Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Applications of Fuzzy-Logic-Wavelet-Based Techniques for Transformers

Inrush Currents Identification and Power Systems Faults Classification.


OMAR A.S.YOUSSEF, Senior Member, IEEE

Abstract- The advent of wavelet transforms (WTs) and fuzzy- ponents to solve the classification problem and the
inference mechanisms (FIMs) with the ability of the first to focus
training is carried out using the information from
on system transients using short data windows and of the second
to map complex and non-linear power system configurations both designer’s experiences and sample data sets.
provide an excellent tool for high speed digital relaying. This The other drawback of this approach is that the
paper presents a new approach to real-time fault classification in number of fuzzy rules increases exponentially with
power transmission systems, and identification of power
respect to inputs and as a consequence 17 rules are
transformers magnetising inrush currents using fuzzy-logic-based
multi-criteria approach [1,2], with a wavelet-based pre-processor framed for 3 inputs only.
stage [3-6]. Three inputs, which are functions of the three line To avoid the needless trip by magnetising inrush
currents, are utilised to detect fault types such as LG, LL, LLG as currents, the second harmonic component is
well as magnetising inrush currents. The technique is based on
commonly used for blocking differential relay in
utilising the low-frequency components generated during fault
conditions on the power system and/or magnetising inrush power transformers, and consequently, it is used as a
currents. These components are extracted using an online discrimination factor between fault and inrush
wavelet-based pre-processor stage with data window of 16 currents. Previous work on transformer protection is
samples (based on 1.0 kHz sampling rate and 50 Hz power
indicated in [19-22]. Some techniques have been
frequency). Generated data from the simulation of an 330∆/33Υ
kV, step-down transformer connected to a 330 kV model power adopted to identify the magnetising inrush and
system using EMTP software were used by the MATLAB internal faults. In [23] a modal analysis in
program to test the performance of the technique as to its speed of conjunction with a microprocessor-based system
response, computational burden and reliability. Results are shown was used as a tool for this purpose. In [24] the active
and they indicate that this approach can be used as an effective
tool for high-speed digital relaying, and that computational power flowing into transformer is used as a
burden is much simpler than the recently postulated fault discrimination factor which is almost zero in case of
classification. energisation. In [25] wavelet-based system is used.
This paper proposes an application of a new
Index Terms- fuzzy sets – knowledge-based systems – decision
making – fuzzy control – wavelet transforms – wavelet analysis -
multi-criteria-fuzzy-logic- based technique with an
protective relaying – digital signal processors – protective online wavelet-based pre-processor stage to identify
relaying – transient analysis – transformers - Phase selection fault conditions in order to determine the nature of
single-pole autoreclosure the fault and to distinguish between them and
transformer magnetising inrush currents. The
property of multi-resolution in time and frequency
I. INTRODUCTION provided by wavelets is utilised to accurately locate
transient components while simultaneously retaining
Almost all theories developed in the field of information about the fundamental frequency and its
transmission lines protective schemes are based on low-order harmonics, which facilitates the detection
deterministic computations on a well-defined model of transformer inrush currents and power system
of the system to be protected. This results in fault conditions. A sample 330∆/33Υ kV step-down
difficulty because of the complexity of the system power transformer connected to a model 330kV
model, the lack of knowledge of its parameters, the three-phase power system was simulated using the
great number of information to be processed, and the EMTP software [26]. The line currents were then
difficulty in taking into consideration any system processed using an online wavelet transform
variations as the rules are fixed. Recently, the algorithm [3-6] to extract the low-frequency
application of the fuzzy set theory is introduced to (characteristic) components generated during faults
solve uncertainty problems [7-8]. All situations that as well as magnetising inrush conditions. The
are not characterised by a simple and well-defined performance of the proposed model was extensively
deterministic mathematical model can be more easily tested using test sets. The paper is organised as
handled in terms of the fuzzy-set theory, in which follows. Section II presents brief overview of both
simple rules and a number of simple membership wavelet transform analysis and fuzzy logic
functions (MFs) are used to derive the correct result. techniques, and outlines the proposed scheme.
Some of the fuzzy logic applications in power system Section III demonstrates the simulation and
protection are included in [9-18]. One such discusses the test results. Section IV draws the
application to fault classification problem is found in conclusions.
[18]. However, this approach needs sequence com-

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.
level-4 and -1(scale-2) for a BC fault are shown in
II. ALGORITHM Fig.5.
Ia
A. Sample Power System
Coeff. Coeff. vector
The simple three-phase power system, shown in LD1Ia HD1Ia lengths
vectors
Fig. 1, was chosen for the purpose of generating line
currents under normal and fault conditions. System
HD2Va
parameters are listed in the Appendix. Database of LD2Ia
HD1Ia L(Ia)

line currents is built up for various types of faults HF- level 1


under different system conditions, using EMTP LD3Ia HD3Ia L(HD1Ia )

software. As an example of the simulation results, the HD2Ia

HF- level 2
power spectral densities of an inrush current and a
typical BCG fault on BB2 are shown in Fig.2. LD4Ia HD4Ia L(HD2Ia

50 km HD3Ia

Ia, b,c 330∆/33Υ kV HF-level 3


L(HD4Ia )

GS π
BB1 BB2 BB3 HD4Ia

HF-level 4 L(HD4Ia )

Fig. 1. Model system under consideration.


(Ia)Inrush (Ia)Fault -3
x 10
4 0.04 5 4 LD4Ia L(LD4Ia )
PSD(Ia)Inrush PSD(Ia)Fault LF-level 4
2
0.02 0 2
A

-2 0 -5 0 Fig. 3 Four-level wavelet decomposition of line current Ia.


(Ib)Inrush (Ib)Fault
x 10
-3
HD, LD: HF, and LF decomposition coefficients respectively.
1 0.03 5 6
PSD(Ib)Inrush PSD(Ib)Fault
L (HD2 Ia): Length of the HF decomposition coefficients.
0 0.02 4
0 Ia Ib Ic
A

-1 0.01 2

-2 0 -5 0
(Ic)Inrush (Ic)Fault -3
x 10
1 0.04 5 6 A
PSD(Ic)Inrush PSD(Ic)Fault
0 0 4
0.02
A

-1 -5 2

-2 0 -10 0 0 100 200 0 100 200 0 100 200


0 50 100 100 200 300 200 250 100 200 300
Sample frequency Sample frequency DWT of Ia DWT of Ib DWT of Ic

Fig. 2 Power spectral density (PSD) of line currents I a,b,c


Le
during inrush current, and an BCG fault on BB2. vel

B. Online wavelet-based pre-processor [3-6]


To remove harmonic, non-harmonic components
50 100 150 200 50 100 150 200 50 100 150 200
and to extract the low-frequency (characteristic)
components, an online WT technique with sliding Fig. 4 Four-level discrete wavelet transform of line currents
data window approach are used. Taking into Ia,b,c during an AG fault at BB3.
consideration that the maximum number of W Ia2 , W Ia4 - B C Fault
1
decomposition and reconstruction levels of WT 6W Ia4 W Ia2
analysis is governed by log2(N), where N is the 0.5
A

number of samples in the data window (16 samples 0

in this work), level-4 wavelet transform components -0.5


W Ib2 , W Ib4 - B C F ault
(approximation components) of the three line 1
6W Ib4 W Ib2
currents Ia,b,c are computed on the delta-side (HV) of 0.5
A

the transformer. Those are denoted as WIa4, WIb4, 0

and WIc4. The decomposition process of the wavelet -0.5


W Ic 2 , W Ic 4 - B C Fault
transform analysis is shown in Fig. 3, while the 1

discrete wavelet transform (DWT) of line currents 0


A

Ia,b,c of a typical AG fault are shown in Fig.4. It -1


6W Ic 4 W Ic 2
should be pointed out that due to the dyadic nature -2
1900 1950 2000 2050 2100 2150 2200 2250 2300 2350 2400
of the wavelet transform process, level-4 S am ple

components have frequency around 12.5 Hz Fig. 5 Level-1 (scale 2) and level-4 approximation components
(50Hz/4). Typical approximation components of of line currents Ia,b,c during an BC fault at BB3.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.
C. The Multi-Criteria Fuzzy-Logic Processor [1-3] Figs. 8-10 for an AG, BC and BCG faults
The block diagram in Fig.6 describes the process respectively, where the characteristic (level-4
performed in a fuzzy-set approach. The process uses approximation) components are shown, from which
a collection of fuzzy membership functions (MFs) it is clear that:
and rules, instead of Boolean logic, to reason about AG fault: Sab<< Dab BC fault: Sab>> Dab
data. Crisp values are first transformed into fuzzy BCG fault: Sab> Dab
values to be able to use them to apply rules Where Sab, Dab are the sum and difference of line
formulated by linguistic expressions. Then, the currents Ia and Ib respectively, i.e.
fuzzy system transforms the linguistic conclusion
back to a crisp value. These steps are described as Sab= WIa4 + WIb4 Dab= WIa4 - WIb4
follows: Sbc= WIb4 + WIc4 Dbc= WIb4 - WIc4
Crisp inputs
According to this notation, the square of the
amplitude of the sum and difference of the level-4
System
Fuzzy inputs Fuzzy inference approximation components of line currents Ia,b,c are
description system computed during every sampling interval based on
frequency of 12.5 Hz, i.e.
Fuzzy Rule Fuzzy System ∧ 2 (S ab (k +1) − S ab (k −1) ) 2
Base 2
S ab (k ) = [S ab (k ) + ]
(2πfh ) 2
Defuzzification Where, h: is the sampling interval (1x10-3 sec.) and
f: is the level-4 frequency (12.5 Hz). Also, ratios of
sum/difference are computed, i.e.
Crisp output
SDab =(Sab/Dab)2 & SDbc =(Sbc/Dbc)2
Fig. 6 Fuzzy control system. SDca = SDca /SDca

a. Fuzzification: crisp input(s) values are


transformed into fuzzy sets to be able to use them
for computing the truth-values of the premise of
each rule in the rule base.
b. Inference: the truth- value for the premise of each
logic rule is computed and applied to the
conclusion part of each rule. This leads to one
fuzzy set to be assigned to each output variable
for each rule.
c. Composition: all of the fuzzy sets assigned to
each output variable are combined together to
form a single fuzzy set for each output variable.
d. Defuzzification: the single fuzzy sets are
converted back to crisp values.
The fuzzy processor block represents the heart of the Fig. 7. Current distribution in ∆/Υ transformer during
whole fuzzy-logic process. It evaluates the overall different power system faults.
truth grade of a set of rules that describe the system
in a “ fuzzy “ way. Fault SDab SDbc SDca
HEALTHY H1 H1 H1
D. Derivation of the Characteristic Components AG L1 H2 H2
BC H4 L2 L2
(Level-4 approximation components)
BCG H3 L3 L3
In reference to Fig.7 the currents distribution INRUSH H1 or H3 H1 or H3 H3 or H1
during AG, BC, and BCG faults occurring on the BG H2 L1 H2
power system on the low voltage side of the CA L2 H4 L2
transformer are shown, from which we can observe CAG L3 H3 L3
that during an AG faults, the line currents Ia,b on the CG H2 H2 L1
ABC L2 L2 H4
∆-side are in opposite directions, while during an BC ABG L3 L3 H3
fault, the line currents Ia,b are in the same direction L1 < .01 , L2 = .15 - .25 , L3 = .3 - .8
and are approximately equal. In case of BCG faults H1 < .5 , H2 = .8-1.2 , H3 = 2 - 4 , H4 = 20 - 200
the two line currents Ia,b are in the same direction but Table 1. Fault type classification index.
not necessarily equal. This concept is illustrated in

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.
LG-Case1 LG-Case2
0.5 E. Definitions of MFs
WIa4 WIb4
WIa4
Based on currents distribution in ∆/Υ transformers
0 WIc4 and extensive conducted simulation studies under
A

WIc4
WIb4 different system conditions, it is found that SDab,
-0.5 SDbc, and SDca varies according to the limits
LG-Case3 LG-Case4
0.2 indicated in Table 1. A typical case for an BC fault
WIb4
WIb4 at BB2 are shown in Fig.11. Based on these values,
0 WIc4 the MFs are defined as in Fig.12, of which
A

WIc4

WIa4 WIa4 parameters are:


-0.2 BC Fault BC Fault BC Fault
0.2 0.5 0.5
LG-Case5 LG-Case6
W Ia4 W Ib4 W Ib4 W Ia4
0.5
0 0 0

A
WIa4 WIb4
W Ic4 W Ic4
WIc4 -0.2 -0.5 -0.5
0 WIc4
A

0.5 0.2 0.2


WIb4 WIa4 S =W I +W I
BC b4 c4 S =W I +W I
CA c4 a4
-0.5 0 0 0

A
2000 2050 2100 2050 2100 2150 2200 S =W I +W I
AB a4 b4
Sample Sample -0.5 -0.2 -0.2
0.01 0.5 0.5
Fig. 8 Level-4 approximation components of line currents D
AB
=W I -W I
a4 b4 D =W I -W I
CA c4 a4
Ia,b,c in case of AG fault. 0 0 0
A D =W I -W I
LL-Case1 LL-Case2 BC b4 c4
-0.01 -0.5 -0.5
0.5
WIc4 100 1 4
WIa4 WIb4 |S |/|D |
|S |/|D | BC BC
0 50 AB AB 0.5 2 |S |/|D |
A
A

CA CA

WIc4 WIa4 WIb4 0 0 0


2000 2200 2400 2000 2200 2400 2000 2200 2400
-0.5
Sample Sample Sample
LL-Case3 LL-Case4
0.5
WIa4 WIb4 Fig. 11 Characteristic components SDab , SDbc , SDca , in case of
WIa4 WIb4
an BC fault at BB2.
0
A

WIc4 WIc4
-0.5 1 L1 L2 L3

LL-Case5 LL-Case6
LOW MFs (INPUT)
0.5 0.5
WIc4 WIa4 WIb4
SDab ,SDbc,SDca
0
0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
A

Degree of Membership

WIa4 WIb4 H2 H3 H4
WIc4 1
-0.5
2000 2050 2100 2050 2100 2150 2200 HIGH MFs (INPUT)
Sample Sample 0.5
H1 SDab ,SDbc ,SDca
Fig. 9 Level-4 approximation components of line currents 0
0 5 10 15 20 25
Ia,b,c in case of BC fault.
NORM AG BC BCG MAG BG CA CAG CG AB ABG
1
LLG-Case1 LLG-Case2
0.5 0.5 OUTPUT MFs
WIb4 WIc4
WIa4 WIa4 0
0 0 2 4 6 8 10 12
A

Fault Type Index


WIc4 WIb4
-0.5 Fig. 12 Definitions of Membership functions of inputs and
LLG-Case3 LLG-Case4 output.
0.5
WIb4 WIb4
WIa4 The parameters of the membership functions are:
0 WIa4 1. Low(L) MFs: L1, L2 and L3 are trapezoidal
A

WIc4 functions defined by [ 0 0 .04 .05], [.1 .15 .25 .3]


WIc4
-0.5 and [.2 .3 .9 1.0] respectively.
LLG-Case5 LLG-Case6
0.5 2. High (H) MFs: H1 is a triangular function defined
WIc4 WIb4 by [0 0 .5], while H2, H3 and H4 are trapezoidal
0 WIa4 WIa4
functions defined by [.5 .6 1.4 1.5], [1 2 4 5]
A

and [20 21 200 200] respectively. H4 is shown in


WIb4
-0.5
WIc4 Fig.12 up to 25. MFs associated with different
2000 2050 2100 2050 2100 2150 2200 power system conditions are illustrated in Table.2
Sample Sample

Fig. 10 Level-4 approximation components of line currents


Ia,b,c in case of BCG fault.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.
Power System Condition Membership Functions fault resistance considered to simulate fault currents
1. Healthy H1 --- with the same energisation angles were 0.01Ω, 1.0Ω
2. Mag. inrush H1 H3
3. LG fault L1 H2 and 50.0 Ω. Sampling rate of 1.0 kHz has been
4. LL L2 H4 considered for the algorithm (20 samples per power
5. LLG L3 H3 frequency cycle based on 50 Hz). MATLAB has
Table 2 MFs associated with power system conditions. been used to implement the algorithm using the
three line currents derived from the EMTP. In the
F. Selection of fuzzy inference mechanism(FIM) fault tests the following parameters were considered:
and Definitions of Fuzzy Rules [2] 3 different source capacities, 3 different fault types,
Mamdani inference mechanism has been selected 3 different fault resistance, 2 different fault locations
with the CENTROID defuzzification method,
at BB2 and BB3, and 11 different fault inception
MINIMUM implication, and MAXIMUM
angles. SDab, SDbc and SDca were used as the
aggregation. The fuzzy rules of the FIM are characteristic functions. Typical cases are shown in
illustrated in Table 3. Extreme healthy and faulty Fig. 13-15 for AB, BC, and BCG faults, while the
power system conditions have been considered in fuzzy outputs along with SDab, SDbc and SDca are
the problem domain, thus avoiding any unexpected shown in Figs. 16-19 for AG, BC, BCG and
circumstances. Linear MFs have been selected transformer inrush conditions respectively. In case
(trapezoidal- triangular) thus reducing the of AG fault, the fuzzy output is 2, while it is 3 in
computational burden to minimal. Only three inputs case of BC fault, and 4 in case of BCG. Fuzzy
SDab, SDbc and SDca with seven linguistic variables outputs 5 and 1 correspond to a transition from
(L1, L2, L3, H1, H2, H3, and H4) and one output (OP)
transformer energisation to healthy conditions
with eleven linguistic variables specifying the respectively.
different power system conditions are defined for the The above results reveal that the proposed approach
FIM. is computationally simple in comparison to other
State Fuzzy Rule conventional approaches and yields classification in
Healthy If A is H1 and B is H1 and C is H1 THEN OP = 1 less than half power frequency cycle.
Inrush If A is H1 and B is H1 and C is H3 or Test results show that the proposed wavelet-fuzzy-
If A is H1 and B is H3 and C is H1 or based technique is effective in determining the faulty
If A is H3 and B is H1 and C is H1 THEN OP = 5
AG If A is L1 and B is H2 and C is H2 THEN OP = 2
phase(s) under different fault locations, fault
BG If A is H2 and B is L1 and C is H2 THEN OP = 6 occurring times, presence of fault resistance and
CG If A is H2 and B is H2 and C is L1 THEN OP = 9 variations in source impedances, and that transformer
BC If A is H4 and B is L2 and C is L2 THEN OP = 3 inrushAGcurrents
Fault can be detected.
AG Fault AG Fault
CA If A is L2 and B is H4 and C is L2 THEN OP = 7 0.5 0.5 0.5
WIa4 WIa4
AB If A is L2 and B is L2 and C is H4 THEN OP = 10 WIc4 WIc4
0 0 0
A

BCG If A is H3 and B is L3 and C is L3 THEN OP = 4


WIb4 WIb4
CAG If A is L3 and B is H3 and C is L3 THEN OP = 8 -0.5 -3 -0.5 -0.5
x 10
ABG If A is L3 and B is L3 and C is H3 THEN OP = 11 5 0.5 0.5
S AB=WIa4+WIb4 S BC=WIb4+WIc4
Table 3. Rules of the multi-criteria FIM.
0 0 0
A

S CA =WIc4+WIa4
III. Digital Simulations -5 -0.5 -0.5
0.5 0.5 0.5
DBC=WIb4-WIc4 DCA =WIc4-WIa4
Using the model system described earlier, the 0 0 0
A

proposed technique is tested using simulated data DAB=WIa4-WIb4


-0.5 -0.5 -0.5
from the EMTP. The transformer was simulated 2 4 2
with 330kV, ∆-connected primary, 33kV, Y- |S AB|/|DAB| |S BC|/|DBC| |S CA |/|DCA |
1 2 1
connected secondary with the neutral grounded.
A

With the data given in Appendix, subroutine 0 0 0


2000 2200 2400 2000 2200 2400 2000 2200 2400
BCTRAN in EMTP is used to obtain transformer Sample Sample Sample
parameters. The simulations provide samples of
Fig.13 Characteristic components SDab, SDbc and SDca in case
currents in each phase on the ∆-side of the of an AG fault.
transformer when it is energised or when a fault
occurs on the system on the Υ-side. Varying shunt IV. CONCLUSION
fault resistance, fault location, fault inception angles The paper presented an online application of a
and prefault were considered. Data from the multi-criteria fuzzy-logic-based technique to fault
simulations are used as input to the algorithm to classification and transformer inrush current
identify its response. A total of 594 fault cases and identification. The approach in terms of fuzzy-set
11 inrush cases were simulated to test the various procedures was proved to be effective in
features of the algorithm. The inrush currents used implementing a simple fuzzy procedure (11 rules)
correspond to every 2 ms. of a complete cycle. The to solve a problem that requires more complex

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.
BCG Fault BCG Fault BCG Fault
0.5 0.5 0.5 1000
WIb4 WIb4 WIa4 RatioAB= |WIa4 + WIb4|2 / |WIa4 - WIb4|2
0 0 0 500
A

WIa4
WIc4 WIc4
-0.5 -0.5 -0.5 0
0.5 0.2 0.5 2
SBC=WIb4+WIc4 SCA =WIc4+WIa4 Ratio = |WI + WI |2 / |WI - WI |2
0 0 0 1 BC b4 c4 b4 c4
A

S =WI +WI
AB a4 b4
-0.5 -0.2 -0.5 0
ratioAB / ratioBC
0.2 1 1 3000
D =WI -WI D =WI -WI 2000
AB a4 b4 CA c4 a4
0 0 0
A

DBC=WIb4-WIc4 1000
-0.2 -1 -1 0
Fuzzy Output
100 1 1
|S |/|D | 5
|SAB|/|DAB| BC BC |S |/|D |
CA CA
50 0.5 0.5
A

0 0 0 0
2000 2200 2400 2000 2200 2400 2000 2200 2400 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000
Sample Sample Sample Sample

Fig.14 Characteristic components SDab, SDbc and SDca in case Fig.17 Fuzzy output (1 -3) that corresponds to transition
of an BC fault. from healthy to BC fault conditions respectively.

BCG Fault BCG Fault BCG Fault 10


0.5 0.5 0.5 Ratio = |WI + WI |2 / |WI - WI |2
WIb4 WIb4 AB a4 b4 a4 b4
WIa4 5
0 0 0
A

WIa4
WIc4 WIc4 0
-0.5 -0.5 -0.5
1.5
0.5 0.2 0.5
2 2
SBC=WIb4+WIc4 SCA =WIc4+WIa4 1 RatioBC= |WIb4 + WIc4| / |WIb4 - WIc4|
0 0 0
A

0.5
S =WI +WI
AB a4 b4 0
-0.5 -0.2 -0.5 ratioAB / ratioBC
0.2 1 1 200
D =WI -WI D =WI -WI
AB a4 b4 CA c4 a4
0 0 0 100
A

DBC=WIb4-WIc4
-0.2 -1 -1 0
Fuzzy Output
100 1 1
|SAB|/|DAB| |S |/|D |
BC BC |S |/|D | 5
CA CA
50 0.5 0.5
A

0 0 0
2000 2200 2400 2000 2200 2400 2000 2200 2400 0
1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000
Sample Sample Sample
Sample

Fig.15 Characteristic components SDab, SDbc and SDca in case Fig.18 Fuzzy output (1 -4) that corresponds to transition
of an BCG fault. from healthy to BCG fault conditions respectively.

2 5

RatioAB= |WIa4 + WIb4|2 / |WIa4 - WIb4|2 SDab-bc


1

0 0
10 300

RatioBC= |WIb4 + WIc4|2 / |WIb4 - WIc4|2 200 SDbc-ca


5
100
0 0
ratioAB / ratioBC
10 500

5 SDca-ab

0 0
Fuzzy Output
5 5
Fuzzy Output

0 0
1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 100 200 300 400 500 600 700 800 900 1000
Sample Sample

Fig.16 Fuzzy output (1 -2) that corresponds to transition Fig.19 Fuzzy output (5-1) that corresponds to transition from
from healthy to AG fault conditions respectively. transformer inrush to healthy conditions respectively.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.
algorithms when approached in a deterministic way. [17] H.T. Yang, and C.C. Liao, “ Adaptive Fuzzy Diagnosis for
The technique is proved to be accurate and robust dissolved Gas Analysis of Power Transformers”, IEEE
Transactions on Power Delivery, Vol. 14, No. 4, pp. 1342-
and would perform accurately for various system 1350, October 1999.
conditions. The approach is found to be [18] A. Ferrero, S. Sangiovanni, and E. Zappitelli, “ A Fuzzy –
computationally much simpler than the conventional Set Approach to Fault-Type Identification in Digital
approaches which are currently used for transient Relaying”, IEEE Transactions on Power Delivery, Vol. 10,
No. 1, pp. 169-175, January 1995.
disturbance classification. [19] Ingaki K., Higaki M., Matsui Y., Kurita K. , Suzuki M.,
Yoshida K. , and Maeda T, “Digital Protection Method for
V. REFERENCES Power Transformers Based on an Equivalent Circuit Composed
of Inverse Inductance”, IEEE Transactions on Power Delivery,
Vol. 3, No.4, pp. 1501-1510, Oct. 1988.
[1] Omar A.S.Youssef, “ Combined Fuzzy-Logic-Wavelet-Based
[20] Akimoto Y., Nishida S., Sakaguchi T., “ Transformer
Fault Classification Technique for Power System Relaying”,
Protection Scheme Based on a Model Including Nonlinear
paper # TPWRD-00471-2002, IEEE Transactions on Power
Magnetizing Characteristics”, IEEJ Transactions, Vol. 98-B,
delivery, under print, Vol. 19, No.2, April 2004.
No. 8, pp.703-710, Aug. 1978.
[2] Omar A.S.Youssef, “Applications of Fuzzy Inference
[21] Phadke A.G., and Thorp J.S., “ A New Computer-Based
Mechanisms to Power System Relaying”, paper # TPWRD-
Flux-Restrained Current-Differential Relay for Power
00557-2003, submitted for publication in the IEEE
Transformer Protection”, IEEE Transactions on Power
Transactions on Power delivery, Nov. 2003.
Apparatus and Systems, Vol. PAS-102, No. 11, pp. .3624-
[3] Omar A.S.Youssef, “ Online Applications of Wavelet
3629, Nov. 1983.
Transforms to Power Systems Relaying ”, IEEE Transactions
[22] Wiszniewski A., and Kasztenny B., “ A Multi-Criteria
on Power delivery, Vol. 18, No.4, Oct. 2003, pp. 1158-1165.
Differential Transformer Relay Based on Fuzzy Logic”, IEEE
[4] Omar A.S.Youssef, "New Algorithm to Phase Selection Based
Transactions on Power Delivery, Vol. 10, No.4, pp. 1786-1792,
on Wavelet Transforms", ”, IEEE Transactions on Power
Oct. 1995.
delivery, Vol. 17, No.4, Oct. 2002, pp. 908-914.
[23] T.S.Sidhu, M.S. Sachdev,” On-Line Identification of
[5] Omar A.S.Youssef, " A Wavelet-Based Technique for
Magnetizing Inrush and Internal Faults in Three-Phase
Discrimination between Faults and Magnetising Inrush
Transformers”, IEEE Trans. On Power Delivery, Vol.7, No.4,
Currents in Transformers", IEEE Transactions on Power
Oct. 1992, pp.1885 – 1891.
delivery, Vol. 18, No.1, Jan. 2003, pp. 170-176.
[24] Kukiaki Yabe,” Power Differential Method for
[6] Omar A.S.Youssef, “ Fault classification based on wavelet
Discrimination Between Fault and Magnetizing Inrush Current
transforms”, paper # 01TD069, IEEE, T&D Conference, 28
in Transformers”, IEEE Trans. On Power Delivery, Vol. 12,
Oct. - 2 Nov. 2001, Atlanta, Georgia
No. 3, July 1997, pp. 1109 –115.
[7] Zadeh L. A.," Fuzzy sets”, Information Control, No. 8, 1965,
[25] M.G.Morante, D.W. Nicoletti, “A Wavelet-base Differential
pp. 338-353.
Transformer Protection”, IEEE Trans. On Power Delivery,
[8] H.J. Zimmermann, Fuzzy sets, Decision Making, and Expert
Vol. 14, No.4, Oct. 99, pp.1351-1358.
Systems, Kluwer Academic Publisher, Boston 1987.
[26] “Alternative Electromagnetic Transient Program (ATP) ”.
[9] P.K.Dash, S. Mishra, M.M.A. Salama, and A.C. Liew ”
Classification of Power System Disturbances Using a Fuzzy
Expert System and a Fourier Linear Combiner ”, IEEE VI. Appendix
Transactions on Power Delivery, Vol. 15, No. 2, pp. 472-477, 1. Source parameters
April 2000. Z0 =.52 + j 6.65 Ω , Z1 = 0.32 + j 3.75 Ω
[10] H. Wang, and W.W. Keerthipala, “ Fuzzy-Neuro Approach 2. Transformer parameters
to Fault Classification for Transmission Line Protection”, Three phase 100.0 MVA, 50Hz, 330/33kV, ∆/Y connected
IEEE Transactions on Power Delivery, Vol. 13, No. 4, pp. windings with earthed neutral with the following parameters
1093-1102, October 1998. ∆/Y windings Wp exc.= 70. kW Zp.SC. = 30%
[11] P.K.Dash, A.K. Pradhan, and G. Panda, ” A Novel Fuzzy Ip exc. = 0.24% WZ. exc.= 6453 kW ZZ.SC. = 28%
Neural Network Based Distance Relaying Scheme ”, IEEE IZ. exc. = 56.25% Wp.SC = 312 kW
Transactions on Power Delivery, Vol. 15, No. 3, pp. 902-907, p, z : stand for positive and zero sequences respectively
July 2000. W: stands for losses, exc.: stands for excitation
[12] A. Wiszniewski, and B. Kasztenny, “ A Multi-Criteria 3. T.L. parameters
Differential Transformer Relay Based on Fuzzy Logic”, Two cascaded π-sections each with the following parameters:
IEEE Transactions on Power Delivery, Vol. 10, No. 4, pp. Z0 = 1.16 + j 13.3Ω , Z1= .665 + j 7.5 Ω
1786-1792, October 1995.
[13] W.H. Chen, C.W. Liu, and M.S. Tsai “ On-Line Fault
diagnosis of Distribution Substations Using Hybrid Cause- Biography
Effect Network and Fuzzy Rule-Based Method ”, IEEE
Transactions on Power Delivery, Vol. 15, No. 2, pp. 710-717,
Omar Youssef (SM’03) was
April 2000. born in Cairo, Egypt in 1945. He
[14] B. Kasztenny, E. Rosolowski, M.M. Saha, and B.Hillstrom, “ received the B.Sc., M.Sc., and
A Self-Organising Fuzzy Logic Based Protective Relay – An Ph.D. in Electrical Engineering
from University of Cairo, Faculty
Application to Power Transformer Protection”, IEEE
of Engineering in 1966, 1976, and
Transactions on Power Delivery, Vol. 12, No. 3, pp. 1119-
1979 respectively. From 1966 he
1127, July 1997.
has undertaken lecturing or
[15] H.J. Lee, D.Y. Park, B.S. Ahn, Y.M. Park, J.K. Prk, and S.S.
consulting assignments in Libya,
Venkata ” A Fuzzy Expert System for the Integrated Fault
Nigeria, Saudi Arabia, Iraq, Qatar.
Diagnosis”, IEEE Transactions on Power Delivery, Vol. 15,
No. 2, pp. 833-838, April 2000. On 1999 he has been invited as a Visiting Research Fellow at
[16] B. Kasztenny, E. Rosolowski, J.Izykowski, M.M. Saha, and University of Bath, U.K. He is currently the Deputy Dean to
B.Hillstrom, “ Fuzzy Logic Controller for On-Load Graduate Studies and Research, Faculty of Industrial Education,
Transformer Tap Changer”, IEEE Transactions on Power University of Suez Canal, Suez, Egypt.
Delivery, Vol. 13, No. 1, pp. 164-170, January 1998.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:29:03 UTC from IEEE Xplore. Restrictions apply.

You might also like