You are on page 1of 9

Applications of Fuzzy Inference Mechanisms

to Power System Relaying.


OMAR A.S.YOUSSEF
Ph.D., Senior Member, IEEE
Faculty of Industrial Education
Suez Canal University, Suez , Egypt.

presented before. The three line currents were


Abstract – Most transmission line protective schemes are based
on deterministic computations on a well defined model of the utilised in the multi-criteria algorithm using
system to be protected. This results in difficulty because of the Mamdani's fuzzy inference mechanism (FIM) for the
complexity of the system model, the lack of knowledge of its decision making part of the scheme.
parameters, the great number of information to be processed, and
The key benefit of fuzzy logic is that its knowledge
the difficulty in taking into consideration any system variation as
the rules are fixed. The application of fuzzy logic for exploring representation is explicit, using simple “IF-THEN”
complex, non-linear systems, diagnosis systems and other expert relations. Fuzzy logic systems are simpler, and
systems, particularly when there is no simple mathematical model faster. They reduce the design development cycle,
to be performed. provides a very powerful and attractive solution simplify design complexity, improve control
to classification problems. In this paper, a feasibility study on
the application of different fuzzy reasoning mechanisms to power performance, simplify implementation and reduce
system relaying algorithms is conducted. Those mechanisms are hardware costs. In addition, fuzzy rules are more
namely, Mamdani’s mechanism, Larsen's mechanism, Takagi- expressive than crisp values. On the other hand,
Sugeno's mechanism, and Tsukamoto mechanism. A comparative
their main disadvantages are confined to: (a) Non-
analysis on the application of these fuzzy inference mechanisms
to a novel fault detection and phase selection technique on EHV linear MFs and high MF terms can aggravate the
transmission lines is reported. The proposed scheme utilises only computational complexity. (b) The definition of the
the phase angle between two line currents for the decision MFs and the definition of fuzzy rules are difficult
making part of the scheme. A sample three-phase power system procedure, based on the designer’s knowledge and
was simulated using the EMTP software. An online wavelet-
based pre-processor stage [1-4] is used with data window of 10 should be done with great care, i.e. MFs should
samples (based on 4.5 kHz sampling rate and 50 Hz power reflect different event conditions. (c) The number of
frequency). The performance of the proposed model was rules should be enough, noting that it is a power
extensively tested in each case of fuzzy inference mechanism
function of the number of MF linguistic terms and
using the MATLAB software. The advantages and disadvantages
of each mechanism are reported and compared together. Some of number of inputs. (d) A change in the MF(rule) can
the test results are included in this paper. require a change in rule (MF). (e) Multi-parameter
optimization problem, i.e. MF affects the rules and
consequently the inference method. (f) In particular;
Index Terms – Fuzzy control - fuzzy sets – decision making -
fuzzy logic - knowledge-based systems - fault classification - with FL, there is difficulty in choosing the correct
phase selection. fuzzy rules, especially for complex systems. (g) FL
does not work well in unexpected circumstances. (h)
I. INTRODUCTION There is no systematic approach of tuning the MFs,
sometimes laborious or time consuming. (i) FL is
Single-pole tripping and consequent autoreclosure non-adaptive in nature.
has been recognized as an effective means of This paper proposes a new 1-input, 1-output fuzzy-
improving system performance. Traditional phase based phase selection technique that utilises only
selection schemes [5-9] suffer from some drawbacks the phase angle between two line currents in order
due to complexity of the system models, lack of to determine the faulted phase and to initiate
knowledge of their parameters, effect of remote-end single-phase autoreclosure. A zero-sequence
infeed, fault resistances, mutual coupling from current detector is used to trigger the FIM. The
adjacent parallel lines, etc. Recently, the applications paper is organised as follows: In Section II, some
of the fuzzy set theory is introduced to solve basic concepts of different FIMs are introduced.
uncertainty problems [10-11]. Some of the fuzzy In Section III, the relaying algorithm is briefly
logic applications in power system protection are described. In Section IV, MFs and the response of
included in [12-19]. A proposed fuzzy-logic-based FIMs are derived. In Section V, the selection of the
multi-criteria approach to real-time fault detection FIM is performed. In Section VI, the simulation
and classification [20] and phase selection [21] results are presented. Results from each inference
algorithms in power transmission systems have been mechanism are compared. Conclusions and
suggestions for future work are given in Section
VII.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
II. FUZZY INFERENCE
MECHANISMS (FIMs)
A fuzzy inference system (FIS) uses a collection of
fuzzy membership functions (MFs)and rules, instead
of Boolean logic, to reason about data. Basically, it
constitutes three parts: a rule base containing a
selection of fuzzy rules, a database defining the MF
(fuzzy values) used in the fuzzy rules, and a
reasoning mechanism. This is shown in Fig.1.

Rule Base Data Base


Input ( MFs) Output

Fuzzy
Reasoning

Fig.2 Mamdani's max-min and Larsen's max-product


inference mechanisms.
Fig. 1 Model of an FIS.

3. Takagi-Sugeno's (TS) FIS: It uses crisp functions


A fuzzy logic system is inflected in the following
in the consequents.
basic elements: fuzzification, inference
First order system: output1,2,3 = a1,2,3*x+b1,2,3
(implication), composition (aggregation) and de-
Zero order system: output1,2,3 = b1,2,3
fuzzification. Sometimes the term inference method
3
is used to mean the combination inference and
composition, e.g. in max-min inference (Mamdani's
∑ output i * MFai
FIS) and sum-product inference (Larsen's FIS) They Overall output = i =1
3
are the combination of max composition and min ∑ MFai
inference, or sum composition and product i =1
inference, respectively. Fig. 2 shows these four 4. Tsukamoto's FIS: A modification of TS method in
steps for both mechanisms. The most commonly that the MFs are assumed to be monotonous.
used inference mechanisms are either produce fuzzy Activation levels are computed as for TS method.
sets for defuzzification, e.g. Mamdani's mechanism , Interpolation of the antecedent and consequent MFs
or directly a control value, e.g. Takagi-Sugeno's using an ‘spline’ function
mechanism. Those are briefly described as follows: Output1 = INTERPOLATE(MFa1 , MFc1) ,
1. Mamdani's (max-min) FIS: A linguistic model Output2 = INTERPOLATE(MFa2 , MFc2) ,
that describes the system by means of Output3 = INTERPOLATE(MFa3 , MFc3)
linguistic if-then rules with fuzzy proposition in the And the overall output is calculated as in ( TS ) FIS.
antecedent as well as in the consequent. Implication
is modeled by means of minimum operator, and the 5. Singleton FIS: A special case of the linguistic
resulting output MFs are combined using maximum fuzzy model is obtained when the consequent fuzzy
operator. For one input, three linguistic variables, sets are singleton fuzzy sets. A simplified inference/
the process can be explained as follows: defuzzification method is usually used with this
Effective MFs: MFE1 =min(MFa1,MFc1) , model. More details on the five mechanisms are
MFE2=min(MFa2, MFc2) , MFE3 = min(MFa3 , MFc3) given in the Appendix
Overall MF = max(MFE1 , MFE2 , MFE3 )
Consequently, this can be defuzzified using one of III. ALGORITHM
the defuzzification techniques.
A. Sample Power System: A simplified three phase
2. Larsen's ( max-product ) FIS: Implication is power system, shown in Fig. 3, was chosen for the
modeled using the product operator, while purpose of generating line currents under normal and
propositions are defined by the maximum operator. fault conditions. Transmission line impedances are
Effective MFs: MFE1 =MFa1*MFc1 , listed in Table (1).
MFE2=MFa2* MFc2 , MFE3 = MFa3 * MFc3 Sequence R - Ω/km X - Ω/km Susceptance
Overall MF = max(MFE1 , MFE2 , MFE3 ) Zero 0.3083340 0.986269 2.75263x10 – 6
Which can in turn be defuzzified. Positive 0.0172664 0.303638 3.88293x10 – 6
Table 1. Transmission line Sequence Impedances.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
P ha s e a n gle C iab (Ia ,Ib ) P ha s e a n gle C ibc (Ib,Ic )
1 50 1 50

1 00 1 00
while source parameters are: D eg s D eg s .

ZS =4.0+j 40.0 Ω , Z0/Z1 = 1 50 50

ZR = 0.4+j 4.0 Ω , Z0/Z1 = 1 0 0

SC P ha s e a n gle C ic a (Ic ,Ia ) Z ero S eq . ind e x I0 /I1


2 50 8
GS GR 2 00
6
300 km 1 50
D eg s . 4
1 00
2
50

Fig.3 Model power system. 0 0


0 5 00 1 0 00 0 500 1 0 00
S am p le

Data base of line currents is built up for various types Fig. 4 Phase angles Ciab , Cibc , Cica and zero sequence index
of faults at different locations and different system |I0| / |I1| during four different AG faults.
parameters by using EMTP software [22-23]. A 2- P h a se a ngle C ia b (Ia,Ib ) P ha se a ngle C ib c (Ib ,Ic )
samples data window is used to minimise the effect 2 50 20 0

of dc offset components present in the signal 2 00


15 0

waveform. Based on mother wavelet db8 and 10- D e g .1 50 D e g s.


10 0
samples sliding data window wavelet-based pre- 1 00
50
50
processor is used to remove high frequency
0 0
harmonic and non-harmonic components [1-4].
P h as e an gle C ica (Ic,Ia) Ze ro S e q . in d ex I0/I1
3 00 3
B. Determination of Sequence Components of Line
Currents I0 and I1[20-21] 2 00 2
D e g s.
The symmetrical components are derived by means
of a single linear combination of samples of line 1 00 1

currents. At sampling instant (k), the sequence


0 0
components of line currents are calculated as: 0 500 1 0 00 0 50 0 1 0 00
S am p le
i0(k) = ( ia(k) + ib(k) + ic(k) ) /3
i1(k) = ( ia(k) - ib(k+15) - ic(k-15) ) /3 Fig. 5 Phase angles Ciab , Cibc , Cica and zero sequence index
|I0| / |I1| during five different BG faults.
i2(k) = ( ia(k) - ib(k-15) - ic(k+15) ) /3
Phase angle Ciab (Ia,Ib) Phase angle Cibc (Ib,Ic)
150 300
C. Zero-sequence current index[20-21]
A zero sequence current base indicator can be 100 200
Degs.

Degs.

expressed as:
50 100
∧2
I o( k ) [(i 0( k ) ) + C 32 (i 0(k +1)
2 2
− i 0 (k −1) ) ]
=
0 0

∧2 [(i1( k ) ) 2 + C 32 (i1( k +1) − i1( k −1) ) 2 ] Phase angle Cica (Ic,Ia) Zero Seq. index I0/I1
I 1( k ) 200 10

8
150
1
where , C 3 = ( ) 6
Degs.

2πfh 100
4

D. Phase Angle Criterion[20-21] 50


2

Phase angles of the three line currents Ia,b,c at 0


0 500 1000
0
0 500 1000
sampling interval (k) are calculated as follows : Sample

i a,b,c(k +1) − i a,b,c(k -1) Fig. 6 Phase angles Ciab , Cibc , Cica and zero sequence index
ϕ ia ,b,c( k ) = C1 |I0| / |I1| during four different CG faults.
C 2 (i a,b,c(k +1) + i a,b,c(k -1) ) + i a,b,c(k)
1 + 2 cos θ s2 The phase angle criteria is explained in Table 2 for
where, C1 = , C 2 = cos θ s different fault types, while the rule base for the
2 sin θ s fuzzy decision support system is listed in Table 3.
For sampling rate 4.5 kHz, the sampling angle θs
=4o , C1=21.42 and C2= 0.998.Fuzzy criteria Ciab , AG BG CG
Cibc , and Cica are defined as follows: Ciab L H M
Ciab(k) = ϕia(k) - ϕib(k) , Cibc(k) = ϕib(k) - ϕic(k) Cibc M L H
Cica H M L
Cica(k) = ϕic(k) - ϕia(k) Table 2 Phase angle criterion.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
Fault Rule consequent MFs along with the corresponding FIM
AG If Ciab is LOW THEN trip output is AG. response are shown in Figs.8-15. In each figure four
BG If Ciab is HIGH THEN trip output is BG.
CG If Ciab is MEDIUM THEN trip output is CG.
cases are considered for antecedent and consequent
Table 3 Fuzzy rules. MFs. In Fig. 8, each row represents three antecedent
MFs , three consequent MFs and the resulting
Degree of MF response when using Mamdani’s FIM. The first row
Antecedent MFs
µ is denoted as case 1, second row as case 2, and so
1.0 on. In Fig. 9, the four Mamdani’s FIM responses are
MF1 MF2 MF3 plotted together on the same graph. Fig. 10 indicates
AG CG BG Larsen’s response for the same MFs shown in Fig.8.
Fig.12 represents Sugeno’s FIM responses for the
L M H specified MFs shown in Fig. 11. Figs.13-14 show
Tsukamoto’s FIMs. In Figs.15 and 16, the responses
Phase angle, of the four FIMs for two different sets of antecedent
0 60 120 180
Degs. and consequent are plotted. Fig.17 illustrates the
Degree of MF resulting defuzzified output for three BG, three CG,
µ Consequent MFs and three AG different fault conditions.
1.0 Antecedent MF Consequent MF Tsukamoto FIS Response

Degreeof MF Degree of MF Degree of MF Degree of MF

Degreeof MF Degree of MF Degree of MF Degree of MF


3

Defuz output
1 1

0.5 0.5 2

MF1 MF2 MF3 0 0 1


3

Defuz output
AG CG BG 1 1

0.5 0.5 2

Fault type 0 0 1
3
index

Defuz output
1 1

0.5 0.5 2

1 2 3 0 0 1
4

Defuz output
1 1
Fig. 7 Antecedent and consequent MFs of the FIS. 0.5 0.5 2

0 0 0
0 100 200 0 2 4 0 100 200
Phase Angle,degs. Defuz. Output Phase Angle,degs.
IV. MFs DEFINITIONS
Fig. 8 MFs & corresponding response of Mamdani FIM.
The performance of the power system during Mamdani FIS Response
various types of single-phase-to-ground faults at 3.5

[1]:Case 1
different locations, inception angles and prefault 3 [2]:Case 2

conditions of the system is illustrated in Figs 4-6. [3]:Case 3


[4]:Case 4
Definition of MFs of the phase angle criterion Ciab is 2.5
DefuzzifiedOutput

based on the data obtained from the simulation 2

results, as illustrated in Fig.7, from which it is clear [3]

1.5 [4]
that the disturbance waveforms classification is [1]
[2]

based on three linguistic variables of the phase 1

angle criterion Ciab designated as H(HIGH),


0.5
M(MEDIUM), and L (LOW). The zero-sequence 0 50 100 150
Phase Angle,degs.
200 250

current index ( |I0| / |I1|) is used to initiate the fuzzy


inference mechanism. An extensive simulation Fig. 9 Response of Mamdani FIM .
procedure have been carried out in order to
appropriately define MFs and specify the fuzzy 3.5
Larsen FIS Response

rules. Extreme healthy and faulty power system [1]:Case 1


[2]:Case 2
conditions have been considered in the problem 3
[3]:Case 3
[4]:Case 4
domain, thus avoiding any unexpected 2.5
Defuzzified Output

circumstances. Linear MFs have been selected


(trapezoidal- triangular) thus reducing the 2
[1]

computational burden to minimal. Only single input 1.5


[3]
( Ciab ) with three linguistic variables (LOW- [4]
[2]
MEDIUM –HIGH ), and one output with three 1

linguistic variables (AG –BG – CG ) are defined for 0.5


the fuzzy inference mechanism. A combination of 0 50 100 150
Phase Angle,degs.
200 250

trapezoidal and triangular MFs are considered in


order to get the most appropriate response of the Fig. 10 Response of Larsen FIM .
fuzzy inference mechanism (FIM). Antecedent and

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
Antecedent MF Sugeno FIS Response
egreeof MF Degreeof MF Degreeof MF Degreeof MF

Defuz output
1 3

0.5 2

1
computationally efficient representation than a
0
Mamdani system, the Sugeno system is preferred for

efuz output
1 3

0.5 2 constructing fuzzy models. It works well with linear


techniques. Also, it is suited for modeling nonlinear

D
0 1

Defuz output
1 3

0.5 2 systems by interpolating multiple linear models.


0 1 In Fig. 17, the left column represents BG, CG, and
AG faults under the following conditions: 10. Ω

Defuz output
1 3

0.5 2

1 fault resistance at 30 km from the local end of the


D

0
0 50 100 150 200 250 0 50 100 150 200 250
Phase Angle,degs. Phase Angle,degs.

1
Antecedent MFs

Degree of MF
Fig. 11 MFs & corresponding response of Sugeno FIM. 0.8
0.6
LOW MEDIUM HIGH
Sugeno FIS Response 0.4
0.2
3 [1]:Case 1
0
[2]:Case 2 0 50 100 150 200 250
[3]:Case 3 Phase Angle, Degrees
[4]:Case 4
1
2.5
[1] Cons. MFs
0.8
DefuzzifiedOutput

Degree of MF
0.6
AG Fault CG Fault BG Fault
2 0.4

0.2

0
1.5 0 0.5 1 1.5 2 2.5 3 3.5 4
[2]
Defuzzified Output

[3] [4]
1 D e fu z z if ie d R e s p o n s e o f F u z z y I n f e r e n c e S y s te m
0 50 100 150 200 250 O u tp u t Su geno
Phase Angle,degs. 3

Fig.12 Response of Sugeno FIM.


2 .5
Antecedent MF Consequent MF Tsukamoto FIS Response
Degree of MF

Defuz output

1 1 3 M am dani
0.5 L M H 0.5 AG CG BG 2
2
0 0 1
1 1 3 T su k a m o to
0.5 0.5 2 1 .5

0 0 1
Degree of MF

Defuz output

1 1 3 L a rsen
1
0.5 0.5 2
0 0 1 0 50 100 150 200 250
1 1 3 P h a se A n g le , D e g r ee s
0.5 0.5 2

0 0 1 Fig. 15Antecedent and Consequent MFs and the


Degree of MF

Defuz output

1 1 3 corresponding responses of the four FIMs.


0.5 0.5 2
0 0 1
0 100 200 0 2 4 0 100 200
Phase Angle,degs. Defuz. output Phase Angle,degs. 1 Antecedent MFs
Degree of MF

0.8
Fig. 13 MFs & corresponding response of Tsukamoto FIM. 0.6
LOW MEDIUM HIGH
0.4
Tsukamoto FIS Response
3 0.2
Triangular parameters of conseq. MF:CG Fault
2.8
0
[1]:[.99 1.99 2.99] 0 50 100 150 200 250
2.6
[2]:[1.09 2.09 3.09] Phase Angle, Degrees
[3]:[1.1 2.1 3.1]
[4]:[1. 1.99 3]
1
2.4
[5]:[1. 2 2.5]
Cons. MFs
Defuzzified Output

0.8
Degree of MF

2.2 [4]
[2]
[3] 0.6 AG Fault CG Fault BG Fault
2 [1]
[5] 0.4
1.8
0.2
1.6
0
1.4 0 0.5 1 1.5 2 2.5 3 3.5 4
Defuzzified Output
1.2

1
0 50 100 150 200 250 D e fu z z i f i e d R e s p o n s e o f F u z z y I n fere n c e
4
Phase Angle,degs. u tp u t S y ste m
Fig. 14 Response of Tsukamoto FIS. 3.5 T su k a m o to
3
Sugeno
2.5
M am dani
IV. SELECTION OF FUZZY 2

INFERENCE MECHANISM 1.5

1 L a rs e n

Takagi and Sugeno's model can express a highly 0.5

non-linear functional relation using small number of 0


0 50 100 150 200 250

P ha se A ng le, D e gree s
fuzzy rules. However, the complexity of its
identification procedures made it difficult to be used. Fig. 16 Antecedent and Consequent MFs and the
On the other hand, because it is a more compact and corresponding responses of the four FIMs.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
Approx. of Ia,b,c Phase angles bet. Ia,Ib,Ic
500 250
Iaw Cica
line at 0.035 ms fault inception angle, while the Ibw
200

middle column represents the situation (50 Ω, 150


150

Degs.
Cibc

A
0
100
km, 0.04 ms ), and the right column represents the Icw
50
Ciab

situation ( 0.1 Ω, 290km , 0.056 ms). It is clear from -500 0

Figs.15-17 that: 6
I0/I1
4
Defuz. Output

Faulty Phase Index 1-3


1. Trapezoidal functions for linguistic variable of the 3
4
antecedent MFs (LOW and HIGH), and triangular 2

MF for (MEDIUM) lead to a better FIM response. 2


1

2. Application of Tsukamoto FIM for the power 0 0


0 100 200 300 400 100 200 300 400
system relaying algorithm under consideration is Sample Sample

inadequate. Fig.18 AG fault - Rf =10Ω at 30 km using Sugeno FIS.


3. Application of Mamdani’s, Larsen’s and Sugeno’s 500
Approx. of Ia,b,c
250
Phase angles bet. Ia,Ib,Ic

FIM lead to the same FIM response. Ibw


200 Ciab
Iaw
4. Regarding the computational burden, Sugeno’s 150

Degs.
Cica

A
0
100
FIM proves to be better. Icw
50
Cibc

5. Mamdani’s and Larsen’s FIM have almost the -500 0

same FIM response and computational burden. I0/I1 Defuz. Output


3 4

Faulty PhaseIndex 1-3


BG Fault BG Fault BG Fault 3
2
3.1 3.1 3.1
Defuz. Output

S S 2
3 3 3 1
M L M L 1

2.9 2.9 T L 2.9 T 0 0


T M S 100 200 300 400 100 200 300 400
2.8 2.8 2.8 Sample Sample

CG Fault CG Fault CG Fault


2.2 2.2 2.2 Fig. 19 BG fault - Rf =10Ω at 30 km using Sugeno FIS.
Defuz. Output

T T T
Approx. of Ia,b,c Ratio of Phase Angles
S S S 400 6
2 2 2 Ibw
M M L M L 200
L Iaw 4
Ciab
1.8 1.8 1.8
A

0
AG Fault AG Fault AG Fault 2
-200 Icw
1.2 1.2 1.2 Cica
Defuz. Output

T T T Cibc
-400 0
1.1 1.1 1.1
S S S Phase angles bet. Ia,Ib,Ic output
1 M 1 M 1 M 250 4
L L L Faulty PhaseIndex 1-3
0.9 0.9 0.9 200
3
0 500 1000 0 500 1000 0 500 1000 150
Degs.

Sample Sample Sample 2


100
Fig. 17 Defuzzified outputs of the four FIMs 50
1

0 0
100 200 300 400 0 100 200 300 400
Sample
V. SIMULATION AND TESTING Fig. 20 BG fault - Rf =20Ω at 100 km using Mamdani FIS.
Approx. of Ia,b,c Phase angles bet. Ia,Ib,Ic
Fault data generate from the EMTP software under 200
Iaw
200
Ciab

different system conditions within the problem 100 150


Cica
Degs.
A

0 100
domain are used as input to the MATLAB software Cibc
-100 50
for testing the algorithm in identifying the faulty -200
Icw
Ibw
0
phase. Sampling rate of 4.5 kHz based on a 50 Hz I0/I1 Defuz. Output
power frequency is used. The phase angle Ciab 2.5 4
Faulty PhaseIndex 1-3

2
between the line currents Ia and Ib together with the 1.5
3

normalised zero sequence current index |I0|/|I1| are 1


2

1
considered as input vector for the network. Power 0.5

0 0
system conditions were varied by varying shunt fault 0 100 200
Sample
300 400 100 200
Sample
300 400

resistance, fault location, fault inception angles and Fig. 21 BG fault - Rf =10Ω at 290 km using Tsukamoto FIS.
prefault conditions. In Figs.18-23, typical wavelet Approx. of Ia,b,c Phase angles bet. Ia,Ib,Ic
400 200
approximation components of line currents Ia,b,c are 200
Icw
150
Cibc
Ibw Iaw Ciab
shown for AG, BG and CG faults for different fuzzy
egs.
A

0 100
D

Cica
inference mechanisms along with phase angles ( -200 50

Ciab , Cibc , Cica ) , zero-sequence current index -400 0

(|I0|/|I1|), and the output of the FIS. Test results show 40


I0/I1
4
Defuz. Output
FaultyPhaseIndex1-3

that the proposed fuzzy-based phase selector is 30 3

effective in determining the faulty phase under 20 2

10 1
different fault locations, fault occurring times, 0 0
0 100 200 300 400 100 200 300 400
presence of fault resistance and variations in source Sample Sample

impedances. Fig. 22 CG fault - Rf =20Ω at 100 km using Larsen FIS.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
Approx. of Ia,b,c Phase angles bet. Ia,Ib,Ic
200 300
Icw
Iaw
100
200
autoreclosure for EHV transmission systems”, IEE Proc-
Cibc

egs.
Gener. Transm. Distrib., Vol. 141, No.2, March, 1994.
A

D
Ciab
-100
100
Cica
[9] Bo,Z.Q., Aggarwal, R.K., Johns, A.T., Li, H.Y., Song, Y.H., “
Ibw
-200 0 A New approach to phase selection using fault generated
I0/I1 Defuz. Output
frequency noise and neural networks”, ”, IEEE Trans. On
8 4 Power Delivery, Vol2, No.1, Jan. 1997, pp.106-115.

FaultyPhaseIndex1-3
6 3 [10] Zadeh L. A.,” Fuzzy sets”, Information Control, No. 8,
4 2 1965, pp. 338-353.
2 1 [11] H.J. Zimmermann, Fuzzy sets, Decision Making, and Expert
0 0 Systems, Kluwer Academic Publisher, Boston 1987.
0 100 200 300 400 100 200 300 400
Sample Sample [12] P.K.Dash, S. Mishra, M.M.A. Salama, and A.C. Liew ”
Fig. 23 CG fault - Rf =40Ω at 240 km using Sugeno FIS. Classification of Power System Disturbances Using a Fuzzy
Expert System and a Fourier Linear Combiner ”, IEEE
Transactions on Power Delivery, Vol. 15, No. 2, pp. 472-477,
I. CONCLUSION April 2000.
[13] A. Wiszniewski, and B. Kasztenny, “ A Multi-Criteria
The paper presented an online application of a 1- Differential Transformer Relay Based on Fuzzy Logic”,
IEEE Transactions on Power Delivery, Vol. 10, No. 4, pp.
input, 1-output fuzzy-logic-based technique to 1786-1792, October 1995.
single-phase auto-reclosing of transmission lines. A [14] W.H. Chen, C.W. Liu, and M.S. Tsai “ On-Line Fault
simple fuzzy procedure to solve a problem that diagnosis of Distribution Substations Using Hybrid Cause-
requires more complex algorithms when approached Effect Network and Fuzzy Rule-Based Method ”, IEEE
Transactions on Power Delivery, Vol. 15, No. 2, pp. 710-717,
in a deterministic way is used. A comparative study April 2000.
on different FIMs in order to select the most [15] B. Kasztenny, E. Rosolowski, M.M. Saha, and B.Hillstrom, “
appropriate response to the relaying algorithm was A Self-Organizing Fuzzy Logic Based Protective Relay – An
found to be Sugeno’s response with only 3-rules Application to Power Transformer Protection”, IEEE
Transactions on Power Delivery, Vol. 12, No. 3, pp. 1119-
with linear MFs having three linguistic variables for 1127, July 1997.
both antecedent and consequent. The technique is [16] H.J. Lee, D.Y. Park, B.S. Ahn, Y.M. Park, J.K. Prk, and S.S.
proved to be fast due to the low computational Venkata ” A Fuzzy Expert System for the Integrated Fault
burden of the Sugeno’s mechanism, accurate and Diagnosis”, IEEE Transactions on Power Delivery, Vol. 15,
No. 2, pp. 833-838, April 2000.
robust and would perform accurately for various [17] B. Kasztenny, E. Rosolowski, J.Izykowski, M.M. Saha, and
system conditions. However, feature extraction B.Hillstrom, “ Fuzzy Logic Controller for On-Load
have to be enhanced in order to improve MFs Transformer Tap Changer”, IEEE Transactions on Power
definitions Also, in determining optimal structure of Delivery, Vol. 13, No. 1, pp. 164-170, January 1998.
[18] H.T. Yang, and C.C. Liao, “ Adaptive Fuzzy Diagnosis for
MFs, methods to find out a solution near a global dissolved Gas Analysis of Power Transformers”, IEEE
minimum is necessary. Besides, due to the available Transactions on Power Delivery, Vol. 14, No. 4, pp. 1342-
dedicated processors that allow real-time execution 1350, October 1999.
of fuzzy procedure, e.g. rule chip TG005MC, [19] A. Ferrero, S. Sangiovanni, and E. Zappitelli, “ A Fuzzy –Set
Approach to Fault-Type Identification in Digital Relaying”,
defuzzifier chip TB005PL,etc. fuzzy-logic-based IEEE Transactions on Power Delivery, Vol. 10, No. 1, pp.
relaying algorithms should change the way that 169-175, January 1995.
hardware is currently defined. [20] Omar A.S.Youssef, “Combined Fuzzy-Logic-Wavelet-Based
Fault Classification Technique for Power System Relaying ”,
paper #TPWRD-0047, accepted for publication in the IEEE
VII. REFERENCES Transactions on Power delivery, March 2003.
[21] Omar A.S.Youssef, “A Novel Fuzzy-Logic-Based Phase
[1] Omar A.S.Youssef, “ Online Applications of Wavelet Selection Technique for Power System Relaying ”, Electric
Transforms to Power Systems Relaying ”, IEEE Transactions Power Systems Research Journal, paper # EPSR 1909, under
on Power delivery, Vol. 18, No. 4, Oct. 2003, pp.1158-1165. press, Oct. 2003.
[2] Omar A.S.Youssef, "New Algorithm to Phase Selection Based [22] “Alternative Electromagnetic Transient Program (ATP ) ”.
on Wavelet Transforms", IEEE Transactions on Power [23] H.W.Dommel, “ Electromagnetic Transient Program ”,
delivery, Vol. 17, No. 4, Oct. 2002, pp.908-914. August 1986, Boneville Power Administration, USA..
[3] Omar A.S.Youssef, " A Wavelet-Based Technique for Omar Youssef (SM’2003)
Discrimination between Faults and Magnetising Inrush was born in Cairo, Egypt in 1945.
Currents in Transformers", IEEE Transactions on Power He received the B.Sc., M.Sc., and
delivery, Vol. 18, No. 1, Jan. 2003, pp.170-176. Ph.D. in Electrical Engineering from
[4] Omar A.S.Youssef, “ Fault classification based on wavelet University of Cairo, Faculty of
transforms”, paper # 01TD069, IEEE, T&D Conference , 28 Engineering in 1966, 1976, and 1979
Oct. - 2 Nov. 2001, Atlanta, Georgia. respectively. From 1966 he has
[5] IEEE committee report,” Single Phase tripping and Auto undertaken lecturing or consulting
Reclosing of Transmission Lines”, IEEE Transactions on assignments in Libya, Nigeria, Saudi
Power Delivery, Vol. 7, No. 1, pp. 182-191, January 1992 Arabia, Iraq, Qatar. On 1999 he has
[6] IEEE Working Group, Single-pole switching for stability and been invited as a Visiting Research
reliability, IEEE, Vol. PWRS-1, No.2, 1986, pp.25-36. Fellow at University of Bath, U.K. He is currently the Deputy
[7] GEC measurements, “ Protective relaying application guide”, Dean to Graduate Studies and Research, Faculty of Industrial
GEC Alsthom Measurement Limited. Education, University of Suez Canal, Suez, Egypt. Email:
[8] Aggarwal, R.K., Johns, A.T., Song, Y.H., Dunn, R.W., Fitton, oyoussef@link.net
D.S., “Neural network techniques based adaptive single-pole

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
VIII. Appendix: Fuzzy Inference Mechanisms
Architecture: For two-input , one-output FIS, the ith fuzzy rule 1.Input to the system is (x0 , y0).
from the rule-base is: 2. Fuzzified input is (x0 , y0).
Ri : if x is Ai and y is Bi then z is Ci 3. Firing strength of the ith rule is Ai(x0) ∧ Bi(y0).
Fact: x is x0 and y is y0 4. The ith individual rule output is Ci’(w) = [ Ai(x0) ∧ Bi(y0) ] →
Consequence: z0 Ci(w).
As an example, the process of reasoning in Mamdani FIS is as 5. Overall system output (union of the individual rule outputs)
follows: C= C1’ ∪ ….. ∪ Cn’.

Mamdani Larsen Takagi-Sugeno Tsukamoto Singleton


Fuzzy implication Minimum operator Product operator ----------------- ----------------- -----------------
method
Aggregation Maximum operator Maximum operator ----------------- ----------------- -----------------
(composition ) of
propositions
Firing(activation) α1=A1(x0) ∧ B1(y0) α1=A1(x0) ∧ B1(y0) α1=A1(x0) ∧ B1(y0) α1=A1(x0) ∧ B1(y0) α1=A1(x0) ∧ B1(y0)
level of the rules, α2=A2(x0) ∧ B2(y0) α2=A2(x0) ∧ B2(y0) α2=A2(x0) ∧ B2(y0) α2=A2(x0) ∧ B2(y0) α2=A2(x0) ∧ B2(y0)
i= 1,2
Individual rule C’1(w)= α1∧C1(w) C’1(w)= α1 C1(w) First order TS: α1=C1(z1) C1 , C2
outputs C’2(w)= α2∧C2(w) C’2(w)= α2 C2(w) Z1* =a1 x0 + b1 y0 α2=C2(z1)
( min. operation ) (prod. Operation ) Z2* =a2 x0 + b2 y0 (crisp)
Overall system C(w)=C’1(w)∨ C’2(w) C(w)=
output C(w)= (α1∧C1(w) ) ∨ α1C1(w) ∨ α2C2(w) ----------------- ----------------- -----------------
(α2∧C2(w) )
Overall Using one of the Using one of the Z0=(α1Z1*+α2 Z2*) / z0=(α1z1*+α2 z2*) / z0=(α1C1+α2 C2) /
deterministic defuzzification defuzzification (α1 + α2 ) (α1 + α2 ) (α1 + α2 )
output methods methods (Crisp ) (Crisp– cog method) (Crisp– cog method)
Rule-base with n- C(w)= C(w)= z0= Z0= z0=
rules, αI : is the n ∨ i=n i=1 (αi Ci (w)) n n n n n n
firing level of the ∨ (αi ∧ Ci (w)) ∑ αizi* / ∑ αi ∑ αiZi* / ∑ αi ∑ αiCi* / ∑ αi
ith rule, i=1,2….,n i=1 i=1 i=1 i=1 i=1 i=1 i=1

Remarks Inference method with Differs from Linear functions as Differs from (TS) in Linguistic F. model
MFs as outputs in Mamdani’s method outputs in rules . that MFs are with singleton con-
rules only in the choice of 1.Zero-order model: monotonous. Values sequence F. sets and
the product operator Constant consequent for zi* are the fuzzy mean defuz-
for implication 2.First-order model: solutions to Zi(zi)= zification method
Linear consequent αi

min Mamdani Larsen Takagi-Sugeno Tsukamoto

A1 B1 C1

α1 α1
u v w z1 = a1 x + b1 y z1

A2 B2 C2

α2 α2
x0 u y0 v w z2 = a2 x + b2 y z2

=
∑ α
z
i i
=
∑ α z
i i

∑α ∑α
z z

i i

Defuzzification

Fig. A1 The five main Inference mechanisms.

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.
9

Authorized licensed use limited to: Durban University of Technology. Downloaded on April 28,2023 at 16:06:39 UTC from IEEE Xplore. Restrictions apply.

You might also like