Differential Preotection
Differential Preotection
Differential Preotection
A Novel Algorithm for Enhancement of
Security and Reliability of Power Transformer
Protection
Prof. A. A. Hossam Eldin, Fellow, IEEE, M. A. Refaey,
Email: [email protected] Email: [email protected]
Electrical Engineering Department,
Alexandria University
Alexandria, Egypt
426
II. Power Transformer Model Φ j,k (t) = 2 j/2 Φ(2 j t k) (3)
ψ j,k (t) = 2 j/2 ψ(2 j t k) (4)
The single line diagram for the power system under study is
shown in Fig 1. It consists of a three phase 220 kV source If j and k are integers, then the basis functions are scaled by
connected to a three phase, three windings, power a factor of 2 j , and translated k time units. If the time
transformer (300 MVA, YYD 220/132/66 kV) via a 20km
transmission line. The tertiary winding is left opened. When resolution of a discretely sampled function is , then Φ j,k
the tertiary winding is opened the whole primary current is (t) or ψj,k (t) is the same as Φ j0(t – kΔt) or ψj0(t – kΔt) .
drawn by the secondary winding except for a small amount
of circulating current in delta tertiary winding. So, the Let be a discretely sample current function of length 2 N .
transformer model can be approximated and considered as The wavelet transform coefficients of is given by:
two windings transformer. It should be noted that the
differential protection current transformers on the primary 2 N - j -1 J 2 N - j -1
and secondary windings are connected in delta. i (t ) = a jk f jk (t ) + å
å å d y (t )jk jk (5)
k =0 j =1 k =0
( )
where, J J £ N is the number of levels of
decomposition, or depth of the tree representing the
decomposition algorithm. Each set of wavelet transform
coefficients {ajk} and {d jk} for 1 £ j £ J , represents a
bandpass filtered and downsampled version of the original
signal i(t).
Fig 1 Proposed power system model
The decomposition process can be iterated, with successive
Simulation of the power system during steady state, external approximations being decomposed in turn, so that one
faults, energization conditions, and internal faults is based signal is broken down into many lower resolution
on 100% of transformer windings using PSCAD/EMTDC components. In reality, the decomposition can proceed only
simulation package. until the individual details consist of a single sample. In
practice, a suitable number of levels will be selected based
III. Discrete Wavelet Transform on the nature of the signal.
The output of DWT is a finite number of wavelet Since dyadic DWT is being used, the first detail will cover
coefficients. It turns out that scales and positions based on from fs /2 to fs /4, and second detail from fs/4 to fs /8, and so
powers of two are chosen. Our analysis will be much more on, where f s is the sampling frequency. The applied
efficient, accurate and of smaller execution time. The sampling frequency here is 20 kHz corresponding to 400
scaling function and mother wavelet are functions that samples per power frequency cycle.
satisfy the following twoscale equations where n is an
integer: [5] The used mother wavelet is Daubechies with four
F (t ) = 2 å h n F(2 t - n ) (1) coefficients “db4”. The second step is to determine the
n
number of resolution levels. The number of the data
window samples is 64, which means that the number of
y (t ) = 2 å g n y (2 t - n ) (2) resolution levels can be up to 6 levels. The optimum number
n of resolution levels in the analysis was chosen to be 5
Where, Φ(t) is the scaling function and ψ(t) is the mother levels.
wavelet for some set of coefficients H = { h n }and G = {
The frequency band represented by each level depends on
n
g n } such that g n = (- 1 ) h 1 - n . That is, G is the reverse the number of resolution levels as well as the sampling
frequency used. Applying DWT to the sliding data window
of H, interleaved with sign changes. Each wavelet scaling
of 64 samples size, with sampling frequency of 20 kHz,
function and mother wavelet function is represented by a set
resulted in 5 levels of resolutions with 5 details and an
of coefficients H and G. These coefficients are lowpass and
approximation whose frequency bands are listed in table1.
highpass Finite Impulse Response (FIR) filter coefficients,
respectively. The wavelet basis functions Φ j,k (t) and
ψj,k(t) are scaled and translated copies of Φ(t) and ψ(t).
427
Table 1 Frequency bands of the result DWT components signals are asymmetric of periodic nature and decaying
amplitude; so, the MAD ratio wave is a noisy periodic wave
Level Wavelet Component Frequency band and each period posses a component of high amplitude
number component type (Hz) corresponding to the sharp edges in the inrush currents.
1 d1 Detail 10000:5000
2 d2 Detail 5000:2500 It should be noted that this discriminating criterion was
3 d3 Detail 2500:1250 applied to different details but the expected mentioned
4 d4 Detail 1250:625 results were not noticed clearly except when using detail d3
5 d5 Detail 625:312.5 and detail d4.
5 a5 Approximation 312.5:156.25
IV. Feature Extraction
It is a more robust estimator of scale than the sample
variance or standard deviation. The differential current
signals
Id1 = (Iap Icp) (Ias Ics) (7)
Where, Iap, Ibp, Icp, Ias, Ibs and Ics are the primary and
secondary power transformer currents
First of all, the MAD ratio of d4 to d3 (MAD of d4 divided
by that of d3) is expected to be zero under normal
conditions of differential currents (steady state sine wave).
This is due to the fact that the wavelet coefficients under
normal conditions are approximately zero. When a certain
fault is introduced to the power system model, an
instantaneous rapid change in the MAD ratio is actuated
which is equivalent to the wavelet decomposition
coefficients at sharp edges. Then the MAD ratio wave is to
be a step noisy signal corresponding to the asymmetry in the
transformer phase currents after fault inception.
The second case introduced here is the inrush current Fig. 2 Flowchart of the proposed algorithm
with/without residual magnetism. In this case, the current
428
V. Simulation Results shape is completely different from that in figures 35. The
wave shape is of repeated nature having rapid rising and
The performance of the proposed technique was evaluated falling instants corresponding to the rapid change instants in
for different types of internal and external faults. the inrush current waveforms.
1. Internal phase to ground Fault: It is a single phase to
ground on phase B of the transformer secondary side
upstream the secondary circuit breaker.
2. Double phase to ground fault: It is a fault between two
phases B and C on the transformer secondary side
upstream the secondary circuit breaker.
4. External faults: The external faults are modeled in the
form of a single phase B to ground fault occurring on
the secondary side of the transformer downstream the
secondary circuit breaker.
5. Inrush current: In order to study the magnetizing Fig 3 MAD ratios in case of internal single linetoground fault
current accurately, a load is connected in parallel with
the transformer primary circuit breaker in order to
enable source voltage build up. Also, the saturation is
enabled through the saturation curve settings defined by
the user. The model of inrush current with no residual
magnetism is studied. Then, the residual magnetism
effect is taken into consideration through adding two
additional circuit breakers to simulate previous
switching operations.
Figure 3 shows the case of an internal fault (single phase to
ground). It is clear from the ratio plot concerning Id2 and Id3
that the MAD ratio is approximately zero for normal
operation until the time of fault application a rapid change is
noticed in the MAD ratio values, then it nearly saturates
about a certain value. I d1 is not the same because it depends Fig 4 MAD ratios in case of double linetoground fault
on the currents from the nonfaulty phases so, it is always
approximately zero.
Figure 4 shows the case of an internal fault (double phase to
ground). The MAD ratio plot concerning Id1, Id2 and Id3 is
approximately zero for normal operation until the time of
fault application a rapid change is noticed in the MAD ratio
values, then it nearly saturates about a certain value. Here
all differential currents depend on the faulty phases, so all
ratio plots have almost the same shape.
429
[4] H. KhorashadiZadeh, “FuzzyNeuro Approach to Differential
Protection for Power Transformer”, IEEE TENCON, Vol. 3, Nov.
2124, 2004, pp. 279282.
[5] M. M. Eissa, ”A Novel Digital Directional Transformer Protection
Technique Based on Wavelet Packet”, IEEE Transactions on Power
Delivery, vol. 20, no. 3, pp. 18301836, JULY 2005.
Fig 6 MAD ratios in the case of an inrush current without residual
magnetism
Fig 7 MAD ratios in the case of an inrush current with residual
magnetism
VI. CONCLUSIONS
The proposed technique can differentiate precisely between
internal and external faults as well as inrush currents. The
speed of decision making is good; where any change in the
current signals is detected and analyzed rapidly. The
algorithm is flexible and can be easily adapted to new
elements or functionalities in the power system.
Generalization capability of the proposed algorithm is
achieved. Execution time of the novel technique is superior
to previously applied ones.
I. REFERENCES
[1] S. A. Saleh, M. A. Rahman, "Modeling and protection of three
phase power transformer using wavelet packet transform," IEEE
Transactions on Power Delivery, vol. 20, no. 2, pp. 12731282,
April 2005.
[3] M.Shin, C. Park, and J. Kim, “Fuzzy LogicBased Relaying for
Large Power Transformer Protection” IEEE Transactions on Power
Delivery, vol. 18, no. 3, July 2003, pp. 718723.
430