Professional Documents
Culture Documents
Extensive Events Database Development Matlab To Fault Location Distribution
Extensive Events Database Development Matlab To Fault Location Distribution
II. DEVELOPMENT OF A FAULT LOCATION KNOWLEDGE BASED additional information which helps to solve the problem of
METHOD -KBM. fault location. Examples of additional information are:
The implementation of a KBM for fault location requires of Telephone calls from affected customers; classification of the
methodological planning, analysis of techniques to be used, as load in each node as commercial, residential, industrial; and
well as the availability of well conditioned information to train weather conditions, among others.
and validate the method. Information which allows characterizing the faulted system
is known as descriptors [14]. Some of them are:
A. Methodological stages - Characterization of voltage sags: The fundamental of
Figure 1 shows the required stages to develop a KBM to voltage registered at the substation has a variable magnitude
solve the problem of fault location. during the pre-fault, fault and post-fault intervals. The
Conditiom characterization of sags allows extracting descriptors which
LHISYSTEMhiru;
Infornia nt
n
can be associated with faults.
- Transient signal analysis: Frequency and duration of the
transient caused by fault are used as descriptors. These
values are related to the location of fault because these direct
relation to the electrical parameters of the faulted system.
- a and P coefficients analysis: These coefficients are obtained
by transformation of the time samples of the three phase
fault currents, using Clark-Concordia transform.
- Load current pattern analysis: Characterization of the
currents registered during faults helps to fault location. It
contains the performance of the protective devices which
operates to fault clearing.
Fig. 1. Methodological scheme to develop a KBM. III. MODELING THE DISTRIBUTION SYSTEM AND PROTECTIVE
The simulation stage allows acquire the measurements RELAYING
from the system under fault. Information from the measured Alternative Transient Program-ATP was selected for
system variables is obtained from waveforms of voltage and simulation purpose, because of it is widely accepted by
current, among other type of information described in literal researchers in electrical engineering and its tested
C. After that, a data base is developed to save this information performance on this type of studies.
as ordered input-output pairs. Inputs are characteristics known
in this paper as "descriptors", while the outputs correspond to Figure 2 shows a 25 kV system from SaskPower and Light,
in Saskatoon, Canada used in this paper as a test system [15].
the current fault location of the registered faults. In this system, five protective elements have been added as it
There are several algorithms which are capable to infer is shown: A main breaker (ANSI 52) with a inverse
knowledge rules from examples. To do that, normally it is overcurrent relay (ANSI 51), placed at the distribution
necessary of learning or training stage. In case of supervised substation; a recloser (R) located at the beginning of the line
learning, training is performed by presenting the ordered pairs between nodes 3 and 4; a fuse in the line between nodes 5 and
of input-output data set to the learning algorithm. 10 (Fl); another fuse in the line between nodes 6 and 9 (F2),
Finally there is a validation stage where the effectiveness is and the last fuse in the line between nodes 11 and 13 (F3). In
checked. Error is considered by comparison between the this application case, the protective system was configured in
KBM output and the real value of the fault location. the "fuse saving" scheme, as it is shown in Fig. 3 [16].
B. Commonly used techniques in the KBM implementation. 2 3 4 5 8 9 1o 11 12
/a a
/a
Normally KBM formulates the solution of a problem by
means of a classification algorithm that uses the data set 1 13
~~~~~~F
A. Distribution system modeling Figure 6 shows the operation and the logic of the MODEL
The power system implementation by using ATP is simple; used to simulate the recloser operation.
thanks to the graphical user interface ATPDraw [8]. Several Variables
Counters
power elements could be selected from a complete list, which Timers
includes generators, transmission lines, passive elements as
resistances inductances and capacitances, switches, among
others. However, there are not available protective devices,
and that is the reason of develop a code by using ATP-
MODELS. This allows creating new elements by a general
programming structure, supported by a set of simulation tools
used to help in representation and study of the variables [8].
SD
,% N
,
>, Delay Curves (Rec) Fig. 4. Working scheme of the fuse model.
-- _- Total clearing (fuse) MODEL: FUSE
'Z. Minimun melt (fuse) INPUT (MODEL 's Signals Input data)
Fast Curves (Rec) DATA (Inverse time curve. Modified by the end user)
Current VAR (Variables used to measure the RMS value of line currents)
Fig. 3. Typical time coordination curves for the distribution system feeder. (Time, timer and zero cross detection variables)
(Output variables)
B. Fuse modeling OUTPUT (Output variables)
zone, these elements do not need additional devices, as current IF IRMS>=IPK THEN (Over current detection)
(Calculation of the fuse operation times according to DIAL
transformers (CT's), or additional breakers. Fuse contains andA,B andp inverse time curve parameters)
(Timers starting)
itself both the relay and breaker functions [16]. Fig. 4 ENDIF
presents the operation scheme and the programming logic IF TEMP> Top THEN (Timer ending)
used to simulate fuse behavior. In Fig. 5 the programming ENDIF
(Fuse operation in next current zero crossing)
code using fuse MODELS is shown.
ENDIF
C. Recloser modeling ENDEXEC (End)
Recloser device contains itself the integration of two ANSI Fig. 5. Example of the fuse model code.
protective functions (50/51), a breaker function (52), and the Variables
Counters (C)
START
logic control. Reclosers are provided with two or more inverse Timers (T)
time operating curves, but in this case and to simulation
purpose, the protective module is settled with two inverse time
curves as shown in Fig. 3. In case of fault, recloser detects it
and operates in a determined number of times with the fast
curve and a predefined reclosing time tRl. If fault persist, Calculates Top
Timer start
-
-
recloser operates in a number of times with the slow curve and
reclosing time tR2. These elements are normally used in
distribution systems to reduce the influence of the transient
faults on continuity indexes.
The code implemented by using MODELS to simulate the
recloser operation, is similar to the fuse code presented in
Fig. 5. Recloser measures the RMS current and when an
overcurrent is detected, it starts its operation process. In the
presented application, recloser was settled with two fast and
one slow operation curves. Recloser has the possibility to have
single or thee phase trips. In case of single phase faults
reclosers trips the faulted phase. For two-phase or three-phase Fig. 6. Working Scheme of the Recloser model.
faults, recloser gives a three phase trip. In case of auto
extinguished faults when the system returns to normality state, D. Simulatingfaults at the distribution system
the recloser comes back to is normal state avoiding false trips. ATP executes all MODEL's code existing in the circuit in
4
each simulation step. Time variable cannot be modified by a character format. The name of each ATP file contained in
user.An example of current signal during a single phase fault the text file is "FallaXXBusYYRFZZ.atp", where XX
in node 21 is presented in Fig. 7, where the protective corresponds to the type of fault (from the 01 to the 11), YY is
devices actuations are highlighted. the node where the fault is simulated (from 01 to 25 nodes of
[A]-- the test system), and ZZ is the fault resistance value (from 01
375- to 06 for the six fault resistance values presented above).
250- First
reclosure
.... ---- Second
reclosure
Fu se
blow
2
out
B. Faulted circuit generation
(Rec) (R ec)
1 25-
This process is made from the ATP code of the power
system drawn in the ATPDraw. This file contains the
-125--Faault
First fast Second fast
protective devices developed by the using MODELS.
-250
operation
(Rec)
operation
(Rec)
To obtain a new ATP file with a new fault, it is enough to
-375-
modify the fault resistance value and the fault node location
from the original ATP file. Each copy of the original file is
VS10 0.11i 0.22 0.33 0.44 [seci associated to a different fault situation.
Fig. 7. Line current measured at the distribution substation during a fault in
As example, a 10X phase B to ground fault in node 5 is
node 21.
presented. The base file code at lines 1225, 1226 and lines
from 1291 to 1297 contains the information shown in Fig. 9.
IV. AUTOMATIC FAULT GENERATION BY USING A
COLLABORATIVE STRATEGY BETWEEN ATP & MATLAB®. C 1 2 3 4 5 6 7 8
C 345678901234567890123456789012345678901234567890123456789012345678901234567890
To develop a data base which contains voltage and current C < n 1>< n 2><refl><ref2>< R >< L >< C >
waveforms registers, different types of faults were simulated C < n 1>< n 2><refl><ref2>< R >< A
RA STNC 40.
>< B ><Leng><><>0
0
in the test system shown in Fig. 2. This is necessary since RB STNC 10. 0
RC STNC 40. 0
the KBM needs of a considerable amount of examples to
learning and validation stages. Fig. 9: ATP code details.
Automatic fault generation consists of make a predefined The first four lines in Fig. 9 help to identify each column
number of copies of the ATP file which contains the of the text and it serves as template guide to maintain the
distribution system, each one with the all 11 possible fault position of text in ATP. Next two lines are commentaries and
types, fault resistance and faulted node. To have a complete
help the user to identify the initial and final nodes where each
data base, faults were simulated in each system node. In
device is connected, in addition its resistance, inductive or
addition, faults were simulated using six different fault capacitance values.
resistances: 0.5Q, 5Q, 10Q, 20Q, 30Q, 40Q [17][18]. Next lines contain information about position and fault
resistance values. In this case, information presented can be
Automatic fault simulation by means of the interaction interpreted as: Exist an element connected between nodes RA
between ATP and Matlab is shown in Fig. 7, and it is and STNC; the element is a resistance of 10 X. The final zero
following described: number placed at the 80 column indicates that the user does
not want read any electric variable from this element, if a
variable as current or voltage wants to be registered, number
1 or 2 respectively should be placed. For the fault type control
we use a set of switches as shown in Fig. 9, by closing or
opening switches it is possible to obtain any of the 11 fault
types described before.
Lines from 1306 to 1311 contain the information of fault
switches control type, as shown in Fig. 10.
/swuErCH
C < n 1>< n 2>< Tclose ><Top/Tde >< Ie ><Vf/CLOP >< type >
STNC -1. 2. 0
N5A RA 2. 2. 0
N5C RC 2. 2. 0
N5B RB .04 2. 0
Fig. 10. ATP code details of the switch to performs faults.
Figure 10 presents the code of switches used to generate
the fault. Switches configuration is presented in Fig. 11. The
information contained in Fig. 10 is next described:
Fig. 8. Interaction scheme between ATP and MATLAB for automatic fault - A switch between STNC and ground (when there is a blank
generation.
space is interpreted as ground); the closing time is -1 sec, it
A. Fault list generation indicates that this switch is closed since simulation starts.
Matlab was used to generate the fault list. The fault list is The opening time is 2 sec.
obtained by means of the concatenation of text and numbers in - A switch connected between node 5 phase A and node RA;
5
the closing time is 2 sec., however the simulation time is voltage and its RMS values, measured at the distribution
only 1.i s. This means that the switch remains opened during substation were obtained. The size of the data base for this
all simulation time. Same situation is for phase C. system is 1.04GB, having saved only the mat files. A
The last switch connected between node 5 phase B and node sampling frequency of 40 kHz and a simulation time of 1,1
sec were used.
RB. The closing time is 40ms, and the opening time is 2 sec.
After that, the switch closes itself and remains closed during 930 faults cases were automatically simulated, and the
the remaining simulation time. This switch generates the simulation time was almost 36 hours, using a Pentium IV 280
GHz. It is because the number of faults performed, the high
phase B to ground fault.
sampling frequency and the long simulation time required to
register the operations of all protective devices. In addition, it
was observed that the models used as protective devices
reduce the speed of the simulation. According to, it shows that
it is not logical to make these simulations manually, because
the long time required by a qualified operator.
As example, from a three-phase of 40 X fault resistance in
Fig. 11. Switch used to generate a fault at node 5. node six. Transient frequencies are 3790 Hz, 3780 Hz and
C. Generating the automatic execution file. 3784 Hz, respectively for phases A, B and C. Transient
duration are 5.05 ms, 5.35 ms and 5.48 ms in each phase.
To automatic execution of the circuits previously Phase C transient signal is depicted in figure 13
generated, an instruction file batch processing with .bat 20.
extension is used. The file to execute all the simulations is [k\~
presented in Fig. 12.
101
atpwnt filename.atp *-R
del for *.tmp *.dbg *.lis
0-
move filename.pI4 new-directory
Fig. 12. File used to execute all simulations. 10 \X
The first instruction executes the .atp archives generated in
the previous stage. During the .atp file simulation some files
as .is, .dbg, .tmp and pl4 are generated. In this case, the -20_
descriptors are obtained for each specified fault file. These 95-
sets are used to train and validate the fault location KBM.
V. RESULTS -95-
VI. CONCLUSIONS [10] Z. Gaing. "Wavelet-Based Neural Network for Power Disturbance
Recognition and Classification" IEEE Transactions on Power Delivery,
Power quality is a fundamental aspect to distribution Vol. 19. 2004. pp 1560-1568.
utilities. The supply continuity indexes are very important [11] Li-Xing W. "A Course in Fuzzy Systems and Control". Prentice Hall.
because these reflect the economical and social impact of 1997
[12] J. Weston, C. Watkins. "Multi-Class Support Vector Machines".
interruptions at the end users. It also influences facilities due Proceedings on VII European Symposium on Artificial Neural Networks,
to the economic implications of poor service quality. 1999. Pags: 219-224.
The simulation of protective devices and schemes widely [13] Cormane, A. Vargas, H. Ord6foez, G. "Modelo Estadistico para la
Localizaci6n de Fallas en Sistemas de Distribuci6n" III Simposio
used in power distribution systems reflects a real circuit Internacional sobre Calidad de la Energia Eletrica SICEL 2005
behavior. It allows extracting useful information as Bogota D.C., Colombia. Noviembre 16-18 de 2005. (In Spanish).
descriptors, required by the knowledge based methodologies [14] Mora J. Serrano H., Ord6foez G. "Mejoramiento de los Indices de
to solve the fault location problem. To perform simulations Calidad del Servicio mediante la Caracterizaci6n de Sefnales para
Localizaci6n de Fallas en Distribuci6n". III International Symposium of
tools well recognized as ATP and Matlab were used. The Power Quality 2005. IEEE Branch - Colombia, Bogota, Noviembre
protective device MODELS here described, can be useful for 2005. (In Spanish).
different applications of real power distribution systems. [15] Das R. "Determining the locations of faults in distribution systems".
Canada, 1998, 206 p. Doctoral thesis. University of Saskatchewan
The automatic simulation of faults has allowed creating a Saskatoon.
data base that satisfactory describes the distribution system [16] Lewis J. "Protective relaying-Principles and applications", Ed Marcel
behavior under several fault conditions. Dekker, Inc. New York 1987, pp 390-409
The methodology here proposed and the fault data base [17] Lee S., Choi M., Kang S., Jin B., Lee D., Ahn B. "An intelligent and
Efficient Fault Location and Diagnosis Scheme for Radial Distribution
developed help to solve the fault location problem. This Systems" IEEE Transactions on Power Delivery, Vol. 19, No. 2, April
information is used mainly by the knowledge based methods 2004. Pages 524-531.
(KBM), but it also serves to test model based methods (MBM) [18] Dagenhart J. "The 40- Ground-Fault Phenomenon" IEEE Transactions
oriented to fault location. on Industry Applications, vol. 36, no. 1, pages 30-32, 2000.
VII. REFERENCES
VIII. BIOGRAPHIES
[1] Mora, J. "Voltage Sag Characterization and Classification for J. J. Mora Received his BSc in Electrical
Diagnosis in Electric Power Quality Domain", Master Thesis. University Engineering from Industrial University of Santander
of Girona, Spain - 2003. (UIS), Colombia in 1996, MSc in Electrical Power
[2] Bollen, M. "Understanding Power Quality Problems: Voltage Sags and from UIS in 2001, MSc in Information Technologies
Interruptions". IEEE Press. 2000. from University of Girona, Spain in 2003. He is
[3] IEEE Std C37. 114 "IEEE Guidefor Determining Fault Location on AC currently professor of the Electrical Engineering
Transmission and Distribution Lines" Power System Relaying School at the Technological University of Pereira,
Committee 2004. Colombia. He is also working towards his PhD at
[4] IEEE Std C62.48. "IEEE Guide on Interactions between Power System University of Girona. Member of GPE (Col) research
Disturbances and Surge-Protective Devices" Surge-Protective Devices group on Energy Systems and eXit (Sp) group of
Committee. 1995 Control and Artificial Intelligence. Interest Areas: Power Quality, Transient
[5] Mora J., Perez S., Rodriguez J. "Hybrid model based on agents structire Analysis, Protective Relaying and Artificial Intelligence.
to locate faults in power distribution systems". Scientia et Technical
Journal. Vol 30, May 2006, Pags 55-61 (In Spanish). J. Melendez. Obtained his BSC in
[6] Mora J., Mufnoz S., Carrillo G. "Tecnicas algoritmicas de localizaci6n Telecommunication Engineering at the Universitat
de fallas como alternativa para reducir el efecto de las salidas en Politecnica de Catalunya (UPC, Spain) in 1991 and
sistemas de potencia - Una revisi6n" III Simposio Internacional sobre the Ph.D. degree in Engineering by the Universitat de
Calidad de la Energia Electrica SICEL 2005 Bogota D.C., Colombia. Girona (UdG) in 1998. Titular professor UdG. Interest
Noviembre de 2005. (In Spanish). Areas: Power Quality, Knowledge-based techniques
[7] Thukaram D., Khincha H. and Vijaynarasimha H. "Artificial Neural Areas......
for fault detection, diagnosis and supervision of
Network and Support Vector Machine Approach for Locating Faults in i......_..... industrial processes and its application to real process.
Radial".Distribution Systems". IEEE Transactions on Power Delivery,
Vol. 20, No. 2, April 2005. Pp 710 721
[8] ATP Rulebook. Comite Argentino de Usuarios del EMTP/ATP, J. C. Bedoya. Received his BSc in Electrical
Argentina (CAUE). Marzo de 2002. Engineering from Technological University of Pereira
[9] Waissman, J. Aguilar, M. Dahhou, B. "Construction d'un Modele (UTP), Colombia in 2005. MSc student of Electrical
Comportemental pour la Supervision de Procedes: Application a une Engineering. He is assistant researcher at the
Station de Traitement des Eaux", Doctoral These, Institut National Electrical Engineering School at UTP. Member of
Polytechnique de Toulouse. France, 2000. GPE (Col) research group on Energy Systems. Interest
Areas: Power Quality, Transient Analysis, and
Artificial Intelligence.