IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO.

4, AUGUST 2015 1843

A Novel Method for Noninvasive Estimation of

Utility Harmonic Impedance Based on Complex
Independent Component Analysis
Farzad Karimzadeh, Saeid Esmaeili, and Seyed Hossein Hosseinian

Abstract—This paper presents a new noninvasive method for A wide variety of methods has been developed to measure
calculating utility harmonic impedance at the point of common the utility harmonic impedance at the PCC. These methods can
coupling (PCC). The proposed method is based on a statistical be categorized into two types: 1) the invasive methods and 2)
signal-processing technique, known as independent component
analysis (ICA). The complex ICA technique is applied to the equa- noninvasive methods. In invasive methods, the utility harmonic
tions derived from Northon equivalent circuit model at the PCC impedance is extracted from the variations of voltage and cur-
in order to estimate the utility harmonic current values. Then, the rent which are caused by the injected disturbances or (inter)har-
estimated values of the utility harmonic current are used in an monic currents to the connected network [1]–[4]. Due to the sig-
optimization problem to calculate the utility harmonic impedance. nificant level of excitation, a wideband frequency result is ob-
Due to considering the utility harmonic current variations in
utility harmonic impedance calculation, the proposed method is tained from invasive methods. Despite sufficient accuracy, the
relatively robust against the background harmonic fluctuations. need for an expensive high-speed data-acquisition system and
The results obtained from computer simulation and a real case disturbance injection devices has restricted the application of
study verify the effectiveness of the proposed method. invasive methods. Moreover, the invasive methods cannot be
Index Terms—Background harmonic, harmonic impedance, performed repeatedly, because the injected disturbances may be
noninvasive method, power quality (PQ). harmful to the network especially in medium-voltage (MV) and
high-voltage (HV) systems.
On the other hand, in noninvasive methods, the natural varia-
I. INTRODUCTION
tions of voltage and current caused by harmonic load variability
are used to estimate the utility harmonic impedance at the PCC.

U TILITY harmonic impedance is an important parameter

in the harmonic study of electrical networks. It is gener-
ally used at the point of common coupling (PCC) for designing
In contrast to invasive methods, the noninvasive methods can
be applied continuously without imposing disturbances into the
network. Moreover, the required data for noninvasive methods
filters, verification of harmonic limit compliance, predication of can be simply achieved from a low-cost power-quality (PQ)
system resonance, and evaluating the harmonic voltage emis- analyzer.
sion level. Determining utility harmonic impedance is compar- Up until now, different noninvasive methods have been
atively a challenging task. Besides computer simulation tech- developed to estimate the utility harmonic impedance at the
niques, which need detailed data of system components, direct PCC [5]–[7]. The main problem associated with these methods
measuring of utility harmonic impedance is also favorable in is their susceptibility to the background harmonic fluctuations
many applications. The major advantage of the direct measuring which influences the precision and reliability of results. In
of the utility harmonic impedance is its independence from de- [5], a method based on the covariance characteristic of the
tailed system data which are often hard to access. Furthermore, random vector is proposed to eliminate the negative effect of
the measurement can be carried out from time to time to con- the background harmonic. Nevertheless, the proposed method
sider the effect of system changes on the value of the utility gives poor results where background harmonic fluctuations
harmonic impedance. are dominant at the PCC. Another method based on data
selection is proposed in [6]. In this method, first, the com-
Manuscript received June 21, 2014; revised September 30, 2014 and De- plex least-square regression is utilized to estimate the utility
cember 15, 2014; accepted January 18, 2015. Date of publication February 03, harmonic impedance during specific time periods. Then, two
2015; date of current version July 21, 2015. Paper no. TPWRD-00736-2014.
independent data-selection techniques are employed to select
F. Karimzadeh is with the Energy Department, Graduate University of Ad-
vance Technology, Kerman 7631133131, Iran (e-mail: [email protected]). the reliable utility harmonic impedance samples. Although
S. Esmaeili is with the Electrical Engineering Department, Shahid Bahonar the practical aspects of the aforementioned method are shown
University of Kerman, Kerman 7616914111, Iran (e-mail: [email protected].
distinctly in this paper, the background harmonic still has
ir).
S. H. Hosseinian is with the Electrical Engineering Department, Amirkabir its own negative effect on results especially in cases where
University of Technology, Tehran 1591634311, Iran (e-mail: hosseinian@aut. the variations of the background harmonic are high. In [7], a
ac.ir).
power spectral density of current and a cross power spectral
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. density of current and voltage are used to estimate the utility
Digital Object Identifier 10.1109/TPWRD.2015.2398820 harmonic impedance. In addition, a series of conditions is

0885-8977 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 1844 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 4, AUGUST 2015

defined to detect the utility harmonic impedance samples which

are caused more by consumer-side variations. Similar to the
previous methods, this method is vulnerable to the background
harmonic fluctuations and is inefficient in cases where voltage
variations are mostly caused by the utility side.
In this paper, a novel noninvasive method is proposed to es-
timate the utility harmonic impedance at the PCC. The pro-
posed method is based on a well-known statistical technique,
the so-called complex independent component analysis (ICA).
The ICA technique is able to estimate the unobserved signals
from the observed mixtures without knowing the way the un-
observed signals are mixed [8]. This technique was previously
used in power systems for load forecasting and harmonic load
identification [9]–[11]. In this paper, the ICA technique is ap-
plied to the equations derived from Northon equivalent circuits
of consumer and utility at the PCC in order to estimate the utility
harmonic current values. Then, the normalized values of the
estimated utility harmonic current are used in an optimization
problem to calculate the utility harmonic impedance. According
to consideration of the utility harmonic current variations in the
utility harmonic impedance calculation, the proposed method
is comparatively robust against background harmonic fluctu-
ations. The performance of the proposed method is evaluated
through utilizing computer simulation and field tests.
This paper is organized as follows: in Section II, the ICA
technique is concisely introduced. The proposed method is de-
scribed in Section III. The computer simulation and field test
are employed to demonstrate the effectiveness of the proposed
Fig. 1. Complex FastICA algorithm.
method in Section IV. Some considerations about the proposed
method are discussed in Section V. Finally, conclusions are
drawn in Section VI.
are unknown. The goal of the ICA method is to estimate the
separating matrix , and then obtain the source signals by
II. INDEPENDENT COMPONENT ANALYSIS
ICA is the statistical signal-processing technique to recover (3)
the unobserved signals or “source signals” from observed mix-
tures without knowing the way the source signals are mixed. In where the matrix is the estimation of source signals and
an ICA model, the observed signals are obtained from the linear stands for Hermitian transpose. Assumptions must be satisfied
combination of the source signals. This linear mixing model can in order to estimate the matrix by the ICA technique [10]
be expressed as follows [8]: as follows.
1) The source signals are statistically independent.
(1) 2) At most, one of the source signals is Gaussian distributed.
3) The number of observations is greater than or equal to
where is the -dimensional the number of sources .
vector of observed values, is There is a computationally efficient and robust approach to
the -dimensional vector of source values, is the estimate the ICA model called the fast-fixed point algorithm
matrix of mixing coefficients called the mixing matrix, and (FastICA) [8]. Here, the extended type of FastICA that is suit-
is the time index for . In (1), the source signals able for the complex ICA model is employed to estimate .
vector and mixing matrix are unknown variables. In The FastICA algorithm for the complex ICA model is repre-
the complex model of ICA, the vectors of , , and the sented in Fig. 1. In this algorithm, is a smooth even func-
matrix are complex values. Equation (1) can be represented tion, denotes the statistical expectation, for
in matrix form as follows: are the columns of matrix , and is the Hermitian trans-
pose of vector . There are different choices for the function
(2) [12]; however, here preference is given to ,
which is robust against outliers and suitable for different density
where is the observed signals matrix with dimensions, functions. In the aforementioned algorithm, ,
and is the source signals matrix with dimensions. and . The complete description of the com-
In (2), the source signals matrix and the mixing matrix plex FastICA algorithm is available in [12].

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 KARIMZADEH et al.: NOVEL METHOD FOR NONINVASIVE ESTIMATION OF UTILITY HARMONIC IMPEDANCE 1845

In (8), and are directly measured at the PCC. If is

determined in some way, the parameter can be easily calcu-
lated. In this paper, the ICA method is utilized to estimate .
However, it needs some considerations which are discussed in
the following text.
In practice, numerous samples of voltage and current are
recorded by the power-quality (PQ) measurement instrument
at the PCC. By assuming that the utility harmonic impedance
remains constant during the measurement interval, (6) and (7)
Fig. 2. Norton equivalent circuit model at the PCC.
can be represented in the matrix form as follows:

(9)
Before applying the FastICA algorithm to the data, the
observed signals are preprocessed by centering and whitening where
in order to simplify the ICA algorithm. Centering transforms
the observed signals to zero-mean variables and whitening (10)
linearly transforms the observed and centered signals, such that
the transformed signals are uncorrelated, have zero mean, and (11)
their variances equal unity [10].
(12)
III. PROPOSED METHOD

A. Problem Formulation where is the number of samples measured during

the time interval, ,
The Norton equivalent circuit model is prevalently used to , ,
evaluate the harmonic voltage emission at the PCC. In this and are the -dimensional vectors.
model, the utility and consumer are represented with their Expression (9) is quite similar to the ICA model, where the
Norton equivalent circuits (Fig. 2). This model can be applied observed signals are the measured voltage and current at the
for each individual frequency. According to Fig. 2 and by using PCC and , the source signals are the equivalent current
the superposition law, the voltage and current at the PCC can sources of the utility and consumer sides and , and the
be defined as follows: mixing matrix is the impedance matrix . In order to estimate
the separation matrix by using ICA, the triple conditions
(4) of the ICA technique must be satisfied. The third condition
is obviously satisfied since the number of observed signals
(5) is equal to the number of source signals. But the first and
second conditions require more considerations. According to
where and are the equivalent harmonic impedances of the first and second conditions, the source signals, that is, the
the utility side and consumer side, respectively, and and and vectors must be statistically independent and have
are the equivalent harmonic current sources of the utility side non-Gaussian distribution. It is shown in [9] that the voltage
and consumer side, respectively. and current measured in power networks can be decomposed
In practice, the equivalent impedance of the utility is de- into the fast- and slow-varying components by utilizing the
termined by short-circuit capacity of the network at the PCC. linear filter. Based on this principle, the and matrices can
Hence, the equivalent impedance of utility is relatively smaller be separated into the fast- and slow-varying components as
than the equivalent impedance of consumer at the fundamental follows:
frequency. Furthermore, since the equivalent impedance of
utility and consumer are inductive in most cases, it means (13)
that even at a higher harmonic order, the equivalent harmonic (14)
impedance of the utility would still be much smaller than that of
the customer at nonresonance frequencies [5]. So by assuming where and are the fast and slow-varying components of
, (4) and (5) are changed as follows: matrix , and and are the fast- and slow-varying compo-
nents of matrix . It can be simply deduced that the fast-varying
(6) component of the matrix is produced from the fast-varying
(7) component of the matrix, and the slow-varying component of
the matrix is produced from the slow-varying component of
Combining (6) and (7), the utility harmonic impedance can be the matrix. Hence, we have
obtained as follows:
(15)
(8)
(16)

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 1846 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 4, AUGUST 2015

It is shown in [11] that the fast-varying components of har- where and represent the real and imaginary parts of
monic loads currents are statistically independent and have non- vector , respectively; gives the mean value of vector
Gaussian distribution. Since the variations are related to the ; and scales the range of vector values in [0,1]. The
distributed harmonic loads in the utility side, and variations relationship of can be defined as follows:
are related to the harmonic loads in the consumer side, one can
deduce that the fast-varying components of and have the
same properties as harmonic loads currents, that is, they are (22)
statistically independent and have non-Gaussian distribution.
Therefore, the ICA technique can be applied to the fast-varying
components of the measured voltage and current at the PCC and, where and are the minimum and maximum
consequently, the separating matrix can be estimated. The values of vector , respectively.
separating matrix is then used to recover the slow-varying In this paper, the exhaustive search method is utilized to solve
components of the matrix as follows: the optimization problem. According to this method, first, a list
of all potential solutions to the problem is generated. Then,
(17) the potential solutions are evaluated one by one and the one
that gives the minimum is chosen. The upper limit, lower
where the matrix consists of the estimated slow-varying com- limit, and the resolution of real and imaginary parts of are
ponents of source signals and . There are two indetermi- critical parameters in solving the aforementioned optimization
nacies in the estimation of the ICA model which exist due to problem with an exhaustive search method. These parameters
the fact that the source signals and the way they are mixed are can be determined as a function of the fundamental short-circuit
unknown. The first indeterminacy is related to the ordering of impedance of the utility at the PCC. Empirically, it is recom-
source signals, that is, the vector could be the first or the mended that the limitations of the real and imaginary parts of
second row of , and the second indeterminacy is related to the are set to , and , respec-
scaling of source signals. The prior information about the source tively, and the resolution of changes is also set to .
signals is needed to eliminate two aforementioned indetermina- It should be noted that the aforementioned limitations can be
cies. Here, we have no prior information about the vector. sufficient for calculating utility harmonic impedance up to the
Hence, it is necessary to find a way to clear these two indeter- 25th harmonic even if there is a resonance condition in the fre-
minacies. In order to distinguish the order of the vector in the quency response of utility impedance.
matrix, the assumption of is utilized. According The aforementioned optimization problem can also be solved
to this assumption, the current at the PCC is approximately by other optimization methods.
equal to . Therefore, there should be a linear correlation be-
tween the and vector. This point is the key idea to find
B. Method Guideline
the order in the matrix. It means that the row of , which
has less correlation with , is , and the row of , which The procedures of the proposed method are summarized as
has a greater correlation with is . follows.
Although the ordering indeterminacy is solved by using the 1) Measure the and samples during the appropriate
correlation concept, the scaling indeterminacy still exists. In the time interval (for example, 1-h time interval with a data
following text, we represent that it is not necessary to eliminate resolution of 1 sample/s).
the scaling indeterminacy of . Instead of that, the normalized 2) Decompose and into the fast- and slow-varying
values of suffice to estimate the utility harmonic impedance. components by utilizing the linear filter.
The utility harmonic impedance estimation can be simply con- 3) Apply the ICA technique to the fast-varying components
verted to an optimization problem as follows. According to (8), of and and estimate the separating matrix .
the slow-varying component of can be defined as a function 4) Employ separating matrix to estimate the
of slow-varying components of and .
(18) 5) Use the correlation concept to find the order of in the
matrix.
The value of that minimizes the following objective function 6) Find by solving the optimization problem.
is the answer of the optimization problem:

minimize (19) C. Three-Phase Measurement

where In the three-phase system, voltages and currents may have

considerable imbalances. The single-phase measurement for
calculating utility harmonic impedance may lead to inaccurate
(20) results in this situation. In order to avoid the erroneous calcu-
lation, the Fortescue inverse transformation is used to convert
the voltages and currents of three phases into the positive-,
(21) negative-, and zero-sequence components. Then, the utility

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 KARIMZADEH et al.: NOVEL METHOD FOR NONINVASIVE ESTIMATION OF UTILITY HARMONIC IMPEDANCE 1847

harmonic impedance can be calculated for each sequence. The

Fortescue inverse transformation is defined as follows:

(23)

where and can be either a voltage or current

vector.
In noninvasive methods, the harmonic current source of the
consumer side is considered as the source of excitation for cal-
culating utility harmonic impedance. For certain harmonics and
sequences, the sufficient harmonic excitations are commonly
not available to calculate utility harmonic impedance. Here, the
following criterion is empirically defined to filter the harmonic
sequences which have insufficient harmonic excitation:

(24)

where is the mean value of the -sequence harmonic cur- Fig. 3. Profiles of simulated harmonic current sources for case 2. (a) Real part
rent samples and is the mean value of the fundamental of the utility harmonic current source. (b) Imaginary part of the utility harmonic
component of positive-sequence current samples at the PCC. current source. (c) Real part of the consumer harmonic current source. (d) Imag-
inary part of the consumer harmonic current source.
In (24), the subscript represents the positive-, negative-, or
zero-sequence component.
TABLE I
IV. SIMULATION AND FIELD TEST VERIFICATION CHARACTERISTICS OF THE SIMULATED MODEL

A. Computer Simulation
In this section, the model represented in Fig. 2 is simulated
to examine the performance of the proposed method. Here,
the utility harmonic impedance is assumed to be constant and
has a smaller value in comparison with consumer harmonic
impedance. The consumer harmonic impedance is defined by
normal distribution with the expectation of and PCC are caused more by the utility side. The characteristics of
variance of . As mentioned before, the harmonic current the simulated network for two cases are represented in Table I.
sources of the utility and consumer are composed of slow- The symbol stands for the Laplace distribution
and fast-varying components. In [11], a typical harmonic load with the expectation of and variance of .
profile is used to simulate the slow-varying component of the The proposed method is applied to the voltage and current of
harmonic current source. Moreover, the Laplace distributed PCC, and the utility harmonic impedance is calculated. The fast-
random variables are added to the slow-varying component and slow-varying components of and are separated by
to simulate the fast-varying component. It should be noted an 8-point moving average filter. Fig. 4 demonstrates the fast-
that the Laplace distributed random variables are statistically and slow-varying components of the real and imaginary parts of
independent; therefore, they are suitable to be considered as and related to case 2. The correlation analysis for de-
a fast-varying component. Similar to [11], here, two typical tecting the order of the vector in the matrix is represented
harmonic load profiles in accordance with the actual measure- in Fig. 5 for cases 1 and 2. In this figure, and represent
ments are allotted to and to simulate the slow-varying the samples of the first and second rows of the matrix, respec-
components of harmonic current sources. In addition, zero tively. As can be seen, in both cases, the samples magnitudes of
mean Laplace distributed random variables with specified the first row of the matrix have greater correlation with the
variances values are added to the normalized harmonic load magnitudes of the samples. Therefore, it can be simply in-
profiles to consider the fast-varying components. Fig. 3 illus- ferred that the samples of the second row of the matrix belong
trates the real and imaginary parts of and generated from to the utility-side harmonic source.
aforementioned instruction. There are 1440 samples for the The simulation results obtained for two cases are listed in Ta-
entire day, that is, one sample per minute. Two cases are defined bles II and III, respectively. The results obtained from the pro-
to evaluate the effectiveness of the proposed method. In the first posed method are compared with the results obtained from dif-
case, the fluctuations are larger than the ones, implying ferent well-known noninvasive methods. In Tables II and III,
that the variations of voltage at the PCC are caused more by the methods 1, 2, and 3 are the methods proposed in [5]–[7], respec-
consumer side. In the second case, the fluctuations are larger tively. Two data-selection approaches are proposed in [6] to esti-
than the ones, implying that the variations of voltage at the mate the utility harmonic impedance. In this paper, the threshold

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 1848 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 4, AUGUST 2015

TABLE II
SIMULATION RESULTS FOR CASE 1

TABLE III
SIMULATION RESULTS FOR CASE 2

The error between the estimated value of the utility harmonic

Fig. 4. Slow- and fast-varying components of the real and imaginary parts of impedance and the actual value is calculated through
harmonic voltage and current at the PCC (Case 2). (a) Real part of the harmonic the following equation:
voltage. (b) Imaginary part of the harmonic voltage. (c) Real part of the har-
monic current. (d) Imaginary part of the harmonic current.

Error (25)

The error between the real part of the estimated utility harmonic
impedance and the actual value , and the error between
the imaginary part of the estimated utility harmonic impedance
and the actual value can be calculated through (25).
According to the simulation results for case 1, all of the
methods have acceptable accuracy in the estimation of the
utility harmonic impedance. These results are reasonable be-
cause the influence of the utility harmonic current fluctuations
(background harmonic) on the harmonic voltage of PCC is
trifle. In case 1, the VR is 60% (which is more than 20%),
and NS is 28. It means that in method 2, the utility har-
monic impedance is estimated from the second data-selection
approach.
In contrast to case 1, the effect of the utility harmonic current
fluctuations is dramatically increased in case 2. The simulation
results indicate that the proposed method gives the best result for
case 2. This fact proves the robustness of the proposed method
against the background harmonic fluctuations. In case 2, the VR
is 11.8% (which is less than 20%), and NS is zero. It means that
in method 2, the utility harmonic impedance is estimated from
Fig. 5. Correlation analysis for detecting the vector in the matrix. Plots
of (a) samples versus samples for case 1. (b) samples versus
the first data-selection approach.
samples for case 1. (c) samples versus samples for case 2.
(d) samples versus samples for case 2. B. Field Test Verification
In this section, the practical feasibility of the proposed
method is evaluated by utilizing actual data. The measurement
of variance ratio (VR) is set to 20% for the first data-selection has been carried out at the secondary of a 1-MVA 20/0.4-kV
approach, and the threshold of is set to 0.9 for the second distribution transformer. This transformer is one of the quadru-
data-selection approach. The number of samples selected by the plet transformers of an ore-processing factory. This factory
second data selection is represented by NS. The best answer be- is a part of the Golgohar iron ore mine, which is located in
tween two data-selection approaches is reported as the result of Kerman. The one-line diagram of the electrical network of
method 2 in Tables II and III. For methods 1 and 2, the length this factory is demonstrated in Fig. 6. According to Fig. 6,
of each subinterval is set to 10 samples. the TR2 transformer supplies the combination of linear and

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 KARIMZADEH et al.: NOVEL METHOD FOR NONINVASIVE ESTIMATION OF UTILITY HARMONIC IMPEDANCE 1849

Fig. 6. One-line diagram of the ore-processing factory electrical network.

nonlinear loads. The linear loads are the induction motors, and
the nonlinear loads are the adjustable speed drives (ASDs).
The aggregation of the upstream network harmonic current and
the injected harmonic currents of TR3 and TR4 transformers
nonlinear loads make the background harmonic at the PCC.
The voltage and current at the PCC are measured by using a
MAVOWATT70 PQ analyzer. Since the phase data of voltage
and current are required by the proposed method, the proposed
method is sensitive to the synchronization of sampling. Here,
the voltages and currents are sampled synchronously by the
measurement instrument. The discrete Fourier transform (DFT)
tool is used to obtain harmonic voltage and current parame-
ters. According to the IEC 61000-4-7 standard [13], the DFT
window width is 10 cycles (for a 50-Hz system) with rectangular
weighting. The calculated harmonic parameters are averaged in
one second. All of the harmonic phases refer to the fundamental Fig. 7. Harmonic spectra of the (a) measured voltage and (b) measured current
at the PCC.
voltage in order to eliminate the phase errors caused by the fun-
damental frequency variation. The total measurement duration
is one hour and the data resolution is 1 sample/s. This way, 3600
samples are gathered for calculating utility impedance in dif-
ferent harmonics. Fig. 7 illustrates the typical spectra of voltage
and current which are calculated by performing DFT on 10-cy-
cles voltage and current snapshots. The THD of voltage for three
phases is about 1.75% and the THD of current is about 25% for
phases A and C, and about 20% for phase B. It can be seen from
Fig. 7 that significant imbalances exist in some harmonics of
voltage and current. In order to evaluate the background har-
monic level, a measurement has been performed at the PCC
when the TR2 transformer loads are out of service. Fig. 8 shows
the measured voltage spectrum in this situation. The voltage
spectrum obviously represents the considerable role of back- Fig. 8. Harmonic spectrum of the measured voltage at the PCC before con-
ground harmonic voltage in the harmonic voltage level at the necting the TR2 transformer loads.
PCC. This fact implies that this case study is well chosen to ex-
amine the robustness of the proposed method against negative
effects of the background harmonic. and the negative sequence of the utility harmonic impedance can
The three-phase measured voltages and currents of the 3rd, be calculated for 3rd, 5th, and 11th orders.
5th, 7th, 11th, and 13th harmonic orders are converted to the In order to estimate the utility harmonic impedance, the pro-
positive, negative, and zero sequences by employing (23). As posed algorithm is implemented in a PC using Intel Core i5
an example, Fig. 9 shows the magnitudes of positive, negative, CPU (3.30 GHz) and 4 GB of RAM by MATLAB software.
and zero sequences of the third harmonic voltage and current. The computation time required to estimate each of the harmonic
Applying the criterion of (25) for different sequences of har- impedances is almost 11213 s. Most of the computation time
monics indicates that the positive sequence of the utility har- is allocated to the exhaustive search method. Since the funda-
monic impedance can be calculated for 3rd, 7th, and 13th orders, mental short-circuit impedance of the utility at the PCC is ap-

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 1850 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 4, AUGUST 2015

Fig. 12. Correlation analysis for detecting the vector in the matrix in
Fig. 9. Magnitudes of positive, negative, and zero sequences of the third har- the third harmonic positive-sequence impedance calculation. (a) Magnitude of
monic (a) voltage and (b) current at the PCC. the first row samples of the matrix versus and (b) magnitude of the
second row samples of the matrix versus .

method 2. In method 2, the threshold of the VR is set to 20%

for the first data-selection approach, and the threshold of is
set to 0.9 for the second data-selection approach. The number
of samples in each subinterval is chosen as 60, so the length of
each subinterval is 60 s. The values of VR and NS parameters
are shown in Table IV. Similar to the simulation section, the
best answer between two data-selection approaches is reported
as the result of method 2. Because of the enormous errors,
Fig. 10. Magnitude of the positive sequence of the 3rd harmonic voltage at the
the results related to methods 1 and 3 are not represented in
PCC. (a) Slow-varying component. (b) Fast-varying component. Table IV. The harmonic impedances of the transformer that are
provided by the manufacturer are considered as the reference
value of the utility harmonic impedance. This assumption is
reasonable to some extent since the utility impedance in dis-
tribution networks is commonly dominated by the connection
transformer [14].
It can be seen from the results that in most cases, the errors as-
sociated with the proposed method are significantly smaller than
the ones associated with method 2. The exception is in the nega-
tive-sequence impedance of the 3rd harmonic where the results
obtained from method 2 are more accurate. The evident weak-
ness of method 2 is in the estimation of the resistive part of the
Fig. 11. Magnitude of the positive sequence of the third harmonic current at harmonic impedance. In contrast, the resistance values are rel-
the PCC. (a) Slow-varying component. (b) Fast-varying component. atively well estimated by the proposed method. The exceptions
are the 7th and 11th harmonics where the estimated resistance
values have slight differences with the reference values. In con-
proximately equal to 0.01 , the limitations of real and imag- trast to the resistive part, method 2 has better performance in the
inary parts of in the exhaustive search method are consid- estimation of reactance. The relative errors of positive and neg-
ered as [0,1] and , respectively, and the resolution of ative sequences of the third harmonic reactance obtained from
changes is set to 0.0001 . The fast- and slow-varying compo- method 2 are smaller than the ones obtained from the proposed
nents of measured harmonic voltages and currents are separated method. For other harmonics, the proposed method gives more
by an 8-point moving average filter. As an example, the slow- accurate results in comparison with method 2.
and fast-varying components of the third harmonic positive se-
quence of voltage and current are represented in Figs. 10 and V. DISCUSSIONS
11, respectively. The correlation analysis is utilized to detect the As mentioned before, the proposed method is established
order of the vector in the matrix. Fig. 12 shows the corre- upon the assumption of . This assumption is not
lation analysis for the calculation of the positive sequence of the satisfied in some cases. In the following text, some experi-
3rd harmonic utility impedance. As expected, there is a remark- mental points about this problem are discussed:
able linear correlation between and one of the rows of
the matrix. Applying the correlation analysis for other har- A. Harmonic Filter or Power Factor Correction Capacitor
monics demonstrates that in the 7th, 11th, and 13th harmonic Installed on the Consumer Side
orders, the second row of their and in 5th harmonic order the In the case that the harmonic filter or the power factor correc-
first row of its represents the vector. tion capacitor is installed on the consumer side, the magnitude
In Table IV, the impedance results obtained from the pro- of may be smaller than or equal to the magnitude of . Two
posed method are compared with the results obtained from instructions are recommended for utilizing the proposed method

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 KARIMZADEH et al.: NOVEL METHOD FOR NONINVASIVE ESTIMATION OF UTILITY HARMONIC IMPEDANCE 1851

TABLE IV in the new metering point. Since the characteristic of cable is

CALCULATED UTILITY HARMONIC IMPEDANCE IN THE REAL CASE STUDY usually available, by using a suitable model for cable (for ex-
ample, the model), the utility impedance can be estimated.

C. Existence of Resonance in the Utility Impedance Spectrum

The existence of resonance in the utility impedance spectrum
may lead to an increase of the magnitude in a specific fre-
quency. In this case, the magnitude of may be greater than or
equal to the in that frequency and, consequently, the utility
and consumer side affect the harmonic current at the PCC. It is
clear that the proposed method has no efficient performance in
this situation since the and cannot be distinguished in
the matrix by using linear correlation analysis. Although fur-
ther study is needed to solve this problem, the experiences of
the authors obtained from comprehensive field tests show that
even if there is a resonance frequency in the utility impedance
spectrum, the magnitude of still remains smaller than the
magnitude of in that frequency and, consequently, the pro-
posed method can still work effectively.

VI. CONCLUSION
In this paper, a new noninvasive method is proposed to es-
timate the utility harmonic impedance at the PCC. The pro-
posed method is developed based on the Norton equivalent cir-
cuit model at the PCC and ICA technique. The utility back-
ground harmonic current is estimated by the complex FastICA
approach. The estimated utility harmonic current is then used
in an optimization problem to calculate the utility harmonic
impedance. The results obtained from computer simulation and
field test prove the capability of the proposed method in the es-
timation of the utility harmonic impedance especially in cases
in this situation. First, since the measurement duration for cal- that the background harmonic fluctuations are dominated at the
culating utility harmonic impedance using the proposed method PCC. The proposed method does not require an expensive data
is relatively short (for example, one hour), the filter or capac- gathering instrument or other extra elements. It only needs the
itor can be disconnected temporarily in order to eliminate the voltage and current samples which can be simply achieved by a
negative effects of them on utility harmonic impedance calcu- low-cost PQ analyzer at the PCC.
lation. If disconnection of the filter or capacitor is not possible,
the integration approach, which is discussed in [5], is recom- REFERENCES
mended. In this approach, the metering point is replaced to the [1] A. Robert and T. Deflandre, “Guide for assessing the network harmonic
impedance,” presented at the Int. Conf. Elect. Distrib., Birmingham,
point between the filter (or capacitor) and consumer. So the cal- U.K., Jun. 1997, Joint CIGRE/CIGRE Working Group CC02.
culated impedance at the new metering point is or [2] M. Nagpal, W. Xu, and J. Sawada, “Harmonic impedance measurement
. Since the parameters of the harmonic filter and using three-phase transients,” IEEE Trans. Power Del., vol. 13, no. 1,
pp. 272–277, Jan. 1998.
power factor correction capacitor are commonly available; the [3] P. Bettini and A. D. Lorenzi, “400-kV-network harmonic impedance
utility harmonic impedance can be easily calculated. evaluation from harmonic distortion measurements at RFX site,” Eur.
Trans. Elect. Power, vol. 8, no. 1, pp. 31–38, Feb. 1998.
B. Long Underground Cable Supplying the Load [4] W. Xu, E. E. Ahmed, X. Zhang, and X. Liu, “Measurement of network
harmonic impedances: Practical implementation issues and their solu-
There are not very uncommon cases when the long under- tions,” IEEE Trans. Power Del., vol. 17, no. 1, pp. 210–216, Jan. 2002.
[5] J. Hui, H. Yang, S. Lin, and M. Ye, “Assessing utility harmonic
ground cable supplying the load is overcompensated due to ad- impedance based on the covariance characteristic of random vectors,”
ditional reactive power compensation. Therefore, the equivalent IEEE Trans. Power Del., vol. 25, no. 3, pp. 1778–1786, Jul. 2010.
characteristic of load is capacitive so that the magnitude of the [6] J. Hui, W. Freitas, J. C. M. Vieira, H. Yang, and Y. Liu, “Utility har-
monic impedance measurement based on data selection,” IEEE Trans.
consumer impedance decreases with frequency up to some Power Del., vol. 27, no. 4, pp. 2193–2202, Oct. 2012.
limit frequency. In this situation, the magnitude of the consumer [7] D. Borkowski, A. Wetula, and A. Bien, “New method for noninvasive
impedance may be smaller than or equal to the magnitude measurement of utility harmonic impedance,” presented at the Power
Energy Soc. Gen. Meeting, San Diego, CA, USA, 2012.
of the utility impedance . In order to estimate the utility [8] A. Hyvarinen and E. Oja, “Independent component analysis: Algo-
harmonic impedance by using the proposed method in this sit- rithms and applications,” Neural Netw., vol. 13, pp. 411–430, May/Jun.
uation, replacing the metering point to the point between the 2000.
[9] H. W. Liao and D. Niebur, “Load profile estimation in electric transmis-
cable and consumer is recommended. Hence, the combination sion networks using independent component analysis,” IEEE Trans.
of cable and utility impedance is calculated by proposed method Power Syst., vol. 18, no. 2, pp. 707–715, May 2003.

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.
 1852 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 30, NO. 4, AUGUST 2015

[10] E. Gursoy and D. Niebur, “Harmonic load identification using complex Saeid Esmaeili was born in Rafsanjan, Iran, in
independent component analysis,” IEEE Trans. Power Del., vol. 24, 1976. He received the B.Sc. degree in power elec-
no. 1, pp. 285–292, Jan. 2009. trical engineering from K. N. Toosi University
[11] E. Gursoy, “Independent component analysis for harmonic source of Technology, Tehran, Iran, in 1999, the M.Sc.
identification in electric power systems,” Ph.D. dissertation, Elect. degree in power electrical engineering from Iran
Comput. Eng. Dept., Drexel University, Philadelphia, PA, USA, 2007. University of Science and Technology, Tehran, in
[12] E. Bingham and A. Hyvarinen, “A fast fixed-point algorithm for inde- 2002, and the Ph.D. degree in electrical engineering
pendent component analysis of complex valued signals,” Int. J. Neural from Amirkabir University of Technology (Tehran
Syst., vol. 10, pp. 1–8, 2000. Polytechnic), Tehran, Iran, in 2007.
[13] Electromagnetic Compatibility (EMC)—Part 4: Section 7: Testing and Currently, he is an Assistant Professor in the De-
Measurement Techniques—General Guide on Harmonics and Inter- partment of Electrical Engineering at the Shahid Ba-
harmonics Measurements and Instrumentation, for Power Supply Sys- honar University of Kerman, Kerman, Iran. His research interests include anal-
tems and Equipment Connected Thereto, IEC 61000-4-7, 2002. ysis and optimization of electrical power systems, power quality, and dynamics
[14] T. Pfajfar, B. Blažič, and I. Papič, “Harmonic contributions evaluation of electrical power systems.
with the harmonic current vector method,” IEEE Trans. Power Del.,
vol. 23, no. 1, pp. 425–433, Jan. 2008.

Seyed Hossein Hosseinian was born in Iran in

Farzad Karimzadeh was born in Kerman, Iran, 1961. He received the B.Sc. and M.Sc. degrees in
in 1986. He received the B.Sc. degree in electrical electrical engineering from Amirkabir University
engineering from Shahid Bahonar University of of Technology (AUT), Tehran, Iran, in 1985,
Kerman, Kerman, Iran, in 2008, the M.Sc. degree in and 1988, respectively, and the Ph.D. degree
electrical engineering from K. N. Toosi University in electrical engineering from the University of
of Technology, Tehran, Iran, in 2011, and is currently Newcastle, New Castle, U.K., in 1995.
pursuing the Ph.D. degree in electrical engineering Currently, he is a Professor of the Electrical
from the Graduate University of Advance Tech- Engineering Department at AUT. His special
nology, Kerman. fields of interest include transients in power
His research interests are power quality and systems, power quality, as well as restructuring
harmonics. and deregulation in power systems.

Authorized licensed use limited to: NWFP UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on August 01,2022 at 15:35:31 UTC from IEEE Xplore. Restrictions apply.