Transactions on Power Delivery

considered, a nonlinear set of equation must be solved. For Proc. 37th Annu. North Am. Power Symp. 2005., pp. 219–224, 2005.
[9] L. Garcia-Santander, P. Bastard, M. Petit, I. Gal, E. Lopez, and H.
such, the proposed method is implemented in two steps. Opazo, “Down-conductor fault detection and location via a voltage
Firstly, the line capacitance is neglected and the LSE is based method for radial distribution networks,” IEE Proc. - Gener.
calculated with a moving window on the signals, resulting in Transm. Distrib., vol. 152, no. 2, pp. 180–184, 2005.
[10] N. I. Elkalashy, M. Lehtonen, H. a. Darwish, A. M. I. Taalab, and M. a.
parameter estimation for each sample. In turn, as estimated
Izzularab, “DWT-based detection and transient power direction-based
parameters pass through an abrupt change at the fault location of high-impedance faults due to leaning trees in unearthed MV
inception instant, the proposed approach can also be used to networks,” IEEE Trans. Power Deliv., vol. 23, no. 1, pp. 94–101, 2008.
detect the HIF occurrence. The second step was developed in [11] S. Gautam and Brahma, “Detection of High Impedance Fault in Power
Distribution Systems Using Mathematical Morphology,” IEEE Trans.
order to consider the line capacitance. It consists on an Power Syst., vol. 28, no. 2, pp. 1226–1234, May 2013.
iterative search technique to estimate the parameters that [12] A. S. Bretas, M. Moreto, R. H. Salim, and L. O. Pires, “A Novel High
minimize a residual function. This is carried out with the Impedance Fault Location for Distribution Systems Considering
Distributed Generation,” in Transmission Distribution Conference and
values estimated in the linear LSE as a starting point. Test Exposition: Latin America, 2006. TDC ’06. IEEE/PES, 2006, pp. 1–6.
results on the study system indicate that the consideration of [13] A. H. A. Bakar, M. S. Ali, C. Tan, H. Mokhlis, H. Arof, and H. A. Illias,
line capacitance resulted in similar accuracy. “High impedance fault location in 11kV underground distribution
systems using wavelet transforms,” Int. J. Electr. Power Energy Syst.,
A comparative analysis was made with a state-of-the-art vol. 55, pp. 723–730, Feb. 2014.
method. Results show that when a more complete HIF model [14] D. S. Gazzana, G. D. Ferreira, a. S. Bretas, a. L. Bettiol, a. Carniato, L.
is considered, the state-of-the-art method cannot give an F. N. Passos, a. H. Ferreira, and J. E. M. Silva, “An integrated technique
for fault location and section identification in distribution systems,”
accurate estimation of the fault distance. On the other hand, Electr. Power Syst. Res., vol. 115, pp. 65–73, Oct. 2014.
the proposed approach gives a more accurate estimation of the [15] N. Zamanan and J. Sykulski, “The evolution of high impedance fault
fault distance providing as well the fault parameters modeling,” in 2014 16th International Conference on Harmonics and
Quality of Power (ICHQP), 2014, pp. 77–81.
estimation.
[16] M. R. M. Castillo, J. B. a London, N. G. Bretas, S. Lefebvre, J. Prévost,
The analytical formulation presented in this work considers and B. Lambert, “Offline detection, identification, and correction of
a HIF model in equation derivation. As a result, this allows the branch parameter errors based on several measurement snapshots,” IEEE
location of the fault using one terminal signals, which clearly Trans. Power Syst., vol. 26, no. 2, pp. 870–877, 2011.
[17] M. Ghazizadeh-Ahsaee, “Accurate NHIF locator utilizing two-end
represents a contribution to the state of the art. In view of the unsynchronized measurements,” IEEE Trans. Power Deliv., vol. 28, no.
test results, the method is insensitive to the fault inception 1, pp. 419–426, 2013.
instant. [18] R. H. Salim, K. C. O. Salim, and A. S. Bretas, “Further improvements
on impedance-based fault location for power distribution systems,” IET
Results are encouraging and show potential for online Gener. Transm. Distrib., vol. 5, no. 4, pp. 467–478, 2011.
applications. Still, considering the rapid advances in remote [19] N. I. Elkalashy, M. Lehtonen, H. a. Darwish, M. a. Izzularab, and A. M.
monitoring technology, an interconnected network of I. Taalab, “Modeling and experimental verification of high impedance
arcing fault in medium voltage networks,” IEEE Trans. Dielectr. Electr.
monitoring devices can be conceived to monitor different Insul., vol. 14, no. 2, pp. 375–383, Apr. 2007.
branches of the distribution system. The combination of this [20] W. C. dos Santos, B. A. de Souza, N. S. D. Brito, F. B. Costa, and M. R.
scheme with the proposed HIF location approach would C. Paes, “High Impedance Faults: From Field Tests to Modeling,” J.
Control. Autom. Electr. Syst., vol. 24, no. 6, pp. 885–896, Sep. 2013.
considerably increase the reliability and accuracy of the fault [21] L. A. Aguirre, Introdução à Identificação de Sistemas: técnicas lineares
detection and location. e não-lineares aplicadas a sistemas reais, 3th. ed. Belo Horizonte:
Editora UFMG, 2007.
[22] D. Luenberger and Y. Ye, Linear and Nonlinear Programming, 3th ed.
REFERENCES
Boston, MA: Springer US, 2008.
[23] W. H. Kersting, “Radial distribution test feeders,” IEEE Trans. Power
[1] IEEE Power System Relaying Committee, “High Impedance Fault Syst., vol. 6, no.3, pp. 975-985, 1991.
Detection Technology,” 1996.
[2] A. E. Emanuel, D. Cyganski, J. A. Orr, S. Shiller, and E. M. Leonardo Ulises Iurinic (S´11) received the BSc degree in electromechanical
Gulachenski, “High impedance fault arcing on sandy soil in 15 kV engineering from the National University of Missiones (UNaM), Argentina in
distribution feeders: contributions to the evaluation of the low frequency 2009, and the MSc degree in electrical engineering from the Federal
spectrum,” IEEE Trans. Power Deliv., vol. 5, no. 2, pp. 676–686, Apr. University of Rio Grande do Sul (UFRGS), Brazil in 2012. He is currently
1990. pursuing the PhD degree in electrical engineering at UFRGS.
[3] M. B. Djurić, Z. M. Radojević, and V. V. Terzija, “Digital signal
processing algorithm for arcing faults detection and fault distance A. R. Herrera-Orozco received the BSc and MSc degree in Electrical
calculation on transmission lines,” Int. J. Electr. Power Energy Syst., Engineering from Technological University of Pereira (UTP), Colombia, in
vol. 19, no. 3, pp. 165–170, Mar. 1997. 2010 and 2013, respectively. He is currently pursuing the PhD degree in
[4] M. M. Alamuti, H. Nouri, R. M. Ciric, and V. Terzija, “Intermittent fault Electrical Engineering from the Federal University of Rio Grande do Sul.
location in distribution feeders,” IEEE Trans. Power Deliv., vol. 27, no.
1, pp. 96–103, 2012. Renato Gonçalves Ferraz (M’14) received the BSc, MSc, and PhD degrees in
[5] S. Kulkarni, S. Santoso, and T. a. Short, “Incipient fault location electrical engineering from the Federal University of Rio Grande do Sul,
algorithm for underground cables,” IEEE Trans. Smart Grid, vol. 5, no. Brazil, in 2005, 2010, and 2014, respectively. He is currently an Assistant
3, pp. 1165–1174, 2014. Professor at the University of Caxias do Sul, Brazil.
[6] A. F. Sultan, D. J. Fedirchuk, and G. W. Swift, “Detecting Arcing
downed-wires using fault current flicker and half-cycle asymmetry,” Arturo Suman Bretas (M´99) (SM´12) received the BSc and MSc degrees in
IEEE Trans. Power Deliv., vol. 9, no. 1, pp. 461–470, 1994. Electrical Engineering from the University of São Paulo, Brazil in 1995 and
[7] Z. M. Radojević, V. V. Terzija, and M. B. Djurić, “Numerical algorithm 1998 respectively. At Virginia Tech, in 2001, he received his PhD degree. He
for overhead lines arcing faults detection and distance and directional is currently a Professor of the Electrical and Computer Engineering
protection,” IEEE Trans. Power Deliv., vol. 15, no. 1, pp. 31–37, 2000. Department at the University of Florida.
[8] F. M. Uriarte and V. Centeno, “High-impedance fault detection and
localization in distribution feeders with microprocessor based devices,”

U.S. Government work not protected by U.S. copyright.