Professional Documents
Culture Documents
DOI 10.1007/s00202-016-0364-7
ORIGINAL PAPER
Received: 9 August 2014 / Accepted: 11 April 2016 / Published online: 26 April 2016
© Springer-Verlag Berlin Heidelberg 2016
Abstract The “setting-less” protection approach is an inno- Keywords Nonlinear dynamic systems · Power system
vative protection scheme with increased dependability and protection · Power transformers · State estimation
security. This method is based on dynamic state estimation,
which continuously extracts the operating condition of a pro-
tection zone from real-time measurements and the protection 1 Introduction
zone dynamic model. The dynamic state estimation computes
the probability that the measurements fit the dynamic model In the area of power system protection, security indicates the
of the protection zone within the measurement accuracy via ability for protective relays not to trip for normal or tolerable
the Chi-square test. The probability represents the goodness conditions, while dependability is the ability of relays to trip
of fit of measurements to the dynamic model of the protection for any intolerable fault conditions of devices under protec-
zone. We refer to it as the confidence level that the protection tion. In fact, it is critical for a protection scheme to guarantee
zone is healthy. High confidence levels imply that the pro- both security and dependability, but in many cases, protective
tection zone operates normally, while low confidence levels relaying settings are selected as a trade-off among conflict-
indicate that internal abnormalities/faults may have occurred. ing factors. For example, in case of transformer protection,
Note that the proposed method does not require coordination overcurrent relay monitors transformer winding currents and
with other protection schemes which sometimes compromise trips breakers when the monitored currents exceed preset
security and/or dependability. In this paper, the setting-less thresholds, considering the currents as fault currents. How-
protection approach is described and applied to transformer ever, the overcurrent relay may interpret high inrush currents
protection. The feasibility of the setting-less transformer pro- (for example, during energization or during an abrupt volt-
tection is verified with numerical experiments of several case age change) as internal fault currents and falsely trip. This
studies (e.g., transformer energization, external faults, inter- degrades the protection security [1]. In addition, legacy dif-
nal faults, and high impedance faults). The method has been ferential protection may cause false trips because inrush
verified in the laboratory and there are plans for field testing. currents appear as internal faults.
In an attempt to increase the security and dependabil-
ity of transformer protection, many protection algorithms
have been proposed to differentiate inrush currents [1–3];
desensitization of protective differential relays, additional
time-delay settings for differential relays, harmonic-restraint
The work was supported by EPRI, PSERC, DoE/NETL, and NEC. methods, waveform-shape identification, and dwell-time
methods. However, these methods primarily depend on man-
B Sungyun Choi
ual settings which many times compete with other objectives
schoi4ever@gmail.com
such as speed and selectivity. The end result is that settings
1 Korea Electrotechnology Research Institute, Uiwang, Korea are selected as a compromise among competing factors. For
2 School of Electrical and Computer Engineering, Georgia example, the second harmonic restraint method, which iden-
Institute of Technology, Atlanta, GA 30332-0250, USA tifies inrush currents by monitoring the magnitude of second
123
284 Electr Eng (2016) 98:283–297
harmonic currents, requires manual settings for a threshold voltage law or Faraday’s law) relevant to the protection zone.
of the magnitude; a typical setting is 20 % of the fundamental Dynamic characteristics of components enable the proposed
current. Therefore, if the magnitude exceeds the threshold, method to analyze transient phenomena by means of the
then it is concluded that the monitored current is not the dynamic state estimation [7], [8] and, thus, the proposed
fault current, but the inrush current. The additional percent- method can detect any violation of physical laws based on
age adds to the desensitization of the relay settings. There are the component dynamic model and real-time measurement
many cases where the settings cannot guarantee secure and data.
dependable operation; for instance, high impedance internal A possible implementation of the proposed setting-less
faults during transformer energization will be characterized protection is shown in Fig. 1. A relay that is responsible
with high harmonics and relatively low internal fault current, for protecting a component monitors terminal voltages and
a condition that may not be recognized as internal fault by currents of the component with high sampling rates. Addi-
legacy differential protective relay. tionally, the relay collects other information including the
In the meantime, there are efforts to use artificial neural status of circuit breakers and component measurements (e.g.,
network to discriminate inrush currents and fault conditions tap setting or temperature). With all the measurement data
[14,15], but these methods require training periods to fully and the information collected, the dynamic state estimation is
achieve their functionality. In [16], an effective method that performed, generating dynamic operating states of the com-
uses wavelet transform is proposed for indirect symmetri- ponent in real time. Then, the well-known Chi-square test
cal phase shift transformers. This method, however, initially is used to quantify the goodness of fit of measurements to
checks whether the differential currents exceed the prede- the component dynamic model, thereby providing the prob-
fined thresholds to activate its main function. In [17], a new ability that normalized measurement errors are distributed
sympathetic inrush detection method that calculate online within expected bounds. This probability is referred to as the
magnetic flux by integrating measured voltages on the tips confidence level.
of the transformer is proposed, but the sudden change con- In general, a high confidence level implies that there is no
dition of currents beyond the preset thresholds is required to discrepancy between model and measurements, so it can be
initiate the proposed algorithm. concluded that the device operation is normal. Conversely, if
This paper presents a new approach to increase the secu- the confidence level is low, then the model is not consistent
rity of transformer protection without compromising the with measurements. The main reason for this inconsistency
dependability. The new approach is based on dynamic state is any internal fault of the device under protection as long as
estimation for monitoring the health of the protection zone, measurements are trustable. At the same time, the proposed
in this case the transformer [4–6]. This method produces protection scheme can detect violation of operating limits
the real-time operating conditions of the transformer from and may trip the device for predetermined violations as illus-
real-time measured data and the dynamic model of the trans- trated in Fig. 1. It is necessary to point out that the setting-less
former under protection. It provides the confidence level (i.e., protection method requires no setting such as pick-up cur-
the goodness of fit of the model to the measurements). The rents or restraining quantities, because the method basically
confidence level represents the health condition of the pro- depends on the confidence level, which can be obtained by
tected device; for example, low confidence level indicates the dynamic state estimation. However, it can accept limits
there are internal abnormalities in the transformer. The new on operating quantities, such as transformer temperature or
method does not require any settings, so it is coined “setting- transformer current on specific coils (including through fault
less” protection. The setting-less protection approach was current).
introduced in [6]. This paper presents an update with better
transformer models and extensive numerical experiments to
validate the approach. 3 Dynamic state estimation
123
Electr Eng (2016) 98:283–297 285
⎡ ⎤
protection can be used to form the measurement models, ..
.
and thus the dynamic state estimation can capture dynamic ⎡ ⎤ ⎢ ⎢⎡ ⎤T ⎡ ⎤⎥⎥
f 1 (t) ⎢ v(t) v(t) ⎥
characteristics of the device as long as the characteristics are ⎢
⎢ f 2 (t) ⎥ ⎢ ⎢ y(t) ⎥ ⎢ ⎥⎥
f (t) = ⎢ ⎥ = ⎢⎢ ⎥ Q i ⎢ y(t) ⎥ ⎥ , (2)
⎣ v(tm ) ⎦ ⎥
modeled and included in component dynamic models. Once ⎣ f 1 (tm ) ⎦ ⎢ ⎣ v(tm ) ⎦ ⎥
the states variables have been estimated, the Chi-square test ⎢ ⎥
f 2 (tm ) ⎢ y(tm ) y(tm ) ⎥
calculates the confidence level. ⎣ ⎦
..
.
3.1 Protection zone model ⎡ ⎤ ⎡ ⎤
b1 (t − h) M11 M12
⎢ b2 (t − h) ⎥ ⎢ M21 M22 ⎥ i(t − h)
⎢ ⎥=⎢ ⎥
The derivation of the protection zone model is shown in the ⎣ b3 (t − h) ⎦ ⎣ M31 M32 ⎦ 0
Appendix for a three-phase transformer. The dynamic model b4 (t − h) M41 M42
is expressed in the form of a quadratic algebraic compan- ⎡ ⎤
ion form (i.e., QACF). The generalized syntax of the QACF N11 N12
⎢ N21 N22 ⎥ v(t − h)
model is as follows: +⎢ ⎥
⎣ N31 N32 ⎦ y(t − h) , (3)
⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤ N41 N42
i(t) K1 L 11 L 12 L 13 L 14 v(t)
⎢ 0 ⎥ ⎢ K 2 ⎥ ⎢ L 21 L 22 L 23 L 24 ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ y(t) ⎥
⎣ i(tm ) ⎦ = ⎣ K 3 ⎦ + ⎣ L 31 L 32 L 33 L 34 ⎦ ⎣ v(tm ) ⎦ where K is the constant vector; L, M, and N are the device-
0 K4 L 41 L 42 L 43 L 44 y(tm ) model matrices; h is the integration time step; t is the current
⎡ ⎤ ⎡ ⎤
f 1 (t) b1 (t − h) time; tm is the intermediate time (i.e., the half point in the
⎢ f 2 (t) ⎥ ⎢ b2 (t − h) ⎥ interval [t to t − h]); i(t) is the instantaneous value of termi-
+⎢ ⎥ ⎢ ⎥
⎣ f 1 (tm ) ⎦ − ⎣ b3 (t − h) ⎦ , (1) nal currents at t; v(t) is the instantaneous value of terminal
f 2 (tm ) b4 (t − h) voltages at t; y(t) is the internal state variable at t; f (t) is
123
286 Electr Eng (2016) 98:283–297
the quadratic nonlinear terms at t; and Q i are the matrices All equations in (1) of which left terms are zero can
for the quadratic terms. become virtual measurements as follows:
3.2 Measurement model formulation and state 0 = K 2 + L 21 v(t) + L 22 y(t) + L 23 v(tm ) + L 24 y(tm )
estimation + f 2 (t) − b2 (t − h), (11)
0 = K 4 + L 41 v(t) + L 42 y(t) + L 43 v(tm ) + L 44 y(tm )
The measurement model is obtained from the quadratic
+ f 2 (tm ) − b4 (t − h). (12)
algebraic companion form of the device model by simply
expressing the measured quantity as a function of the states
The virtual measurements are expressed as follows:
of the protection zone model. The generalized form of a mea-
surement model is: 0 = Equ(k) + ηr , (13)
zr = h r (x) + ηr , (4) where Equ(k) is the right terms of the k-th row in (11) or
123
Electr Eng (2016) 98:283–297 287
where ν is the degree of freedom, m is the number of measure- A transformer can experience many operating conditions
ments, and n is the number of state variables. The Chi-square including a variety of faults: (a) steady state, (b) transformer
critical value is calculated by summing all squares of normal- energization, (c) overexcitation, (d) external faults, and (e)
ized measurement errors as follows: internal faults. In steady-state conditions, voltages and cur-
rents are near sinusoidal with a frequency of near 60 Hz.
m
During transformer energization, high inrush currents with
h i (x̂) − z i 2
ζ = , (18) harmonics are possible because of the nonlinear character-
σi
i=1 istics of transformer cores. Although inrush currents are
typically five to eight times larger than the rated current, the
where ζ is the Chi-square critical value, x̂ is the best estimate setting-less protective relay will not trip since the inrush cur-
of states, h i (x) is the function of state variables of the ith mea- rents are normal characteristics of transformers. Likewise,
surement, z i is the measured value of the ith measurement, overexcitation is also a possible operating condition when
and σi is the standard deviation of the i-th measurement. the system is experiencing overvoltage. Relays should not
Then, the confidence level is obtained as follows: trip for overexcitation condition unless the conditions persist
to the point that the transformer starts to overheat. Moreover,
Pr[χ 2 ≥ ζ ] = 1.0 − Pr[χ 2 ≤ ζ ] = 1.0 − Pr(ζ, ν) (19) the setting-less protective relay must not operate external to
the transformer faults, yet the relay must trip when internal
faults occur.
where χ 2 is the chi-square random variable; ζ is the Chi-
square critical value; Pr[χ 2 ≥ ζ ] is the probability that χ 2 is
larger than ζ , indicating the confidence level; both Pr[χ 2 ≤ 5 Numerical experiments
ζ ] and Pr(ζ, ν) are the cumulative distribution function of χ 2
with the degree of freedom, ν. This paper presents numerical experiments to verify the fea-
The confidence level indicates the health index of a device sibility of the setting-less transformer protection. For this
(or a zone) under protection. If the confidence level drops to purpose, a test system has been used to simulate the various
zero or low values for several cycles, then the device is in a conditions and events. The WinIGS-T program was used for
fault condition. this purpose [9]. The simulation results are stored in COM-
123
288 Electr Eng (2016) 98:283–297
TRADE format, and the data in the CONTRADE files are is the sign function, and n is the exponent. The exponent
played back into the setting-less protective relay. n is set to eleven, representing the magnetization charac-
Figure 2 shows the single-line diagram of a test power teristics of typical transformer core material. The detailed
system. It consists of a 15 kV–450 MVA-rated generator, component dynamic model of the transformer is presented in
an 18 kV–850 MVA-rated generator, step-up transformers, Appendix.
a step-down transformer, transmission lines, distribution Several case studies are presented in this paper: (a) trans-
lines, loads, and capacitor banks. The setting-less protection former energization, (b) an external fault condition (i.e.,
scheme is tested with the step-down transformer, which is through fault), (c) an internal fault condition, and (d) other
marked with a red dashed square in Fig. 2. The transformer fault conditions including three-phase external fault, phase-
is three-phase, two-winding, delta-wye-connected, saturable to-phase internal fault, three-phase internal fault, and high
iron core, 35 MVA, and 230 kV/25 kV. The magnetizing cur- impedance internal fault. For each study case, the simulated
rent and the flux linkage are as expressed in the following data include the six terminal voltages and six terminal cur-
equation: rents of the transformer with a sampling rate of 20,000 Hz.
These voltages and currents are input to the setting-less pro-
λ(t)
n tective relay as measurement data. The sampling rate used
i m (t) = i 0
sign(λ(t)), (20)
λ0
in this case is high for the present day relays. The proposed
setting-less protection scheme works equally well for differ-
where i m (t) is the magnetizing current, λ(t) is the mag- ent sampling rates [6].
netic flux linkage, i 0 and λ0 are equation constants, sign()
123
Electr Eng (2016) 98:283–297 289
5.1 Case study 1: transformer energization the setting-less relay would not trip, and this decision is cor-
rect since inrush currents are natural characteristics of the
In this case study, the switch connected to the left of the step- transformer due to the saturable core.
down transformer was initially open, so the transformer was
disconnected. The transformer is energized at time 3.495 s by
closing the switch. Inrush currents are observed on the high
voltage side of the transformer. Figure 3 illustrates the six 5.2 Case study 2: external fault condition
terminal voltages and seven terminal currents of the trans-
former. Inrush currents are clearly shown in the primary side This event involves a single phase-A-to-ground fault at
of the transformer (i.e., XFMR1 bus) after the switch was LOAD1 bus in Fig. 2. As shown in Fig. 5, the fault was initi-
closed. ated at 11.995 s and cleared at 12.102 s. Indeed, this external
Figure 4 shows results of the dynamic state estimation fault generated high fault currents, which was also observed
during the test period. The figure shows examples of mea- at the transformers, but the setting-less protective relay recog-
surements superimposed on the estimated measurement—in nized that the fault was external and would not take action. In
this case, the difference between actual measurements and other words, the external fault should be cleared by the pro-
estimated is very small and one cannot distinguish the two tective device that protects the zone where the fault occurred.
curves (measurement and estimated measurement). The con- The test results for this study case are illustrated in Fig. 6.
fidence level is maintained 100 % even when inrush currents Note that the confidence level remained 100 % during the
were flowing through the step-down transformer, indicating external fault period in spite of the high electric currents. As
that there was no internal fault in the transformer. Hence, shown in Figs. 5 and 6, the fault currents due to the external
fault were high enough that may cause false trip of over-
123
290 Electr Eng (2016) 98:283–297
current relays, but the setting-less protection would not trip faults of devices under protection. Figure 8 indicates that the
breakers because the confidence level remained 100 %. estimated values through measurements at the primary side
are much different from the measured values during internal
5.3 Case study 3: internal fault condition faults. Moreover, the confidence level is maintained at zero
during the internal faults, and there are little nonzero spikes;
In this case study, the phase-A terminal at the primary side among them, the noticeable nonzero confidence levels are
of the step-down transformer (i.e., XFMR1 bus) was shorted 13.22 and 5.98 %.
to ground at 11.995 s and cleared at 12.101 s. As described
in Fig. 7, during the internal fault, the phase-A voltage at the
bus XFMR1 was nearly zero, and high fault current flowed 5.4 Case study 4: other fault conditions
through the phase A of the XFMR1 bus. The setting-less
protective relay should be able to detect this internal fault by In addition to the previous case studies, case study 4 presents
computing the confidence level. the various fault conditions, which are listed in Table 1. For
As depicted in Fig. 8, the confidence level dropped to zero case study 4–1, the three-phase fault occurred outside the
when the internal fault occurred. Note that in this simulation, step-down transformer and is cleared. However, for case stud-
the fault was extinguished and, then, the confidence level was ies 4–2, 4–3, and 4–4, internal fault conditions are tested at
back to 100 %. These changes of the confidence level imply the primary side of the transformer (i.e., XFMR1 bus). Note
that the setting-less protection method can detect internal that the high impedance fault (HIF) occurs in case study 4–4.
123
Electr Eng (2016) 98:283–297 291
For all these cases, fault occurred at 11.995 s and is cleared conditions have impacted on the following estimation even at
at 12.101 s. post-fault conditions. However, in the actual implementation
The confidence level of each case is displayed in Fig. 9 of setting-less protection, the dynamic state estimation will
separately. For case study 4–1, the confidence level is main- be reset to initial values on detecting the internal fault and
tained at 100 % even during the faults; this is correct because tripping breakers, and therefore it is more significant that the
the fault location is outside the transformer under protection. confidence level drops to zero as soon as fault occurs. In this
In case of case studies 4–2, 4–3, and 4–4, the results indicate sense, the case study 4–2, 4–3, and 4–4 validates the feasibil-
that the confidence level becomes zero during the internal ity of the setting-less protection in various fault conditions;
fault, which exactly follows the concept of the proposed pro- in case of case study 4–3, the spike of confidence level during
tection scheme. From Fig. 9, it is observed for case study fault conditions can be ignored since the duration of spike is
4–2 that right after the fault is cleared, the confidence level is much short.
recovered to 100 % and then slips to zero momentarily. More- Case study 4–4 represents the high impedance fault case
over, for case study 4–4, the confidence level is maintained at one of the transformer terminals. The given fault resistance
to be zero even after the fault is cleared, and then it becomes is 10 kilo-ohms, which is enough to make waveforms of each
100 % again with a slight dip in a moment. These phenomena terminal look normal as illustrated in Fig. 10; it seems like
are due to the nature of the dynamic state estimation that the that there is no harmonic or high fault current. It should be
previously estimated values affect the consecutive estimation pointed out that the proposed method successfully detects
instants; in other words, badly estimated results during fault the high impedance internal fault as shown in Fig. 9.
123
292 Electr Eng (2016) 98:283–297
The numerical experiments indicate that in the study cases, The proposed setting-less protection method is demonstrated
the confidence level dropped to zero only when an inter- for a transformer via a series of numerical experiments.
nal fault occurred. The health status of the transformer is For this purpose, a three-phase, two-winding, saturable-
normal when inrush currents or external fault currents flow core, delta-wye-connected transformer is modeled, which
through the transformer, thus generating confidence levels is then used for testing the setting-less protection scheme.
of 100 %. Accordingly, the setting-less protective relay pro- A variety of case studies are tested: transformer energiza-
tects the transformer by successfully and reliably detecting tion, external fault conditions, internal fault conditions, and
internal faults. The relay does not compromise the security high impedance fault conditions. The results of the numerical
by correctly identifying inrush currents and through faults experiments indicate that the confidence level is maintained
as “tolerable” conditions for the transformer. It is concluded at 100 % during transformer energization or during exter-
that the proposed setting-less protection scheme improves nal fault conditions, but drops to zero when an internal
dependability and security. fault occurs in the transformer. The numerical results clearly
123
Electr Eng (2016) 98:283–297 293
123
294 Electr Eng (2016) 98:283–297
0 = g(i(t), x(t), t), (21) equations are converted to a set of linear and quadratic alge-
dx(t) braic equations, which are the component dynamic model
= f (i(t), x(t), t), (22)
dt expressed as (1), (2) and (3) in Sect. 3. This procedure is
x(t) = [ v(t) y(t) ]T , (23) shown here for a three-phase transformer bank.
Figure 11 describes the equivalent circuit of the single-
where x(t) is the vector of state variables, i(t) is the vector phase transformer model, which contains two resistances r1
of terminal currents, v(t) is the vector of terminal voltages, and r2 , two inductances L 1 and L 2 , a shunt core resistance
y(t) is the vector of internal states, and t is the current time. rc , a shunt inductance L m , and a magnetizing current i m (t).
These equations are first quadratized, i.e., in case of the exis- The currents i 1L (t) and i 3L (t) represent the terminal currents
tence of nonlinearities of degree larger than two, additional i 1 (t) and i 3 (t), respectively.
variables are introduced so that the resulting differential and The model equations for the single-phase transformer can
algebraic equations do not include any nonlinearities greater be expressed as follows:
than degree two [10], [11]. Subsequently, the quadratic inte-
gration method is used to integrate (21), (22), and (23) with a
i 1 (t) = i 1L (t), (24)
time step determined by three consecutive measured samples
[12], [13], (i.e., by the sampling rate of the data acquisi- i 1 (t) + i 2 (t) = 0, (25)
tion system). As a result, a set of algebraic and differential i 3 (t) = i 3L (t), (26)
123
Electr Eng (2016) 98:283–297 295
2
i 3 (t) + i 4 (t) = 0, (27) λ(t)
0 = y1 (t) − , (33)
rc i 1 (t) + rc N i 3 (t) = rc i m (t) + e(t), (28) λ0
0 = i m (t) − z(t), (29) 0 = y2 (t) − y1 (t)2 , (34)
d 0 = y3 (t) − y2 (t) , 2
(35)
0 = v1 (t) − v2 (t) − e(t) − r1 i 1L (t) − L 1 i 1L (t), (30)
dt 0 = y4 (t) − y1 (t)y3 (t), (36)
d λ(t)
0 = v3 (t) − v4 (t) − N e(t) − r2 i 3L (t) − L 2 i 3L (t), (31) 0 = y5 (t) − y4 (t) , (37)
dt λ0
d
0 = e(t) − λ(t), (32) 0 = −i 0 y5 (t) + z(t), (38)
dt
123
296 Electr Eng (2016) 98:283–297
2 (15) 4 (17)
b B
2 (36) 5 (39)
17-25
(51-59)
c C
3 (37) 6 (40)
26-34
(60-68)
N
7 (41)
where i(t) is the currents, v(t) is the voltages, N is the turn Based on the index mapping, the matrices and vectors (i.e.,
ratio of the transformer, y(t) and z(t) are the internal states, K , L, M, and N ) of the single-phase transformer model in
i 0 and λ0 are the constants, and t is the current time. By (39), (40), and (41) are integrated into one, finally forming
applying the quadratic integration and model quadratization the following three-phase transformer model:
into equations, the single-phase transformer dynamic model
is obtained as follows:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
i 1φ (t) v1φ (t) i 3φ (t) v3φ (t)
⎢ 0 ⎥ ⎢ y1φ (t) ⎥ ⎢ 0 ⎥ ⎢ y3φ (t) ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ i 1φ (tm ) ⎦ = K 1φ + L 1φ ⎣ v1φ (tm ) ⎦ ⎣ i 3φ (tm ) ⎦ = K 3φ + L 3φ ⎣ v3φ (tm ) ⎦
0 y1φ (tm ) 0 y3φ (tm )
f 1φ (t) f 3φ (t)
+ − b1φ (t − h) , (39) + − b3φ (t − h) , (42)
f 1φ (tm ) f 3φ (tm )
⎡ ⎤ ⎡ ⎤
.. ..
. ⎢⎡ .
⎢⎡
⎢ ⎤T ⎡ ⎤⎥⎥ ⎢ ⎤T ⎡ ⎤⎥⎥
⎢ v (t) v1φ (t) ⎥ ⎢ ⎢ v 3φ (t) v3φ (t) ⎥
⎢⎢ 1φ ⎥
f 1φ (t) ⎢ y (t) ⎥ ⎢ ⎥⎥ f 3φ (t) ⎢⎢ (t) ⎥ ⎢ (t) ⎥
⎥⎥
= ⎢ ⎢ 1φ ⎥ Q i ⎢ y1φ (t) ⎥ ⎥ y
= ⎢ ⎢ 3φ ⎥ Q i ⎢ 3φ y
,
⎣ v1φ (tm ) ⎦ ⎥
, ⎣ v3φ (tm ) ⎦ ⎥
(40) (43)
f 1φ (tm ) ⎢ ⎣ v1φ (tm ) ⎦ ⎥ f 3φ (tm ) ⎢ ⎣ v3φ (tm ) ⎦ ⎥
⎢ ⎥ ⎢ ⎥
⎢ y1φ (tm ) y1φ (tm ) ⎥ ⎢ y3φ (tm ) y3φ (tm ) ⎥
⎣ ⎦ ⎣ ⎦
.. ..
. .
i 3φ (t − h) v3φ (t − h)
where the subscript 1φ represents the single phase. b3φ (t − h) = M3φ + N3φ ,
0 y3φ (t − h)
(44)
i (t − h) v (t − h)
b1φ (t − h) = M1φ 1φ + N1φ 1φ .
0 y1φ (t − h)
(41)
where the subscript 3φ indicates the three phase. Note that
Finally, three single-phase transformer models are merged the above model is in the same form (syntax) as the model
into a three-phase transformer dynamic model as illustrated expressed with (1), (2), and (3). The model corresponds to
in Fig. 12. The indices of the single-phase transformer models a transformer bank. A similar model can be developed for
are re-assigned to those of the three-phase transformer model. shell-type or core-type three-phase transformers.
123
Electr Eng (2016) 98:283–297 297
123