IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO.

2, APRIL 2018 789

A Wavelet-Based Transformer Differential Protection

With Differential Current Transformer Saturation
and Cross-Country Fault Detection
Rodrigo Prado Medeiros and Flavio Bezerra Costa , Member, IEEE

Abstract—The current transformer (CT) saturation phe- provided better performance against nuisance trips caused by
nomenon has been one of the main problems for the power CT saturation than the relays with dual bias slopes. However,
transformer differential protection, leading to incorrect current depending on the configuration used by this protective function
measurements and relay misoperation. This paper proposes a fast
and efficient transformer differential protection scheme with addi- (independent or cross-blocking mode) [2], the relay efficiency
tional differential CT saturation and cross-country fault detection may be reduced if internal faults take place during external faults
modules after the external fault detection, all of them based on the (cross-country fault).
differential wavelet coefficient energy with border distortions in Aiming to improve the differential protection performance
order to stabilize the relay during external faults and distinguish under CT saturation, digital signal processing techniques have
accurately CT saturation from cross-country internal faults. The
proposed method was assessed by using representative simulations been employed for recognizing saturation intervals in the CT
of internal faults, transformer energizations, and external faults secondary currents appropriately [1]–[7]. In [6], a multiregion
with CT saturation followed by cross-country internal faults, and adaptive differential relay based on the differential current tra-
good results were achieved. jectory was proposed in order to discriminate internal faults from
Index Terms—Current transformer (CT) saturation, differential other disturbances properly, including CT saturation cases. In
protection, wavelet transform, cross-country faults. [7], a filter for compensating the CT saturation effects during
faults and a filter for suppressing the DC component introduced
by the inrush current were proposed. The wavelet transform
I. INTRODUCTION
has also been used for power transformer differential protection
HE main purpose of the current transformer (CT) is to
T reproduce accurately the primary current in its secondary
winding, considering the criteria of magnitude and phase shift.
during CT saturation after external faults [4]. However, the con-
ventional wavelet coefficients and their variants are affected by
noise and are not sensible to overdamped transients induced by
However, the secondary current can present distortions when faults, limiting their applications in the power system protection
the CT magnetic core saturates, which may impact the protec- [8], [9]. In addition, must methods concerned only for solving
tive relay operation. For instance, phasor estimation can face problems regarding the CT saturation after external faults, pre-
problems when CTs saturate, which can impact the distance senting no reliability for detecting possible cross-country faults,
and overcurrent protective relays [1]. which may offer some concerns for the transformer differen-
According to [2], the CT saturation phenomenon has been one tial protection [10]–[12]. For instance, a high current magnitude
of the main problems for the power transformer differential pro- during external faults may shake and heat the transformer wind-
tection, since it may lead to a false trip during external faults. In ing, and as longer the fault lasts, as greater the risk of it evolving
fact, traditional percentage differential relays with a single slope into an internal transformer fault [12].
do not present a good response when CT saturates, and dual bias Recently, a wavelet-based power transformer differential
slopes are commonly used to restrain relays from tripping under protection was proposed in order to discriminate internal
CT saturation. In addition, besides the well-known efficiency faults from external faults and inrush conditions [13]. In this
against inrush conditions, the harmonic blocking element has method, the conventional differential protection philosophy
was recreated by using high frequency components extracted
Manuscript received January 12, 2017; revised June 24, 2017 and September through the differential wavelet coefficient energy with border
9, 2017; accepted October 10, 2017. Date of publication October 25, 2017; date
of current version March 22, 2018. This work was supported in part by CAPES distortions of the currents by means of the real-time stationary
(Coordenação de Aperfeiçoamento de Pessoal de Nı́vel Superior) and in part by discrete transform (RT-SWT) instead of the differential
CNPq (Conselho Nacional de Desenvolvimento Cientı́fico e Tecnológico). Paper currents. Both phase (87TW) and negative-sequence (87QW)
no. TPWRD-00012-2017. (Corresponding author: Rodrigo Prado Medeiros.)
The authors are with the Federal University of Rio Grande do Norte, wavelet differential units were proposed. As contributions, the
Natal - RN, CEP:59.078-970, Brazil (e-mail: rodrigo.prado@ufersa.edu.br; relay operating time was faster than conventional differential
flaviocosta@ect.ufrn.br). relays and the method presented accurate detection of critical
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. internal faults (turn-to-turn and turn-to-earth faults inside the
Digital Object Identifier 10.1109/TPWRD.2017.2764062 transformer windings). However, this method is susceptible to

0885-8977 © 2017 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.
790 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 2, APRIL 2018

CT saturation during external faults and it is not able to detect

possible cross-country faults.
This paper proposes an improvement in the power transformer
differential protection scheme based on the operating and re-
straining boundary wavelet coefficient energy proposed by [13]
in order to overcome the CT saturation problem during external
faults. Firstly, this paper proposes an external fault detection
logic in order to block the trip logic when an external fault is
detected and avoid the relay misoperation due to an eventual
CT saturation. However, the differential function must trip if a
cross-country fault occurs and remains blocked during CT satu-
ration. Therefore, a differential CT saturation detection logic is
proposed in order to detect and confirm CT saturations during
external faults, remaining the trip logic blocked. However, if a
cross-country fault takes place after the external fault, a cross-
country fault detection logic based on the wavelet-based differ-
ential protection logic 87TW unblocks the trip logic. Therefore,
the proposed method contemplates the 87TW and 87QW func-
tions from [13] (fast and accurate internal fault detection) and
presents as advantages: high effectiveness and security to be
not affected by CT saturations during external faults and good
performance to distinguish cross-country faults from CT satu-
rations after external faults. In addition, all equations and logics
were designed to be executed in the real-time with low compu-
tational burden.
The proposed method was evaluated under a wide variety of
events such as critical internal faults, transformer energizations,
and external faults with CT saturations and followed by cross-
country internal faults. Comparisons with a conventional dif- Fig. 1. The proposed wavelet-based transformer differential protection.
ferential protection were accomplished, and good results were
obtained.
B. The Real-Time Stationary Discrete Wavelet Transform With
Border Distortions
II. THE PROPOSED WAVELET-BASED TRANSFORMER The RT-SWT boundary wavelet coefficients w = {wi H φ and
DIFFERENTIAL PROTECTION wi X φ } of the currents i = {iH φ and iX φ } are computed, at
Fig. 1 depicts the block diagram of the proposed wavelet- sampling k, through inner products of the wavelet filter hψ with
based transformer differential protection with the saturation L samples of the current, as follows [8], [9]:
detection algorithm, which is executed every sampling time.
L −1
Whenever an event is detected through differential wavelet co- 1  ◦
w(l, k) = √ hψ (n)i(k − L + n + 1 + l), (1)
efficient energy variables, external fault/event, internal fault, or 2 n =0
transformer energization can be detected with the related logics.
When an external fault is detected, the algorithm can detect CT where k  Δk − 1 is always the current sampling; 0  l < L;
saturations and cross-country faults properly. Each block in this L is the length of hψ ; Δk  L is the length of sliding window;
diagram is addressed in the remainder of this section. ◦ ◦
i(k + m) = i(k − Δk + m) with m ∈ N ∗ (periodized signal
in Δk samples). In this paper, Δk = fs /f (i.e., one cycle of the
fundamental frequency f ); fs is the sampling rate.
A. Pre-Processing
According to [9], w(0, k) defined in (1) with l = 0 is the
The preprocessing block lies in antialiasing filters followed wavelet coefficients of the conventional SWT at the sampling
by analog-to-digital converters attending the Nyquist criterium k, whereas w(l = 0, k) are additional wavelet coefficients with
in order to get the time-discrete secondary currents (iH φ and border distortions, useful for detecting overdamped transients
iX φ ) of the current transformers CT1 and CT2, respectively. of faults.
The underscript φ represents the phases A, B, and C, as well as
Q, where Q is the negative-sequence index; H and X are high
C. Phase and Magnitude Adjustments
and low voltage winding sides of the transformer, respectively.
The negative-sequence currents are computed through stored Since the SWT is a linear transformation, the algorithm per-
discrete current samples (no phasors) [13]. forms the amplitude and phase shift correction directly in the
MEDEIROS AND COSTA: WAVELET-BASED TRANSFORMER DIFFERENTIAL PROTECTION WITH DIFFERENTIAL CT SATURATION 791

wavelet coefficients (w ) instead of the currents, as follows: since k > Δk − 1. Edif wb
f is computed with no boundary wavelet
⎡ ⎤ ⎡ ⎤ coefficients of the sliding window with length Δk − L [8]:
wi H A wi H A
⎢  ⎥ 1 ⎢ ⎥ 
k
2
⎣ wi H B ⎦= MH ⎣ wi H B ⎦, (2) Edif
wb
f (k) = wdif f (0, n), (9)
T APH
wi H C wi H C n =k −Δ k +L
⎡  ⎤ ⎡ ⎤ or recursively given by:
wi X A wi X A
⎢  ⎥ 1 ⎢ ⎥ Edif 2 2
f (k) = Edif f (k − 1) − wdif f (k − Δk + L) + wdif f (k),
wb wb
⎣ wi X B ⎦= MX ⎣ wi X B ⎦, (3)
T APX (10)
wi X C wi X C
since k > Δk + L − 1.
where T APH and T APX are taps of the CT1 and CT2, re- Unlike [13], where the differential operations were applied
spectively. The transformation matrices MH and MX are arrays to the operating and restraining currents, such differential op-
used for the zero-sequence and phase compensation presented erations were applied to the operating and restraining wavelet
in [14]–[16]. coefficients of the CT secondary currents in this paper. How-
ever, the operating and restraining energy are similar in both
procedures due to the linear properties of the SWT.
D. Differential Wavelet Coefficients
Unlike the operating and restraining currents (Iop and Ir es ) F. Event Detection
computed from CT secondary phasor currents in typical trans-
Based on [13], an event (internal and external faults, trans-
former differential protection algorithms, the proposed differ-
former energization, etc.) is detected if:
ential operations are applied directly in the wavelet coefficients.
The proposed differential wavelet coefficients wdif f = {wi o p φ , Edif
w
f (k − 1)  Edif f ,
wi r e s φ } of 87TW (φ = A, B, and C) and 87QW (φ = Q) ele- (11)
Edif
w
f (k) > Edif f ,
ments are computed as follows:
where kd /fs = k/fs , i.e., the disturbance inception time kd /fs
1 is set to be the current sampling time when these inequalities
wi o p φ (0, k) = (wi H φ (0, k) + wi X φ (0, k)), (4)
2 are true; Edif f = {Eopφ and Er esφ }, with φ = {A, B, C, and
wi o p φ (l = 0, k) = wi H φ (l, k) + wi X φ (l, k), (5) Q}, are thresholds stochastically defined as follows [13]:

3 
k2
wi r e s φ (l, k) = wi H φ (l, k) − wi X φ (l, k), (6) Edif f = Edif f (n),
w
(12)
k2 − k1 + 1
n =k 1
where 0  l < L. As suggested in [13], the SWT wavelet co-
where [k1 /fs k2 /fs ] is a previous steady-state time range.
efficients were obtained with the db(4) considering the first
The event detection logic is only used for detecting the first
decomposition level, which is appropriated for the differential
event after the steady-state. Therefore, whenever an event is
protection with currents sampled at fs = 15.36 kHz, in which
detected, the wavelet-based differential protection (87TW and
db(L) is the Daubechies mother wavelet with L coefficients [17].
87QW units), the energization, and the external fault detection
Therefore, there is one pair of differential coefficients (wi o p
logics are enabled in order to identify the event or sequence of
and wi r e s ) computed with one conventional wavelet coefficient
events (external fault with CT saturation, external fault followed
(l = 0) and three pairs of differential coefficients computed with
by its clearance, etc.). In this fashion, the event detector will only
border distortions (l = 0) for each φ.
be evaluated again in the next steady-state mode operation.

E. Differential Energy G. Transformer Energization Detection

Based on [8], the boundary differential wavelet coefficient The protection method in [13] predicts if the event will be
energy signals Edif
w
f = {Ei o p φ and Ei r e s φ }, termed just as dif-
w w related to a transformer energization when the current energy
ferential energy, are computed from the respective differential is null. If a pre-energization condition is detected, the slopes
wavelet coefficients wdif f = {wi o p φ , wi r e s φ }, as follows: K87T W and K87Q W are temporarily incremented to a value
slightly larger than 1 in order to avoid a possible misoperation
Edif of the differential protection units after the beginning of the
f (k) = Edif f (k) + Edif f (k),
w wa wb
(7)
energization.
since k > Δk − 1. Edifwa
f is due to the first L − 1 boundary
wavelet coefficients of the sliding window, defined as [8]: H. The Wavelet-Based Differential Protection
Based on the classical differential principle, an internal fault
−1
L is detected if [13]:
2
Edif
wa
f (k) = wdif f (l, k), (8)
l=1
Eiwo p φ (k) > K87W Eiwr e s φ (k), (13)
792 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 2, APRIL 2018

Fig. 3. Typical operating points during a turn-to-earth internal fault with a

few percentage of turns computed from: (a) the wavelet coefficient energy;
(b) the phasor-based conventional method.

were detected through (11) at kd /fs . For the sake of illustra-

tion simplicity, the operating energy waveforms in this paper
were multiplied by two in order to highlight the regions where
Eiwr e s φ < 2Eiwo p φ (internal fault detection), and the energy wave-
forms were normalized for Eiwr e s φ = 1 during the steady-state.
During external faults, just the restraining energy presents a
hard increase from the fault inception time and the operating
energy is scarcely sensitized [Fig. 2(a)], attending (15) and not
attending (14), whereas both the operating and restraining en-
ergy waveforms increase substantially from the internal fault
initiation [Fig. 2(b)], and (15) is no longer attended.
Regarding critical internal faults, such as that in a few per-
centage of turns of the transformer winding, the increase of both
operation and restraining energies are not strong, but the oper-
ating energy tends to be the highest in internal faults. Indeed,
the energy increases mainly due to the boundary effects because
the transients are overdamped. As a consequence, (15) is not
Fig. 2. Typical operating and restraining energy during: (a) an external fault;
(b) an internal fault; (c) a critical internal fault. attended and (14) is attended [Fig. 2(c)], and an external fault is
not detected.
Fig. 3 depicts the operating energy points (Eiwo p A , Eiwr e s A ) dur-
where K87W = K87T W for φ = A, B, and C and K87W = ing the first cycle of the internal fault shown in Fig. 2(c) as well
K87Q W for φ = Q are slopes of characteristic curves of 87TW as the related operating points of the conventional differential
and 87QW elements. In this paper, K87T W = K87Q W = 0.5. protection (IopA , Ir esA ), normalized to be compared. The pro-
Therefore, an internal fault is detected at the time k/fs if: posed external fault detection logic detected this event as non
external fault. After few samples, the 87TW detected an internal
Eiwr e s φ (k) < 2Eiwo p φ (k). (14)
fault and a trip was issued. Conversely, a typical conventional
method would not be able to detect this specific fault (oper-
I. External Fault Detection
ating points concentrated in the restraining region). Both the
According to the power transformer differential protection proposed external fault detection logic and the 87TW worked
theory, the ratio Ir es /Iop tends to be greater in external faults to properly due to the border effects of the boundary differential
the transformer protection zone than in internal faults, avoiding wavelet coefficient energies.
a possible relay misoperation [18]. Based on this premise, this The external fault detection logic is essential in order to avoid
paper proposes a simple method to identify external disturbances problems related to the CT saturation during external faults.
of the transformer protection zone if just the restraining energy Therefore, when an external fault is detected through (15), the
exceeds the thresholds, i.e., if the two following conditions, at proposed differential CT saturation detection logic in enable,
the sampling k  kd , are true: such as addressed in the next section.

Eiwr e s φ (k) > Er esφ ,
(15) III. DIFFERENTIAL CURRENT TRANSFORMER SATURATION
Eiwo p φ (k) < Eopφ . DETECTION METHOD
Fig. 2 depicts the phase-A operating and restraining energy Figs. 4–6 depict the CT secondary currents, their related
waveforms during external and internal faults, where the events wavelet coefficients, and the operating and restraining wavelet
MEDEIROS AND COSTA: WAVELET-BASED TRANSFORMER DIFFERENTIAL PROTECTION WITH DIFFERENTIAL CT SATURATION 793

Fig. 4. Internal fault with CT saturation in a single side: (a) current iH A :

(b) current iX A ; (c) wavelet coefficients of iH A ; (d) wavelet coefficients of
iX A ; (e) operating and restraining wavelet coefficient energy.

coefficient energy regarding an internal fault with CT saturation

in a single side, an external fault with CT saturation in a sin-
gle side, and an external fault with CT saturation in both sides,
respectively. For the sake of illustration simplicity, the wavelet
coefficients were normalized in accordance with the noise con-
tent in order to present peaks associated to the fault inception
time and CT saturation higher than one.

A. The Need of CT Saturation Detection

According to [19], the concern of the CT saturation phe-
nomenon on conventional transformer differential relays de- Fig. 5. External fault with CT saturation in a single side: (a) current iH A ;
(b) current iX A ; (c) wavelet coefficients of iH A (SNR=60 db); (d) wavelet
pends whether the fault is internal or external to the transformer coefficients of iX A (SNR=60 db); (e) operating and restraining wavelet coef-
protection zone. The CT saturation is not a problem during inter- ficient energy (SNR=60 db); (f) wavelet coefficients of iH A (SNR=36 db);
nal faults because relays are designed to trip in these situations. (g) wavelet coefficients of iX A (SNR=36 db); (h) operating and restraining
wavelet coefficient energy (SNR=36 db).
However, the more-prevalent concern is the possible misopera-
tion of differential relays for external faults with CT saturation,
in which differential currents can reach the setting values for 1) In the same fashion of the conventional method, the perfor-
the relay operation. In these cases, the relay tends to operate mance of the proposed 87TW and 87QW are not affected
improperly, and external faults are detected as internal faults by the CT saturation in internal faults because the internal
when the saturation starts [13]. fault is detected before the saturation. For instance, the
Regarding the proposed wavelet-based differential protection, internal fault was detected in about 65 μs after the fault
the following statements were observed (Figs. 4–6): inception time in Fig. 4(e).
794 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 2, APRIL 2018

Fig. 6. External fault with CT saturation in both sides: (a) current iH A ;

(b) current iX A ; (c) operating and restraining wavelet coefficient energy.

2) In the same fashion of the conventional method, the perfor-

mance of the proposed 87TW and 87QW can be affected
by the CT saturation in external faults according to the
saturation level. For instance, some wavelet coefficient
energy operating points in Figs. 5(e) and 6(e) increased
during the CT saturation and converged temporarily to
the operation region (Eiwr e s φ < 2Eiwo p φ ) during some CT Fig. 7. Sequence of events: (a) AB external fault with CT saturation followed
saturation intervals. Therefore, internal faults could be by the fault clearance; (b) AG external fault followed by an AG cross-country
wrongly detected if the trip logic was not blocked. internal fault; (c) AB external fault with CT saturation followed by an ABG
cross-country fault.

B. Accurate CT Saturation Detection

The wavelet coefficients have been proposed to detect faults the differential boundary wavelet coefficient energy waveforms
[20], [21] as well as CT saturation intervals accurately [3], [22]. presented relevant spikes during CT saturation intervals even
Indeed, after the fault inception time, the wavelet coefficients with a SNR of 36 db in Fig. 5(h). In agreement with the fault
of the saturated currents presented peaks associated to the CT detection through the wavelet coefficient energy with border
saturation [Figs. 4(c) and 5(d)]. However, the accuracy of the distortions proposed in [8], the differential wavelet coefficient
wavelet coefficients is affected by several parameters such as energy with border distortions is also less susceptible to noise.
the frequency contents of the transients or distortions.
The RT-SWT with border distortions was proposed by [8] in D. The CT Saturation Identification
order to overcome some drawbacks of the conventional SWT
The wavelet coefficient and differential energy peaks after an
and detect faults accurately. In the same fashion, the proposed
external fault are not always associated to a CT saturation. For
boundary differential wavelet coefficient energy is very sensitive
instance, when an external fault is not cleared by the protection
to faults and CT saturations due to both the transients/distortions
system successfully, an internal fault can take place few cycles
and border effects. For instance, the differential energy in
after the external fault (cross-country fault) [10]. The external
Fig. 5(e) presented an expressive increase associated to CT sat-
fault clearance and other switching events can also generate
urations in the current iX A mainly due to the border effects.
wavelet coefficient and energy peaks. As a consequence, the
wavelet coefficient and differential energy peaks may appear
C. The Noise Effects
after the external fault inception time due to transient events.
It is well-known that the wavelet coefficients are affected by Fig. 7 presents three cases of multiple transient events, which
noise [8]. For instance, the CT saturation initiation times could are challenges for the differential protection analysis: an external
be detected through the wavelet coefficients in Fig. 5(d) with fault followed by CT saturation and by its clearance; an external
a SNR of 60 db. However, the same saturations could not be fault followed by a cross-country internal fault; and an external
detected with the currents with a SNR of 36 db. Conversely, fault followed by CT saturation and by a cross-country internal
MEDEIROS AND COSTA: WAVELET-BASED TRANSFORMER DIFFERENTIAL PROTECTION WITH DIFFERENTIAL CT SATURATION 795

Fig. 9. CT Saturation and cross-country fault detection logics at phase A.

Fig. 8. Energy operating points of the: (a) external fault initiation followed
by CT saturation or its clearance (3 cycles: 768 points) in Figs. 7(a) and
(c); (b) cross-country internal fault (2 cycles: 512 points) in Fig. 7(b).

Fig. 10. Single line diagram of the electrical system.

fault. The main concern is that the differential protection must
trip only when the cross-country faults take place in these cases. 1) the trip logic is blocked to avoid misoperation associated
Fig. 8 depicts the trajectory of the wavelet coefficient energy to the CT saturation;
operating points of the 87TW function (Eiwo p A , Eiwr e s A ) for some 2) the differential function 87TW is not blocked, but it can
events shown in Fig. 7. not control the blocked trip logic;
One concern of the wavelet coefficients is how to identify 3) the differential function 87QW is blocked, because it is a
the peaks associated to CT saturation and other events after very sensitive unit.
the external fault inception time, whereas such distinction can 4) the CT saturation detection logic is enabled to maintain
be properly accomplished through the differential energy by the trip logic blocked when CT saturation is detected and
using the differential protection principles. For instance, the unblock it if faults are detected, respectively. This logic
proposed 87TW function based on the differential wavelet co- considers the differential function 87TW based on the
efficient energy could distinguish multiple CT saturations from differential boundary wavelet coefficient energy.
a cross-country fault and other transient events (e.g., external Fig. 9 depicts the CT saturation detection logic. The distinc-
fault clearance) by using simple criteria: tion between cross-country internal faults and CT saturations
1) According to Fig. 7(a), Eiwr e s A > 2Eiwo p A during an exter- is accomplished by means of the differential function 87TW.
nal fault clearance, and an incorrect trip would not be An increment/decrement counter shown in Fig. 9 was used in
provided. order to register the energy operating points of the differential
2) According to Figs. 7(a) and (c), Eiwr e s A < 2Eiwo p A for a function 87TW, as follows: increment the counter if the energy
short time during the CT saturation intervals. This means operating point is located on the operating region and decrement
that most operating points tend to remain in the restriction it if the operating point is located on the restraining region. If
area during CT saturations [Fig. 8(a)]. the counter exceeds the value equal to 1/2 Δk, the trip logic is
3) According to Figs. 7(b) and (c), Eiwr e s A < 2Eiwo p A for a enabled, and the 87TW function can issue a trip.
long period during the cross-country fault. This means
that most operating points tend to remain in the operating
IV. PERFORMANCE ASSESSMENT
area during a cross-country fault [Fig. 8(b)].
Fig. 10 depicts a single line diagram of the electrical power
system used for the assessment of the proposed method. The
E. CT Saturation Detection Logic
power system was modeled using the Alternative Transients
When an external fault is detected by (15): Program (ATP), and consists of two Thevenin equivalents
796 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 2, APRIL 2018

represented by the voltage sources S1 and S2 and their related tapes (400-5 A and 600-5 A) in order to cause the
impedances connected to the primary and secondary windings onset of saturation in the secondary currents (140
of the power transformer, respectively. Details about the current records).
transformer and power transformer modelings are described in 4) Database 4 (single phase-to-ground external fault + single
the appendix of this work. phase-to-ground cross-country internal fault):
Based on [13], the proposed wavelet-based power transformer i) AG external faults on the high voltage bus followed
differential protection with the 87TW, 87QW, and CT saturation by AG, BG and CG internal faults on transformer
and cross-country internal fault detection (Fig. 1) was imple- bushing windings at the same side of the faulted
mented with a sampling rate of 15360 Hz, which is appropriated bus, with θf = {0, 30, 60, 90, 120, 150, 180} elec-
to extract the transients induced by faults and compatible to the trical degrees, and rf = {1, 5, 10} Ω (63 records).
current technology for protective relays. The internal faults take place four cycles after the
The wavelet analysis is significantly affected by the choice of external faults.
the mother wavelet, which can change depending on the appli- ii) AG external faults on the high voltage bus followed
cation. According to [9], [23], the mother wavelet db(4) presents by turn-to-earth internal faults on the phase A wye
better response for detection of transients induced by faults than winding. The percentage of the turns in the inter-
other wavelets from the Daubechies family with higher num- nal faults is equal to e = {5, 10, 15, ..., 95} % (19
ber of coefficients. In addition, the db(4) was properly used records).
in the power transformer differential protection, and presented 5) Database 5 (external fault + cross-country internal fault):
low computational burden [13]. Therefore, the selected mother AG and AB external faults on the high and low voltage
wavelet in this paper was the db(4). buses evolving to ABG and ABC internal faults on trans-
A differential protection scheme contemplating the traditional former windings at the same side of the faulted bus, with
phase differential element (87T) with harmonic restraint, the θf = {0, 30, 60, 90, 120, 150, 180} electrical degrees, and
negative-sequence current differential (87Q), and the harmonic rf = {1, 10} Ω (112 records). The internal faults take
blocking elements with independent and cross-blocking oper- place four cycles after the external faults.
ating modes was reproduced according to [24] to be compared 6) Database 6 (external fault + CT saturation + cross-country
to the proposed differential protection. The used sampling fre- internal fault): The faults are the same of database 5 (112
quency was 960 Hz and the complex phasors were estimated records).
through the full-cycle Fourier algorithm.
The following databases with a wide variety of events were A. Internal Faults (Database 1)
evaluated in this paper:
The traditional 87T differential element usually detects the
1) Database 1 (internal faults): turn-to-turn faults on the
most internal faults in power transformer, except some faults,
phase A wye winding; turn-to-turn faults on the delta
such as turn-to-turn faults and phase-to-ground faults close to
winding between phases A and B; turn-to-earth faults on
the transformer neutral. Therefore, phasor-based transformer
the phase A wye winding; and turn-to-earth faults on the
differential relays usually use the negative-sequence current dif-
delta winding between the A-to-B-winding and the earth.
ferential (87Q) and the restricted earth fault (REF) elements in
The percentage of the turns in the fault is equal to e =
association with the phase current differential element in order
{1, 2, 3, ..., 98} % (392 records).
to provide a better detection of internal faults [25]. The selected
2) Database 2 (transformer energizations): Switching per-
conventional differential scheme with harmonic restraint, har-
formed by the high voltage side (230 kV), with the
monic blocking and the 87Q units ensured 92.60% of success
secondary terminal opened, and changing the high
rate (only 78.83% with the 87T unit). In addition, these func-
voltage circuit breaker closing time at angles θs =
tions provided an average relay operating time of about 19 ms,
{0, 1, 2, ..., 358, 359} electrical degrees (360 records).
with both independent and cross-blocking modes.
The phase A voltage is taken as reference to θs .
The proposed 87TW differential protection element detected
3) Database 3 (external faults with CT saturation):
98.47% of all internal faults, failing just in one case of turn-to-
i) CT saturation due to the burden resistance: AG,
turn fault on the delta side (1% of the short-circuited winding)
AB, ABG, and ABC external faults on the
and five cases of turn-to-earth faults on the wye side (1–5% of
high voltage bus with fault inception angle θf
the short-circuited winding). These cases were covered by the
= {0, 30, 60, 90, 120, 150, 180} electrical degrees,
87QW unit. Therefore, a success rate of 100% was obtained
fault resistance rf = 1 Ω, and load Rb =
with the 87TW + 87QW.
{1, 2, 3, ..., 19, 20} Ω connected to the secondary
The average relay operating time was 214 μs, almost 100
winding of CT1 (560 records). The phase A voltage
times faster than the conventional one, due to the extraction of
is taken as reference to θf .
the high-frequency content of the fault, which is in accordance
ii) CT saturation due to tap configuration errors: AG,
with a high-speed power transformer differential protection.
BG, CG, AB, BC, AC, ABG, BCG, ACG, and ABC
external faults on the high and low voltage bus
with θf = {0, 30, 60, 90, 120, 150, 180} electrical B. Transformer Energizations (Database 2)
degrees and rf = 1 Ω. In this analysis, the CT sat- The proposed energization detection module provided a suc-
uration curves were purposely designed for smaller cess rate of 100%, i.e., no trip was generated. The conventional
MEDEIROS AND COSTA: WAVELET-BASED TRANSFORMER DIFFERENTIAL PROTECTION WITH DIFFERENTIAL CT SATURATION 797

E. External Faults + CT Saturations + Cross-Country Faults

(Databases 5 and 6)
The proposed method detected successfully all the external
faults. Therefore, no trip was generated during this period. After
that, all the saturations were properly detected and no trip during
these periods was generated. After a while, the external faults
evolved to internal faults, and 222 of 224 cross-country faults
were detected (i.e., a success rate of 99.11% was achieved) with
an average relay operating time of 10.12 ms due to the time
delay in the operation of the increment/decrement counter for
Fig. 11. Performance of the proposed differential protection under external security distinction between CT saturation and cross-country
faults with CT saturation. faults. The conventional differential protection equipped with
the independent harmonic blocking unit ensured a success rate
of 100%, which reveals the importance of the usage of this
differential method with harmonic blocking function (with inde- configuration regarding the reliability in detecting cross-country
pendent and cross-blocking operating modes) management unit faults, whereas the another conventional version equipped with
blocked the 87Q unit, which was sensitized for all inrush cases, the cross-blocking harmonic unit detected only 87.05% of the
and the 87T with harmonic restraint unit was not sensitized, pro- cross-country faults.
viding security (no trip) for all transformer energization cases.

V. CONCLUSION
C. External Faults + CT Saturation (Database 3)
This paper presented a power transformer differential pro-
The proposed external fault detection method detected suc- tection scheme based on the boundary discrete wavelet
cessfully all the external faults, i.e., no trip was generated, and transform with external fault, CT saturation, and cross-country
the CT saturation detection method was properly enabled for internal fault detection modules. The performance of the pro-
monitoring the CT saturation. posed method was assessed with representative simulations of
Based on [2], the degree of saturation is given by: internal faults, transformer energizations, and external faults
Isat with CT saturations followed by cross-country internal faults.
η =1− , (16) The proposed wavelet-based differential protection with only
Iideal
87TW and 87QW units presented a success rate of 100% for de-
where (no saturation) 0  η  1 (the heaviest possible satura- tecting internal faults with an expressive average relay operating
tion); Isat is the rms saturated secondary current; and Iideal time of 214 μs, whereas a phasor-based conventional protection
is the rms primary current without saturation, reflected to the scheme ensured only 92.60% of success rate with an average
secondary, including dc offset. relay operating time of 19 ms. Therefore, the proposed method
Fig. 11 depicts the saturation degree, the instant (in cycles) was the simplest, fastest, and most accurate.
of the first saturation initiation, and the performance of the pro- Regarding the multiple events to the power transformer,
posed method. Therefore, the proposed CT saturation detection both the proposed and conventional techniques were immune
method was tested in a wide variety of CT saturation degrees to the CT saturation. However, the proposed CT saturation
(light, medium and heavy saturation). The CT saturations were detection module is simpler than the conventional technique,
detected and no trip was generated (100% of success rate). requiring only the inclusion of a wavelet coefficient energy in-
The wavelet-based method proposed by [13], which is not crement/decrement counter instead of the traditional harmonic-
equiped with a CT saturation detection module, detected all the based functions. In addition, the proposed method ensured a
external faults with CT saturation as internal faults, and wrong success rate of 100% in detecting cross-country internal faults
trips were performed. The conventional protection equipped from the database 4, whereas the conventional method was able
with harmonic restraint, harmonic blocking (with independent to detect only about of 89% by using both independent and
and cross-blocking configurations) and 87Q units provided se- cross-blocking modes. Regarding the databases 5 and 6, the
curity during the saturation, and no wrong trip was verified. proposed method ensured a success rate of 99.11% in detect-
ing the cross-country internal faults against 100% and 87.05%
D. Single Phase-to-Ground External Fault + Single of the conventional method with the independent and cross-
Phase-to-Ground Cross-Country Internal Fault (Database 4) blocking modes, respectively. Therefore, the proposed method
was immune to CT saturation and presented good reliability in
The proposed method detected all the single phase-to-ground detecting cross-country internal faults.
faults (cross-country faults), ensuring a success rate of 100%
with an average relay operating time of 10 ms. The conventional
APPENDIX
technique failed in some critical turn-to-earth internal faults,
presenting a success rate of 89% with an average relay operating This section presents details about the power transformer,
time of about one-cycle (16 ms). current transformer, and fault modellings.
798 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 2, APRIL 2018

TABLE I TABLE II
NONLINEAR CHARACTERISTIC OF THE MAGNETIZING BRANCH NONLINEAR CHARACTERISTICS OF THE MAGNETIZING BRANCH OF THE USED
OF THE USED POWER TRANSFORMER CT1 AND CT2 IN DIFFERENT TAPES

i (A) ϕ (Wb) CT1 (800-5 A) CT2 (1200-5 A) CT1 (400-5 A) CT2 (600-5 A)

0.144 498.137 i (A) ϕ (Wb) i (A) ϕ (Wb) i (A) ϕ (Wb) i (A) ϕ (Wb)
0.478 523.044 0.052 0.112 0.054 0.338 0.085 0.034 0.117 0.188
1.211 547.951 0.075 0.225 0.132 1.606 0.128 0.079 0.247 0.750
2.540 572. 858 0.135 0.450 0.175 1.876 0.262 0.263 0.399 1.125
6.446 579.085 0.165 1.125 0.189 2.251 0.542 0.675 0.088 1.38
8.954 585.312 0.301 1.501 0.341 2.626 0.724 0.825 1.358 1.501
15.595 591.538 0.555 1.688 0.561 2.926 1.219 0.938 2.080 1.512
20.396 597.765 0.687 1.876 0.976 3.001 1.967 1.013 2.475 1.519
35.461 603.992 44.856 2.251 9.440 3.477 76.328 1.125 58.055 1.613

The SATURATION support routine of ATP was used to con-

vert the original v − i characteristic curves of the CTs provided
by [19] into an equivalent ϕ − i data set. Then, the magnetizing
branch was modeled in ATP using the card type 98. Table II de-
scribes the current versus flux (i, ϕ) magnetizing characteristics
of the CTs.

Fig. 12. The CT circuit model used in this work. C. The Fault Circuit Model
Fig. 13 depicts the fault modelling composed by:
1) The fault circuit for application of single phase-to-
ground, phase-to-phase, phase-to-phase-to-ground and
three-phase external and internal faults.
2) The switches that model the high and low-voltage circuit
breakers used for simulating transformer energization ma-
neuvers.
3) The internal short-circuit application switches (S1C C ,
S2C C , S3C C , and S4C C ) implemented for simulating, re-
spectively, turn-to-turn faults on the wye winding, turn-to-
Fig. 13. The fault circuit model used in this work. turn faults on the delta winding, turn-to-earth faults on the
wye winding, and turn-to-earth faults on the delta winding
A. The Power Transformer Model (internal faults).

The power transformer model is provided by [13], pre- REFERENCES

senting rated power of 100 MVA, voltage ratio of Vp :Vs =
[1] J. Pan, K. Vu, and Y. Hu, “An efficient compensation algorithm for current
230:69 kV, and YNd1 configuration. The impedances related transformer saturation effects,” IEEE Trans. Power Del., vol. 19, no. 4,
to the primary and secondary windings of the power trans- pp. 1623–1628, Oct. 2004.
former were Rp + jXp = 2.04 + j12.54 Ω and Rs + jXs = [2] M. Stanbury and Z. Djekic, “The impact of current-transformer saturation
on transformer differential protection,” IEEE Trans. Power Del., vol. 30,
1.44 + j38.04 Ω, respectively. Regarding the nonlinear charac- no. 3, pp. 1278–1287, Jun. 2015.
teristic, the power transformer core was modelled by means of [3] Y.-Y. Hong and P.-C. Chang-Chian, “Detection and correction of distorted
the Hevia Hysteresis ATP routine with their partitioned wind- current transformer current using wavelet transform and artificial intelli-
gence,” IET Gen. Transm. Distrib., vol. 2, no. 4, pp. 566–575, Jul. 2008.
ings in order to simulate turn-to-turn and turn-to-earth internal [4] A. Rahmati, “An adaptive differential relay for CT saturation based on
faults (Fig. 13). Table I describes the current versus flux (i, ϕ) wavelet transform,” in Proc. Int. Conf. Adv. Energy Eng., Jun. 2010, pp. 89–
points for the power transformer core magnetizing curve. 94.
[5] F. B. Ajaei, M. Sanaye-Pasand, M. Davarpanah, A. Rezaei-Zare, and
R. Iravani, “Compensation of the current-transformer saturation effects
B. The CT Model for digital relays,” IEEE Trans. Power Del., vol. 26, no. 4, pp. 2531–2540,
Oct. 2011.
Regarding the CT model, the elements related to the leak- [6] H. Dashti and M. Sanaye-Pasand, “Power transformer protection using a
age losses on the primary and secondary CT windings (leakage multiregion adaptive differential relay,” IEEE Trans. Power Del., vol. 29,
no. 2, pp. 777–785, Apr. 2014.
inductances and resistances) were reported in [26]. Data con- [7] E. Hajipour, M. Vakilian, and M. Sanaye-Pasand, “Current-transformer
cerning non-linear modeling of the CT (saturation curve) were saturation compensation for transformer differential relays,” IEEE Trans.
taken from [19], since it was necessary various saturation curves Power Del., vol. 30, no. 5, pp. 2293–2302, Oct. 2015.
[8] F. B. Costa, “Fault-induced transient detection based on real-time analysis
in order to verify several levels of the CT saturation. Fig. 12 de- of the wavelet coefficient energy,” IEEE Trans. Power Del., vol. 29, no. 1,
picts the equivalent circuit model for the CTs used in this paper. pp. 140–153, Feb. 2014.
MEDEIROS AND COSTA: WAVELET-BASED TRANSFORMER DIFFERENTIAL PROTECTION WITH DIFFERENTIAL CT SATURATION 799

[9] F. B. Costa, “Boundary wavelet coefficients for real-time detection of [23] S. Santoso, E. Powers, W. Grady, and P. Hofmann, “Power quality assess-
transients induced by faults and power-quality disturbances,” IEEE Trans. ment via wavelet transform analysis,” IEEE Trans. Power Del., vol. 11,
Power Del., vol. 29, no. 6, pp. 2674–2687, Dec. 2014. no. 2, pp. 924–930, Apr. 1996.
[10] X. Lin, H. Weng, and B. Wang, “Identification of cross-country fault of [24] K. A. Tavares and K. M. Silva, “Evaluation of power transformer differen-
power transformer for fast unblocking of differential protection,” IEEE tial protection using the ATP software,” IEEE Latin Amer. Trans., vol. 12,
Trans. Power Del., vol. 24, no. 3, pp. 1079–1086, Jul. 2009. no. 2, pp. 161–168, Mar. 2014.
[11] S. K. Murugan, S. P. Simon, K. Sundareswaran, P. S. R. Nayak, [25] A. Guzman, N. Fischer, and C. Labuschagne, “Improvements in trans-
and N. P. Padhy, “An empirical fourier transform-based power trans- former protection and control,” in Proc. 62nd Annu. Conf. Protec. Relay
former differential protection,” IEEE Trans. Power Del., vol. 32, no. 1, Eng., Mar. 2009, pp. 563–579.
pp. 209–218, Feb. 2017. [26] IEEE Power System Relaying Committee, “EMTP reference models for
[12] IEEE Guide for Protective Relay Applications to Power Transformers, transmission line relay testing,” Rep. D10, IEEE Power & Energy Society,
IEEE Standard 273-1967 (ANSI C36.91-1972), Feb. 1967. Piscataway, NJ, USA, 2004.
[13] R. P. Medeiros, F. B. Costa, and K. M. Silva, “Power transformer differ-
ential protection using the boundary discrete wavelet transform,” IEEE
Trans. Power Del., vol. 31, no. 5, pp. 2083–2095, Oct. 2016. Rodrigo Prado Medeiros received the B.Sc. and
[14] G. Ziegler, Numerical Differential Protection: Principles and Applica- M.Sc. degrees in electrical engineering from Federal
tions. Hoboken, NJ, USA: Wiley, 2005. University of Rio Grande do Norte (UFRN), Natal,
[15] H. Ferrer, E. Schweitzer, and S. E. Laboratories, Modern Solutions for Brazil, in 2012 and 2014, respectively. He is currently
Protection, Control, and Monitoring of Electric Power Systems. Pullman, working toward the Ph.D. degree at UFRN. His re-
WA, USA: Schweitzer Engineering Laboratories, Inc., 2010. search interest are electric power quality and power
[16] IEEE Guide for Protecting Power Transformers, IEEE Standard C37.91- system protection.
2008 (Revision of IEEE Standard C37.91-2000), May 2008.
[17] S. G. Mallat, “A theory for multiresolution signal decomposition: The
wavelet representation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 11,
no. 7, pp. 674–693, Jul. 1989.
[18] Z. Bo, G. Weller, and T. Lomas, “A new technique for transformer protec-
tion based on transient detection,” IEEE Trans. Power Del., vol. 15, no. 3,
pp. 870–875, Jul. 2000.
[19] IEEE Guide for the Application of Current Transformers Used for Flavio Bezerra Costa (S’05–M’10) received the
Protective Relaying Purposes—Redline, IEEE Standard C37.110-2007, B.Sc., M.Sc., and Ph.D. degrees in electrical engi-
Apr. 2008. neering from Federal University of Campina Grande
[20] N. I. Elkalashy, M. Lehtonen, H. A. Darwish, A. M. I. Taalab, and M. (UFCG), Campina Grande, Brazil, in 2005, 2006, and
A. Izzularab, “DWT-based detection and transient power direction-based 2010, respectively. He is currently a Professor at the
location of high-impedance faults due to leaning trees in unearthed MV School of Science and Technology (ECT), Federal
networks,” IEEE Trans. Power Del., vol. 23, no. 1, pp. 94–101, Jan. 2008. University of Rio Grande do Norte (UFRN), Natal,
[21] F. B. Costa, B. A. Souza, and N. S. D. Brito, “Real-time detection of fault- Brazil. He was a Postdoctoral Researcher at UFCG in
induced transients in transmission lines,” Electron. Lett., vol. 46, no. 11, 2010, a Visiting Researcher at KU Leuven, Belgium,
pp. 753–755, May 2010. in 2011–2012, a Visiting Researcher at INESC-Porto,
[22] S. AsghariGovar and H. Seyedi, “Adaptive CWT-based transmission line Portugal, in 2014, and a Postdoctoral Researcher at
differential protection scheme considering cross-country faults and CT RWTH Aachen University in 2014–2015. His current research interests include
saturation,” IET Gen., Transm. Distrib., vol. 10, no. 9, pp. 2035–2041, power system protection, electric power quality, control of renewable energy
2016. systems, as well as smart-grid solutions.