19th International Conference on Renewable Energies and Power Quality (ICREPQ’21)

Almeria (Spain), 28th to 30th July 2021

Renewable Energy and Power Quality Journal (RE&PQJ)
ISSN 2172-038 X, Volume No.19, September 2021

Detection of the initial region of the current transformer core saturation

I. Odinaev1, Andrew V. Pazderin1, Pavel V. Murzin1, Valeriy A. Tashchilin1, Vladislav O.
Samoylenko1 and B. Ghoziev1
1
Department of Automated Electrical Systems
Ural Federal University
Mira str., 19, 620002 Yekaterinburg (Russia)
Phone/Fax number: +007 343 359-16-15, e-mail: [email protected], [email protected],
[email protected], [email protected], [email protected], [email protected].

Abstract. In the first part of the paper the detailed analysis of solved using detection algorithms based on the methods of
the existing current transformer (CT) saturation detection mathematical analysis (difference quotient) [3] – [11] and
methods with their classification is presented. A new saturation mathematical statistics [12]–[16].
detection method has been proposed in second part of the paper.
Mathematical description of the method is given. In the next The saturation detection method proposed in [3]–[10]
section of the paper a CT model with test scheme is presented. operates based on difference function [18]. During testing
To identify the reliability of proposed method against nose and of [18] it was identified that the method has a high
remanent flux density experiments have been produced in the
fourth section. Comparative analysis between proposed and
sensitivity to white noise which influences on the method
existing methods is also given. Finally, in conclusion a detailed results. In order to avoid this effect, the authors of [9] and
description of the method is given regarding its behavior with [10] using Savitzky–Golay filter [21] performed
respect to remanent flux and noise. preliminary smoothing of measured current. However, as
shown by the test results, the method is unable to provide
Key words. Current transformer (CT), saturation detection during severe saturation. Moreover, if during
detection, flux density, unsaturated section. saturation occurrence measured current has no sharp
decline, the method failing to provide detection (Fig. 1, b).
1. Introduction
In [11] for detecting CT saturation the difference function
proposed in [22] is used. Difference between [18] and [22]
One of the main current measurement sources in electrical functions is that the first one applies current derivative,
power systems are electromagnetic CTs. Information which in saturation cases has a high splash, the last one
obtained from these CTs is used for both protection and based on US section predicts measured current samples in
metering systems. However, as shown in [1], during faults saturated section that allow to determine the difference
the core of such CTs can saturate and as a result a between predicted and saturated samples. Based on the
maloperation of the protection system (PS) may take obtained information the saturation detection problem is
place. According to [2] CT saturation affects sensitivity solved. During testing the method in [22] it was
and time operation of PS as well as on the accuracy of the discovered that the measured current with noise
fault localization method. significantly effects on the method efficiency and as result
it leads to the method efficiency losses.
In order to avoid the negative impact of CT saturation on
the PS operation, it is proposed to use mathematical In [12] based on two adjacent extreme points of the first
methods to detect CT saturation [3] – [16] with derivative of the measured current, obtained last (last and
subsequent correction of the measured current [17] – [20]. penultimate), the following two criteria are formulated:
From the point of digital signal processing view, the CT 1. The length of timeline between these extreme
error compensation problem in saturation mode can be
points must be less than 10 ms i.e.
divided into three sub-problems - segmentation, selection,
and filtering. Segmentation is an operation of dividing
measured current into normal and transient sections
(t last − t penultimate  10ms ).
(modes), selection is a choosing measured current samples 2. The ratio of these extreme points is greater than 1
connected to CT unsaturated section (US) and filtering is a
correction of measured current samples connected to CT
 i 
 last  1 .
saturated section. i.e.
 i penultimate 
 
The most difficult of the listed ones are the selection and
filtering sun-problems. Typically, the selection problem is

https://doi.org/10.24084/repqj19.322 477 RE&PQJ, Volume No.19, September 2021

When criteria 1 and 2 are simultaneously true, the CT
saturation is fixed. The justification for the method is that
differentiation of the measured current reduces its DC
component by a factor less than 1 in the power of the Saturation criterion
derivative order. Thus, due to multiple differentiation of
the measured current, its constant component is reduced to
an insignificant level. By eliminating the filtering time,
the method can detect CT saturation with a time offset of t
= 1.2 ms, that results to an additional delay in the PS
operation. However, the method provides only saturation
existence in the CT core but not the end determination of
the US section. This complicates the selection of well
measured samples of secondary current.

To detect CT saturation in the two-winding power

transformer differential protection scheme the detection
method is proposed in [13] and [14]. Based on the Pearson
correlation coefficient the method establishes connection
between currents on high and low sides of the power Fig. 1. Graphic interpretation of the saturation detection
transformer. It is important to note that usage of the method [16]
method in the busbar differential protection scheme
requires processing of large amount of data that leads to 2. CT saturation detection based on MFD
an increase of computational efforts. Moreover, solution
of the saturation detection problem by the method will be Within this section the detailed description of the MFD
complicated in the case of a single CT. method is proposed. To implement the method, the
following CT parameters are required to be known:
Authors in [15] propose a method that is based on current
behavior in the differential protection scheme. During • magnetizing curve H = f (λ);
external faults resulting to CT saturation the shape of the • secondary circuit resistance R2, Ω;
differential current has two specific sections. The first of • secondary circuit resistance L2, Ω∙s/rad;
them corresponds to US section and the second to • path length l, m;
saturated one. It is worth noting that the method allows to • cross section, s, m2;
detect saturation of those CTs that included in the • turning number of the secondary winding, w2.
differential protection. Furthermore, the method detects
CT saturation in the protection scheme but not the definite The detection is based on the fact that when the CT core
saturated CT. In this case the filtering problem of the enters the saturation region its flux density λ changes
measured current with distorted sections will be quickly and instead of a sine shape (i.e. unsaturated mode)
complicated. it has stationary areas that resemble a trapezoid. In these
areas, the value of the derivative and variance of the λ
The method proposed in [16] is based on the analysis of sharply decreases. For presentation purposes, the upper
the location density of the measured current samples. In plots of the Fig. 2 and Fig. 3 show reference (Ref.) and
severe CT saturation the measured current samples are measured (Meas.) currents of the CT, the lower plots show
located close to time axis and in CT normal conditions the change in the reference flux density λ. As can be seen
these samples near the mentioned axis have a wide scatter. from the lower plots, in normal conditions the λ has a sine
Such scatter is due to the fact that measured current shape and when the CT core saturation occurs its shape
samples are located on the linear part of sine current in takes a form resembling a trapezoid. The second case
normal conditions. Fig. 1 shows the measured current means the measure of the flux density variance σ in
samples (Meas.) with their distribution density and relation to its population mean e in saturated areas is
reference current (Ref.). It is worth noting that the method
greatly reduced which can be seen on the lower plot on the
can detect a severe CT saturation or a saturation that
Fig. 3.
severely limits the measured current samples and as a
result these samples are located close to time axis, Fig. 1,
To estimate the flux density variance σ its calculated λc
a. During CT saturation that leads to slow offsetting of the
must be known. According to [23] the λc can be obtained
measured current from reference, it is impossible to detect
using expression (1).
saturation by the method, Fig. 1, b.
R2
B ( n) = i2 (n) + i2 (n − 1)  t +
In this study on the basis of monitoring the flux density w2 s
(MFD) in the CT core a solution of the saturation , (1)
L
detection problem was proposed. More detailed + 2  i2 (n) − i2 (n − 1) + B(n − 1)
description of the detection pressure is given in the next w2 s
section.

https://doi.org/10.24084/repqj19.322 478 RE&PQJ, Volume No.19, September 2021

 MFD method. Varying N allows to change the method
sensitivity against the various levels of CT core saturation.

Knowing the deviation of the calculated flux density σ

within the interval N, it is possible to determine the
presence of CT core saturation. In order to do so, it is
necessary to compare σ with a threshold value Th, which
corresponds to the slope of the magnetizing curve section
in the range from k to m. The point k, located at the
beginning of the section H = f (λ), corresponds to the first
sample of the magnetization curve and the point m
corresponds to the maximum value of the λ at the rated
Fig. 2. Upper plot – reference and measured currents; load of the CT in the case of peak load operation of a grid.
lower plot – the flux density λ in normal conditions
To calculate Th the usage of the λ(n) values, located on
the magnetizing curve section in the k and m range, is
required. For presentation purposes, the Fig. 4 shows the
magnetizing curve on the basis of which k and m are
determined.

k m
Saturation section

Fig. 3. Upper plot – reference and measured currents; Fig. 4. CT magnetizing curve
lower plot – the flux density λ in CT saturation
The Th is calculated according to equation (4).
In (1), due to unknown initial flux density which is 2

 ( k − e )
m
represented as λ(n-1), there will be a shift between the
(4)
reference λ and calculated λc. However, the shape of the Th = k =1
.
calculated and reference flux density will match exactly m
which is an important fact for the MFD method. As the The occurrence of saturation is determined in accordance
MFD method estimates the location of λc samples relative with (5).
to its population mean e , excluding the initial flux
density λ(n-1) in (1) will not impact on the efficiency of
 ( n )  Th . (5)
the method.
The window length N must be selected so that during
Thus, knowing the behavior of the flux density in the CT faults not leading to CT core saturation the maloperation
of the MFD is blocked, i.e. Th would not exceed the σ(n).
core, its deviation σ from the population mean e can be
To avoid maloperation of the method under normal and
calculated using expression (2). peak load conditions of a grid, the start of the method is
2
1 N triggered by the following expression:
 (n − N ) =   ( n − m ) − e ,
N − 1 m =1
(2)
C  2  I max , (6)

where Imax is an RMS value of current according to peak

where n is the number of the current reading of the λc; m is load conditions of a grid; C is the predicted amplitude of
an integer that serves as a window function; N is the the measured current.
length of the window; e is the population mean.
The C prediction can be explained as follows: let there be
some sine signal with its set of the instantaneous values
The e is calculated by equation (3).
and analytical expression (7).

1 N i ( t ) = С  sin (t + 0 ) , (7)

e =
N
  ( n − m) .
m =1
(3)
where ω is an angular rate; φ0 is an initial phase.
It is worth noting that the length of the window N,
including in deviation calculation, is a crucial factor of the Make an arbitrary measurement and set its measurement
time t = 0, then expression (7) can be rewritten as follows:

https://doi.org/10.24084/repqj19.322 479 RE&PQJ, Volume No.19, September 2021

i ( 0 ) = С  sin (0 ) . (8)

Derivative of expression (7) is:

di
i '(t ) = = С cos (  t +  0 ) . (9)
dt
At t = 0 expression (9) cab be rewritten as:
i ' ( 0 ) = С cos (0 ) . (10)
Fig. 5. Test model
The value of i(0) is known it is an instantaneous value for
which t = 0, further it will be denoted as i0, the derivative 4. Testing of the MFD method
i’(0) will be denoted as i'0, and so the tangent of the slope
angle, drawn between two points that located from left i-1 In this section, the testing of the MFD method was
and right i+1 of the i0, can be determined as following: performed. To determine the method sensitivity with
di i i+1 − i−1 respect to nose (δ) and level of the initial flux density λ0 in
i '0 =  = , (11) the CT core, a δ and λ0 were included in the measured
dt t 2 t current and in the CT core respectively. Including noise
was performed by equation (16).
where Δt is a sampling step.
imeas ( t ) = i2 ( t )  (1 + 1   ) + max ( i2 ( t ) )  ( 2   ) , (16)
On the basis of the expressions (8) and (10) following
equations can be received: where i2(t) is a measured current; δ1 and δ2 are priori
sin (0 ) + cos (0 ) = 1 ,
2 2
(12)
specified values of the noise level; ϑ is a random number
in the range of -1 to 1. ϑ has a uniform distribution.
2 2
 i0   i '0  Checking the MFD sensitivity to a noise
 С  +  С  = 1 (13)
    Within this experiment, based on the CT1 of the test
model, checking the method sensitivity to a noise was
2 provided. To do so a noi se with a level of δ1 = 3% and δ2
 i' 
( i0 ) +  0  = С2
2
(14) = 1% was added to the measured current. Initial phase of
 the modeled fault φA0 was included for phase A and
changed in the range of 0° to 90°. The initial phase range
2 (0o – 90o) was chosen based on dividing the entire cycle of
 i' 
C= ( i0 ) + 0  . the sine current into 4 parts, i.e. 0o – 90o, 90o – 180o, 180o
2
(15)
 – 270o and 270o – 360o. The simulation of the other three
parts was not performed under the assumption that their
Thus, based on US section the predicted amplitude of the influence on the fault transient is similar or reflection of
measured current can be obtained. To improve the predict the chosen part. The CT core initial flux density λ0 for all
accuracy it is proposed to calculate the overage derivative CT was excluded and specified λ0 = 0 T Fig. 6 shows the
i'0 using 3-4 measured current samples related to US results of experiment, where reference and measured
section. currents and the result of the MFD method for phases A
and B during a fault between them are presented. As in
this case of fault the CT of the third phase operates with
3. Test Model description normal conditions, its currents and saturation detection
were not performed.
To perform experiments allowing to determine the MFD
ability a grid was modeled in Matlab/Simulink It is clear seen from Fig. 6 that a noise in the measured
environment, that includes load, system S, transmission current has no negative impact on the result of MFD
line and two-winding power transformer with CTs on both operation. It is worth to note that the variation of initial
high (CT1) and low (CT2) sides. The installed load value phase φA0 has no impact on the MFD as well. The last four
was set equal to 70% of power transformer nominal rate, plots of the Fig. 6 show that during CTs cores saturation
cosφ = 0.8. The configuration elements of the grid and the shape of the measured current in the phase B is
their main parameters are presented on Fig. 5. significantly different from that in the phase A.
Nevertheless, the method allows to detect the start of the
Modeling of CT1 and CT2 was performed on the basis of CT saturation in both cases.
the Jiles-Atherton theory that considers hysteresis loop
and losses on the CT core excitation [24], [25]. The Jiles-
Atherton theory includes next parameters: density flux λ;
magnetizing field H; effective field He; magnetization M
with its reversible Mrev and irreversible Mirr components as
well as anhysteretic magnetization Man.

https://doi.org/10.24084/repqj19.322 480 RE&PQJ, Volume No.19, September 2021

 Fig. 7. CT currents and its saturation detection with
variation of λ0

The experimental results presented on the Fig. 7 show that

the MFD method remains ability to detect the start of the
CT core saturation with initial flux density and even
during short US section (3-4 samples) the method still
stays able to perform accurate detection of the start
saturation.

5. Conclusions
Within the paper a new CT saturation detection method
has been proposed. For correct operation of the method, a
threshold calculation and the method setting are given.
The method sensitivity was checked with respect to a
noise and a level of the initial flux density. Difference
Fig. 6. CT currents and its saturation detection with function-based methods are analyzed. The advantages and
variation of φA0 disadvantages of the existing saturation detection methods
are described.
Checking the MFD sensitivity to an initial flux density
To avoid maloperation of the method in normal and peak
In this experiment it was intended to check the MFD conditions of a grid a new amplitude prediction method
method sensitivity to the CT core initial flux density λ0 has been proposed which was implemented on the basis of
with various levels. Variation of the λ0 was performed in numerical differentiation. It is well known that the
the range of -0.5 to 1.8 T only in the CT core established numerical differentiation has high sensitivity to noise as a
on the phase A. An initial phase was a constant and result their accuracy can be significantly decreased. To
specified as φA0 = 0o. Fig. 7 shows the reference and reduce this effect, a median filtering of the measured
measured currents as well as the result of the MFD current was used. Afterwards, based on 3 or 4 samples of
method for phase A during fault between phases A and B.

https://doi.org/10.24084/repqj19.322 481 RE&PQJ, Volume No.19, September 2021

the measured current related to unsaturated section the [11] J. Herlender, J. Iżykowski, and K. Solak, “Compensation
average predicted amplitude was obtained. of the current transformer saturation effects for
transmission line fault location with impedance-differential
Experiments results of the paper show that proposed relay,” Electr. Power Syst. Res., vol. 182, no. May 2019,
2020, doi: 10.1016/j.epsr.2020.106223.
saturation detection method has a high reliability with [12] S. Bahari, T. Hasani, and H. Sevedi, “A New Stabilizing
respect to initial flux density and noise. It is clearly seen Method of Differential Protection Against Current
from these experiments that during severe saturation, Transformer Saturation Using Current Derivatives,” 2020
when unsaturated section contains only 4 samples of the 14th Int. Conf. Prot. Autom. Power Syst. IPAPS 2020, pp.
measured current at an 80 samples/cycle, the proposed 33–38, 2019, doi: 10.1109/IPAPS49326.2019.9069379.
method remains capable to detect the start of saturation. [13] D. Behi, M. Allahbakhshi, A. Bagheri, and M. Tajdinian,
“A new statistical-based algorithm for CT saturation
detection utilizing residual-based similarity index,” 2017
Acknowledgement 25th Iran. Conf. Electr. Eng. ICEE 2017, pp. 1072–1077,
2017, doi: 10.1109/IranianCEE.2017.7985200.
Proposed results are the outcome of work on the topic [14] A. A. A. Etumi and F. J. Anayi, “The application of
"Development and testing algorithms and technical correlation technique in detecting internal and external
solutions for digital control of power facilities" within faults in three-phase transformer and saturation of current
integrated project "Improvement of approaches to digital transformer,” IEEE Trans. Power Deliv., vol. 31, no. 5, pp.
management of power facilities". 2131–2139, 2016, doi: 10.1109/TPWRD.2016.2572608.
[15] T. Zheng, T. Huang, Y. Ma, Z. Zhang, and L. Liu,
“Histogram-Based Method to Avoid Maloperation of
References Transformer Differential Protection Due to Current-
Transformer Saturation under External Faults,” IEEE
[1] R. Hunt, “Impact of CT errors on protective relays - Case Trans. Power Deliv., vol. 33, no. 2, pp. 610–619, 2018,
studies and analyses,” IEEE Trans. Ind. Appl., vol. 48, no. doi: 10.1109/TPWRD.2017.2712806.
1, pp. 52–61, 2012, doi: 10.1109/TIA.2011.2175879. [16] C. Hong, L. Haifeng, J. Hui, P. Jianchun, and H. Chun, “A
[2] F. das Guerra Fernandes Guerra and W. Santos Mota, scheme for detection and assessment of current
“Current transformer model,” IEEE Trans. Power Deliv., transformer saturation,” Proc. - 9th Int. Conf. Meas.
vol. 22, no. 1, pp. 187–194, 2007, doi: Technol. Mechatronics Autom. ICMTMA 2017, pp. 90–
10.1109/TPWRD.2006.887092. 93, 2017, doi: 10.1109/ICMTMA.2017.0029.
[3] H. Dashti, M. Sanaye-Pasand, and M. Davarpanah, [17] J. Pan, K. Vu, and Y. Hu, “An efficient compensation
“Current transformer saturation detectors for busbar algorithm for current transformer saturation effects,” IEEE
differential protection,” Proc. Univ. Power Eng. Conf., no. Trans. Power Deliv., vol. 19, no. 4, pp. 1623–1628, 2004,
1, pp. 338–343, 2007, doi: 10.1109/UPEC.2007.4468971. doi: 10.1109/TPWRD.2004.835273.
[4] M. Hossain, I. Leevongwat, and P. Rastgoufard, “A [18] Y. C. Kang, U. J. Lim, S. H. Kang, and P. A. Crossley,
Current Transformer (CT) Saturation Detection Method “Compensation of the distortion in the secondary current
for Bus Differential Protection,” Clemson Univ. Power caused by saturation and remanence in a CT,” IEEE Trans.
Syst. Conf. PSC 2018, no. M, pp. 1–5, 2019, doi: Power Deliv., vol. 19, no. 4, pp. 1642–1649, 2004, doi:
10.1109/PSC.2018.8664064. 10.1109/TPWRD.2004.835266.
[5] R. Abd Allah, S. Moussa, E. H. Shehab-Eldin, and M. N. [19] E. Hajipour, M. Vakilian, and M. Sanaye-Pasand,
G. Hamed, “Advanced detection and compensation scheme “Current-Transformer Saturation Compensation for
for current transformers saturation,” Proc. 11th Int. Middle Transformer Differential Relays,” IEEE Trans. Power
East Power Syst. Conf. MEPCON’2006, vol. 2, pp. 481– Deliv., vol. 30, no. 5, pp. 2293–2302, 2015, doi:
486, 2006. 10.1109/TPWRD.2015.2411736.
[6] S. Biswal and M. Biswal, “Detection of current [20] G. Baoming, A. T. de Almeida, and T. E. Ferreira
transformer saturation phenomenon for secured operation Fernando J., “Estimation of primary current in saturated
of smart power network,” Electr. Power Syst. Res., vol. current transformer using flexible neural network,” Trans.
175, no. May, p. 105926, 2019, doi: Inst. Meas. Control, vol. 28, no. 1, pp. 81–91, 2006, doi:
10.1016/j.epsr.2019.105926. 10.1191/0142331206tm164oa.
[7] G. L. Macieira and A. L. M. Coelho, “Evaluation of [21] R. W. Schafer, “What Is a Savitzky-Golay Filter? [Lecture
numerical time overcurrent relay performance for current Notes],” no. July, pp. 111–117, 2011.
transformer saturation compensation methods,” Electr. [22] A. Wiszniewski, W. Rebizant, and L. Schiel, “Correction
Power Syst. Res., vol. 149, pp. 55–64, 2017, doi: of current transformer transient performance,” IEEE Trans.
10.1016/j.epsr.2017.04.005. Power Deliv., vol. 23, no. 2, pp. 624–632, 2008, doi:
[8] L. Yang, J. Zhao, P. A. Crossley, and K. Li, “A current 10.1109/TPWRD.2008.915832.
transformer saturation detection algorithm for use in [23] Y. C. Kang, J. K. Park, S. H. Rang, A. T. Johns, and R. K.
current differential protection,” Proc. - Int. Conf. Electr. Aggarwal, “An algorithm for compensating secondary
Control Eng. ICECE 2010, vol. 3, no. 3, pp. 3142–3146, currents of current transformers,” IEEE Trans. Power
2010, doi: 10.1109/iCECE.2010.767. Deliv., vol. 12, no. 1, pp. 116–124, 1997, doi:
[9] G. Lin, Q. Song, D. Zhang, F. Pan, and L. Wang, “A 10.1109/mper.1997.560670.
hybrid method for current transformer saturation detection [24] D. C. Jiles and D. L. Atherton, “Theory of ferromagnetic
and compensation in smart grid,” in The 2017 4th hysteresis (invited),” J. Magn. Magn. Mater., vol. 61, no.
International Conference on Systems and Informatics 6, pp. 48–60, 1986, doi: 10.1063/1.333582.
(ICSAI 2017), 2017, pp. 369–374. [25] U. D. Annakkage, P. G. McLaren, E. Dirks, R. P.
[10] B. M. Schettino, C. A. Duque, and P. M. Silveira, Jayasinghe, and A. D. Parker, “A Current Transformer
“Current-Transformer Saturation Detection Using Model Based on the Jiles-Atherton Theory of
Savitzky-Golay Filter,” IEEE Trans. Power Deliv., vol. 31, Ferromagnetic Hysteresis,” IEEE Trans. Power Deliv., vol.
no. 3, pp. 1400–1401, 2016, doi: 15, no. 1, pp. 57–61, 2000, doi: 10.1109/61.891531
10.1109/TPWRD.2016.2521327.

https://doi.org/10.24084/repqj19.322 482 RE&PQJ, Volume No.19, September 2021