IET Generation, Transmission & Distribution

Research Article

ISSN 1751-8687
Algorithm for transformer differential Received on 21st September 2015
Revised on 25th April 2016
protection based on wavelet correlation Accepted on 12th May 2016
doi: 10.1049/iet-gtd.2015.1147
modes www.ietdl.org

Daniel Guillén, Hector Esponda, Ernesto Vázquez, Gina Idárraga-Ospina ✉

Programa Doctoral de Ingeniería Eléctrica (DIE), Universidad Autónoma de Nuevo León (UANL), Monterrey, Nuevo Leon, México
✉ E-mail: [email protected]

Abstract: In this study, a new algorithm for discrimination between internal faults and inrush current conditions in power
transformers is presented. First, the proposed algorithm removes the pre-fault conditions of the fault currents. Then, those
currents are processed using the discrete wavelet transform to obtain the wavelet matrix. The correlation matrix is
computed from wavelet matrix to know the eigenvalues associated with each frequency scale and to reduce the
analysed information. This algorithm, which the authors are going to call wavelet correlation modes, is used to identify
inrush and internal fault currents in power transformer. To validate the proposed algorithm, simulations are carried out
in power systems computer-aided design software, and the obtained signals are analysed in order to evaluate its
performance and sensitivity.

1 Introduction methodologies based on WT utilise the detail coefficients of the

wavelet trees.
Differential protection has been adopted as the main protecting In this paper, a new algorithm using the wavelet matrix of the
method of power transformer for several years, because it has DWT (instead of the wavelet trees), and the incremental
simple operation principle and high sensitivity. However, it suffers differential currents is proposed. Once state-stable are removed
from a vital drawback; it will operate when there is an inrush from differential current using a Delta filter [14], the algorithm
current due to the energisation of the transformer. Moreover, those built the wavelet matrix [15] using DWT method. Eigenvalues of
currents are detected only by the current-transformer (CT) of the this matrix are used to produce the wavelet correlation matrix, in
supply side and could be interpreted as an internal fault [1]. which each eigenvalue is associated with the frequency range of
There are two typical ways to avoid the inrush current during the transient phenomenon. Finally, the maximum eigenvalues
transformer energisation, one of them is to apply a time delay define the wavelet correlation modes (WCMs) and they are used to
considering that the high magnitude of the inrush current appears detect a transient condition based on the minimum threshold. The
only during the first cycle; moreover, this phenomenon will decay proposed algorithm, which will be called by the authors WCM
after a few cycles [2]. The other one is to use the harmonic algorithm, is applied to evaluate the differential protection with a
content provided by the differential current signals to set a sliding window of half cycle. Finally, the performance of the new
threshold and to avoid a mis-operation of the protective relay. algorithm is analysed in simulation results using power systems
Currently, the differential relay identifies inrush current based on computer-aided design (PSCAD) software; CT saturation,
second-harmonic blocking [3]. However, this component also overexcitation, non-linear load scenarios are considered. In all
appears when CT is saturated, so that this condition may also cases, the algorithm made a correct identification of internal faults
produce a mis-operation of the protection [4]. Furthermore, with and inrush currents. Therefore, the proposed algorithm may be
the new core materials used in actual transformers [5], harmonic considered for new differential protection schemes.
blocking can fail due to the low harmonic content.
According to the aforementioned, several approaches based on
recognition of the signal waveform, to distinguish faults from 2 Transformer differential protection
inrush currents, have been proposed to solve this problem. For
instance, in [6] differential current waveforms have been analysed Fig. 1 shows a typical differential relay connection diagram. Relay
to identify specific characteristics such as asymmetric arrangement compares an operating current with a restraining current. The
regarding the time. In the same way, correlation techniques have operating current (also called differential current), IO, and the
been introduced to analyse the differential current waveforms, i.e. restraining current, IR, are obtained as [16]
the current waveform is divided into its saturation and


non-saturation stages [7]. There is also possible to find research IO =
I1 + I2

using pattern recognitions such as fuzzy logic, which are also

(1)
IR = K
I1 − I2

applied trying to identify specific characteristics into the signal
waveforms [8, 9].
On the other hand, the wavelet transform (WT) has been included The relay generates a tripping signal if the operating current, IO, is
as a processing tool for differential protection, e.g. in [10] a new greater than a percentage of the restraining current, IR, according to
method based on wavelet packet is implemented to detect inrush
currents from internal faults. Other algorithms use the discrete WT IO = kIR + IP (2)
(DWT) for identifying inrush current, e.g. in [11] a new algorithm
based on the median absolute derivation of the detail coefficients where k is the relay operating characteristic that consists of a straight
is proposed. Also, the energy of detail coefficients has been used line having a slope equal to k. Intersection of this characteristic with
to discriminate inrush currents from internal faults based on vertical axis (IO) defines the relay minimum pickup current, IP [17].
correlation coefficients [12, 13]. It is worth to highlight that the The relay percentage restraint characteristic typically has an excellent

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
& The Institution of Engineering and Technology 2016 2871
 4 Wavelet correlation modes
If a signal f (n) is scaled up to a defined decomposition level, then, it
will be producing a wavelet matrix M(J + 1, n), this matrix is
analysed using its correlation matrix defined by [20]

M × MT
Y= (5)
n

where n is the total sample numbers. Therefore, it has a matrix

Y(J + 1, J + 1) which contains the scaled frequency information of
Fig. 1 Differential relay connection diagram the signal. To discriminate internal faults from inrush currents, the
proposed algorithm uses incremental differential currents [14].
Then, it has three correlation matrices Y1, Y2 and Y3 which are
behaviour, but it has problems to discriminate fault currents from the used to form a total wavelet correlation matrix defined by
false differential currents caused by inrush and transformer
overexcitation. 3
nY i
YT = i=1
(6)
3

In this way, all information will be contained into the total wavelet
3 Wavelet transform correlation matrix, so as its modes associated with each frequency
scale. Moreover, the modes are defined by the eigenvalues of the
WT is very useful for processing signals and this powerful technique wavelet correlation matrix as follows [21]
is able to provide time–frequency information about signals at
specific applications. WT has a continuous version defined by [18]
Y T = LLT (7)
1  
1 t−b where L and T are the left and right eigenvector matrices,
WT(a, b) = √ t(t)c∗ dt (3)
a −1 a respectively, and Λ is the eigenvalues diagonal matrix
⎛ ⎞
l1 0 0 ··· 0
where a is the scale parameter regarding frequency, b is the ⎜0 l2 0 ··· 0⎟
translation parameter in time to the signal f (t) and ca,b(t) is the ⎜ ⎟
⎜ .. ⎟
chosen mother wavelet and * represents the operation of complex L=⎜
⎜0 0 . ··· 0⎟ ⎟ (8)
conjugate. For discrete signals, a and b parameters take discrete ⎜ . .. .. .. ⎟
values 2j and i2j, respectively. Therefore, the WT in its discrete ⎝ .. . . lr−1 . ⎠
form is defined as follows 0 0 0 0 lr

 Therefore, if there is a transient phenomenon, the incremental

DWT(j,i) = 2−j/2 f (n)c(2−j n − i) (4) differential currents produce changes in frequency, these changes
n
modify features in the wavelet matrix that can be analysed using
the eigenvalues produced by the wavelet correlation matrix, so that
Moreover, the DWT coefficients are obtained from filtering each eigenvalue is associated with the frequency range of the
operations and are divided in approximation (cA) and detail analysed signals. For instance, in Table 1 the wavelet correlation
coefficients (cD) [19]. These coefficients are computed following matrix for an internal fault and an inrush current is shown. In fact,
the decomposition algorithm presented in Fig. 2. In this case, a an inrush current produces lower values into the wavelet
sample frequency of 960 Hz is assumed and also its corresponding correlation matrix than an internal fault.
approximation and detail coefficients for each frequency scale are Hence, the information is reduced through its eigenvalues and
shown. only the greatest eigenvalue is selected for identifying internal
According to Fig. 2, the produced coefficients are used to form a faults from inrush currents, since it contains the greatest energy [22].
wavelet matrix and this matrix will be filled with cAJ and cWj of each
scale. The size of this wavelet will be (J + 1, N), where J is the
predefined decomposition level and N is the length of the signal. 5 Proposed algorithm
Therefore, to reduce the features produced by inrush currents and
faults, the wavelet matrix is analysed using its correlation matrix The proposed algorithm is summarised in Fig. 3, and it can be
and eigenvalues; the process is described in the following section. divided in three stages. In the first stage, the line current signals

Fig. 2 Three decomposition levels for a signal

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
2872 & The Institution of Engineering and Technology 2016
Table 1 Wavelet correlation matrices produced by internal fault and inrush current
Internal fault Inrush current

3.94 × 10−1 1.05 × 10−2 −1.83 × 10−3 1.32 × 10−1 2.77 × 10−3 5.94 × 10−5 8.67 × 10−7 −4.63 × 10−4
1.05 × 10−2 1.47 × 100 −1.12 × 100 −2.12 × 101 5.94 × 10−5 6.01 × 10−2 −1.27 × 10−3 2.24 × 10−1
−1.83 × 10−3 −1.12 × 100 9.96 × 100 1.21 × 102 8.67 × 10−7 −1.27 × 10−3 8.77 × 10−1 2.21 × 10−1
1.32 × 10−1 −2.12 × 101 1.21 × 102 1.35 × 104 −4.63 × 10−4 2.24 × 10−1 2.21 × 10−1 5.17 × 102

are acquired from CTs and these are used to obtain incremental transformer ratings due to the proposed normalisation. In this
differential currents. At the second stage, incremental differential sense, the maximum inrush current can be computed as follows [23]
currents are normalised and their wavelet matrices are computed
using the DWT. At the last stage, for each incremental differential √  
2U (2BN ) + BR − BS
current the wavelet correlation matrix is computed using (5), so as I pk =  (9)
the total wavelet correlation matrix (6). Finally, the eigenvalues (v L )2 + R2 BN
are calculated using (7) and (8), such that the maximum
eigenvalue defines WCM value. This procedure is independent of where U is the applied voltage, L represents the air-core inductance
any frequency variation in the power network. Hence, a minimum of the transformer, R is the DC resistance of the windings, BR is the
threshold is used to detect any transient conditions; when it is remanent flux density of the core, BS represents the saturation flux
done, a second threshold must be verified with the aim to know if density of the core material and BN is the normal rated flux density
the event is an inrush or fault current. of the core.

5.2 Thresholds
5.1 Normalised signals
Energisation and fault conditions produce transient information which
In general, inrush current plays an important role during energisation is analysed for identifying the occurred transient phenomenon.
and it is related with relay settings. This current can be computed Therefore, the proposed algorithm requires a minimum threshold in
using a simplified equation [23] or estimated according to its rated order to detect a transient condition. So, the minimum threshold
current, since an inrush current may reach up to eight times the defines when there is a normal condition into the differential
rated current [24]. By this reason, incremental differential currents protection scheme. In this sense, the proposed algorithm uses a
are normalised regarding the maximum expected inrush current. minimum threshold of 0.0001 (empirical value obtained by
Therefore, the proposed algorithm can be applied to different simulation) and this is used to activate the differential protection.
Hence, to discriminate between inrush currents and internal faults,
the algorithm must be verified a second threshold. To avoid a
mis-operation into the differential protection scheme, and taken into
account that the major DC hysteresis loop of the transformer core
can reach 1.5 times its nominal flux [25]. The second threshold is
adjusted to 2, since the incremental differential currents were
normalised regarding the maximum expected inrush current.

6 Test system

To evaluate the proposed algorithm, test simulations are carried out

in PSCAD software. The diagram test system is shown in Fig. 4; a
100 MVA, 115/13.8 kV, 60 Hz, transformer connection
delta-grounded star is employed for simulation. Thevenin
equivalents are 325 and 10 MVA with a phase shift of 20°. The
selected CT ratios are 500:5 and 4000:5 in order to allow the
saturation effects in CT [26].
CT saturation is taken into account with the aim to analyse this
critical effect and to ensure that the proposed algorithm works
properly during energisation, internal and external fault conditions.
So, the inputs to the algorithm are incremental differential
currents; they are obtained using a delta filter to eliminate the
periodicity in signals [14]. Moreover, an X/R ratio of 2.5 is used
for the test system.

7 Results

To evaluate the performance of the proposed algorithm, a sample

frequency of 7.68 kHz is used with a sliding window of 64
samples, which it is equivalent to half cycle of the fundamental
frequency. Therefore, the maximum decomposition level will be
equal to 3, and the analysed signals will be scaled in the following
frequency ranges: 1.92–3.84 kHz for the detail coefficients at the
first scale (cD1), at the second scale is 0.96–1.92 kHz (cD2), 480–
960 Hz for cD3 and 0–480 Hz for the approximation coefficients
(cA3). Hence, the most important information will be contained at
Fig. 3 WCM algorithm the scale 3 and correspond with the frequency characteristics of

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
& The Institution of Engineering and Technology 2016 2873
Fig. 4 Test system

the studied transient phenomena. By this reason, in this paper, only analysed sceneries, the inrush currents are correctly identified by
the greatest eigenvalue produced by the total wavelet correlation the proposed algorithm, allowing the blocking of the differential
matrix is used. protection operation. In fact, seven cases are presented to evaluate
A total of 108 simulations were made considering different the performance of the proposed algorithm.
transient conditions, organised as follows: 32 transformer
energisation, 32 internal faults, 16 external faults, 16
overexcitations and 12 combined cases (modification of power 7.1 Transformer energisation
system parameters). In the whole cases, the proposed algorithm
clearly distinguishes between inrush and fault currents. The Fig. 5 presents the waveforms and algorithm performance for an
internal faults were simulated as phase-to-ground fault in wye side energisation. The saturation effects in CT’s are clearly shown at
using the transformer model of used PSCAD software. In all the primary and secondary currents of Figs. 5a and b, respectively.

Fig. 5 Transformer energisation

a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
2874 & The Institution of Engineering and Technology 2016
Fig. 6 Internal fault
a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

Fig. 7 Internal fault after energisation

a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
& The Institution of Engineering and Technology 2016 2875
The incremental differential currents are shown in Fig 5c. Finally, 7.5 Change in X/R ratio of the system
the proposed algorithm results are presented in Fig. 5d, the event
is an inrush current due to the greatest eigenvalue is lower than 2. The magnitude and duration of inrush currents are affected by the
X/R ratio. In this case, the X/R ratio is modified in the Thevenin
equivalent of 115 kV. Fig. 9 shows the obtained results for a
transformer energisation with a modification on X/R ratio of 50
7.2 Internal fault over its base value. In this case, the greatest eigenvalue produces
values lower than 2; this is clearly shown in Fig. 9d, so the event
An internal fault in transformer is described in Fig. 6, where a is identified as an inrush current.
line-to-ground fault at 20% of the winding is simulated. In this
case, primary currents have higher current values than rated
current, such as it is shown in Figs. 6a and b. This condition is 7.6 Overexciting
correctly detected since the greatest eigenvalue is higher than 2
(established threshold). When the power transformer primary winding is overexciting flows
more power to the primary than the secondary side, meaning to
differential protection scheme a fault condition. This condition is
7.3 Internal fault after energisation considerated a mis-operation because the fault is not produced by
an internal fault [27]. To evaluate the algorithm performance in
In this case, a three-phase fault is simulated into the differential zone this scenario, in 200 ms was energised the power transformer and
at the secondary side 13.8 kV and one cycle after the energisation in 500 ms occurs the primary winding overexciting. It was
(200 ms). The proposed algorithm results are shown in Fig. 7, the considerate a 150% excitation value. To the traditional differential
algorithm is able to identify an internal fault under this condition; protection, when the overexciting was incepted occurs a
the fault is detected at 1.33 ms after fault inception. mis-operation as is shown in Fig. 10. By another hand, the
algorithm response is below to the threshold and the event is
identified as not fault condition as is shown in Fig. 10d.

7.4 External fault

7.7 Energisation and internal B-phase-fault to ground
Saturation effects in CT’s also may appear when external faults limiting the magnitude current
occur near of the differential zone. In this case, the proposed
algorithm is evaluated during external faults to avoid a false In this case, a critical scenario limiting the current magnitude is
tripping due to CT’s saturation. A three-phase fault is simulated at proposed. In a short period of time occurs two events: a power
500 ms (Fig. 8); though incremental differential currents present transformer energisation and a fault inside the differential
important changes during the first cycles, the condition is protection zone with magnitude current limited by a resistor with a
identified as an inrush current since the greatest eigenvalue does value of 1.5 Ω. To the energisation the algorithm identified it as an
not reach the threshold of 2, this is shown in Fig. 8d. inrush current because the largest eigenvalue magnitude was

Fig. 8 External fault

a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
2876 & The Institution of Engineering and Technology 2016
Fig. 9 Change in X/R ratio of the system
a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

Fig. 10 Overexciting
a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
& The Institution of Engineering and Technology 2016 2877
Fig. 11 Energisation and internal B-phase-fault to ground limiting the magnitude current
a Primary currents
b Secondary currents
c Incremental differential currents
d WCM results and traditional differential protection response

below than the threshold. However when the fault occurs, the results, the WCM algorithm is able to identify internal faults from
algorithm identified the event as an internal fault though the fault inrush current in a quarter of cycle and can be implemented with
magnitude was lower than the inrush current magnitude as is different transformer ratings.
shown in Fig. 11.

8 Discussion results 10 Acknowledgment

All simulations are analysed and it is concluded that the proposed The authors acknowledge the financial support given by the
algorithm is able to identify internal faults from inrush currents. In CONACYT (Mexico)/CONICET (Argentina) through the bilateral
this sense, inrush currents at different time of occurrence are cooperation project 189382.
analysed, so that WCM algorithm produces values of 0.1–0.8, due
to that the incremental differential currents were normalised
regarding the maximum expected inrush current.
In the same way, several internal faults are analysed and their 11 References
results are evaluated, producing higher values than the established
threshold. On the other hand, the proposed algorithm is evaluated 1 Zheng, T., Gu, J., Huang, S.F., et al.: ‘A new algorithm to avoid maloperation of
transformer differential protection in substations with an inner bridge connection’,
with external faults at both sides of the transformer; in these cases, IEEE Trans. Power Deliv., 2012, 27, (3), pp. 1178–1185
the WCM algorithm produces lower values than 0.1. Finally, the 2 Chiesa, N., Mork, B.A., Høidalen, H.K.: ‘Transformer model for inrush current
mean time detection for internal faults is equal to 3.5 ms. In calculations: simulations, measurements and sensitivity analysis’, IEEE Trans.
addition, the proposed algorithm is independent to any frequency Power Deliv., 2010, 25, (4), pp. 2599–2608
3 Lin, X., Huang, J., Zeng, L., et al.: ‘Analysis of electromagnetic transient and
variation in the power system. By this, the authors conclude that adaptability of second-harmonic restraint based differential protection of UHV
the proposed algorithm is able to identify inrush current in a power transformer’, IEEE Trans. Power Deliv., 2010, 25, (4), pp. 2299–2307
quarter of cycle and only requires a half cycle window. 4 Vazquez, E., Mijares, I.I., Chacon, O.L., et al.: ‘Transformer differential protection
using principal component analysis’, IEEE Trans. Power Deliv., 2008, 23, (1),
pp. 67–72
5 Tripathy, M., Maheshwari, R.P., Verma, H.K.: ‘Power transformer differential
9 Conclusions protection based on optimal probabilistic neural network’, IEEE Trans. Power
Deliv., 2010, 25, (1), pp. 102–112
6 Negreiros Alencara, R.J., Holanda Bezerra, U., Damasceno Ferreira, A.M.: ‘A
A new algorithm for differential protection is proposed. The method to identify inrush currents in power transformers protection based on the
algorithm uses the highest eigenvalue produced by the total differential current gradient’, Electr. Power Syst. Res., 2014, 111, pp. 78–84
wavelet correlation matrix and its results are validated in 7 Bi, D.Q., Zhang, X.A., Yang, H.H., et al.: ‘Correlation analysis of waveforms in
simulations through a test system. Furthermore, different scenarios nonsaturation zone-based method to identify the magnetizing inrush in
transformer’, IEEE Trans. Power Deliv., 2007, 22, (3), pp. 1380–1385
are analysed in order to avoid a mis-operation during inrush 8 Barbosa, D., Netto, U.C., Coury, D.V., et al.: ‘Power transformer differential
current. The algorithm takes a sliding window of half cycle and protection based on Clarke’s transform and fuzzy systems’, IEEE Trans. Power
produces good results in all simulated cases. On the basis of the Deliv., 2011, 26, (2), pp. 1212–1220

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
2878 & The Institution of Engineering and Technology 2016
 9 Bejmert, D., Rebizant, W., Schiel, L.: ‘Transformer differential protection with 19 Gaouda, A.M., Salama, M.M.A.: ‘DSP wavelet-based tool for monitoring
fuzzy logic based inrush stabilization’, Electr. Power Energy Syst., 2014, 63, transformer inrush currents and internal faults’, IEEE Trans. Power Deliv., 2010,
pp. 51–63 25, (3), pp. 1258–1267
10 Saleh, S.A., Scaplen, B., Rahman, M.A.: ‘A new implementation method of 20 Hendriks, R.C., Gerkmann, T.: ‘Noise correlation matrix estimation for
wavelet-packet-transform differential protection for power transformers’, IEEE multi-microphone speech enhancement’, IEEE Trans. Audio Speech Lang.
Trans. Ind. Appl., 2011, 47, (2), pp. 1003–1012 Process., 2012, 20, (1), pp. 223–233
11 Hossam Eldin, A.A., Refaey, M.A.: ‘A novel algorithm for discrimination between 21 Hu, H., He, Z., Zhang, Y., et al.: ‘Modal frequency sensitivity analysis and
inrush current and internal faults in power transformer differential protection based application using complex nodal matrix’, IEEE Trans. Power Deliv., 2014, 29,
on discrete wavelet transform’, Electr. Power Syst. Res., 2011, 81, pp. 19–24 (2), pp. 969–971
12 Vahidi, B., Ghaffarzadeh, N., Hosseinian, S.H.: ‘A wavelet-based method to 22 Taniguchi, T., Sha, S., Karasawa, Y.: ‘Statistical distribution of eigenvalues of
discriminate internal faults from inrush currents using correlation coefficient’, correlation matrices in i.i.d MIMO channels under Rayleigh fading’. IEEE 16th
Electr. Power Energy Syst., 2010, 32, pp. 788–793 Int. Symp. on Personal, Indoor and Mobile Radio Communications, 11–14
13 Rasoulpoor, M., Banejad, M.: ‘A correlation based method for discrimination September 2005, pp. 537–541
between inrush and short circuit currents in differential protection of power 23 Girgis, R.S., teNyenhuis, E.G.: ‘Characteristics of inrush current of present designs
transformer using discrete wavelet transform: theory, simulation and of power transformers’. IEEE General Meeting Power Engineering Society, 24–28
experimental validation’, Electr. Power Energy Syst., 2013, 51, pp. 168–177 June 2007, pp. 1–6
14 Benmouyal, G., Roberts, J.: ‘Superimposed quantities: their true nature and 24 Gong, M., Zhang, X., Gong, Z., et al.: ‘Study on a new method to identify inrush
application relays’. Proc. of the 26th Annual Western Protective Relay Conf., current of transformer based on wavelet neural network’. Int. Conf. on Electrical
Spokane, WA, 26–28 October 1999 and Control Engineering (ICECE), 16–18 September 2011, pp. 848–852
15 Claude, F., Navarro, G., Ordóñez, A.: ‘The wavelet matrix: an efficient wavelet tree 25 Faiz, J., Saffari, S.: ‘Inrush current modeling in a single-phase transformer’, IEEE
for large alphabets’, Inf. Syst., 2015, 47, pp. 15–32 Trans. Magn., 2010, 46, (2), pp. 578–581
16 Zhang, W., Tan, Q., Miao, S., et al.: ‘Self-adaptive transformer differential 26 Abniki, H., Majzoobi, A., Monsef, H., et al.: ‘Identifying inrush currents from
protection’, IET Gener. Transm. Distrib., 2013, 7, (1), pp. 61–68 internal faults using symmetrical components in power transformers’. 2010 Proc.
17 Hosny, A., Sood, V.K.: ‘Transformer differential protection with phase angle of the Int. Symp. Modern Electric Power Systems (MEPS), 20–22 September
difference based inrush restraint’, Electr. Power Syst. Res., 2014, 115, pp. 57–64 2010, pp. 1–6
18 Ozgonenel, O., Karagol, S.: ‘Transformer differential protection using wavelet 27 ‘IEEE guide for protective relay applications to power transformers’, in IEEE Std.
transform’, Electr. Power Syst. Res., 2014, 114, pp. 60–67 C37.91-2000 (Revision of IEEE Std. C37.91-1985, pp. 1–85

IET Gener. Transm. Distrib., 2016, Vol. 10, Iss. 12, pp. 2871–2879
& The Institution of Engineering and Technology 2016 2879