You are on page 1of 5

A Novel Approach for Online Fault Detection in

HVDC Converters
J. Moshtagh, M. Jannati, H. R. Baghaee and E. Nasr

to ground (DLG) and three phase fault in AC system has been


Abstract— A novel approach for fault detection in high voltage DC detected. One of the drawbacks of this approach is neglecting
transmission systems using neural networks is presented. In the dynamic of HVDC controllers. In this paper, by a novel
presented method, at first, Harmonics of voltage waveform in rectifier criterion and using an adaptive linear neuron (ALINE), new
side are derived rapidly by using an adaptive linear neuron. Then, approach is presented for detection of different types of fault
different types of faults including DC line fault, AC system Fault and
in each part of HVDC system (DC transmission line, AC
Converter’s faults are detected property using proposed criterion. In
the under study voltage DC transmission systems system, rectifier and system and converters). To have more accuracy in the
its controllers and required filters is modeled completely. A proposed proposed method, Dynamic behavior of rectifier and inverters
criterion is tested on a high voltage DC system by computer and its control system is exactly analyzed. Computer
simulation performed in MATLAB/Simulink environment. simulations are performed in MATLAB/Simulink to
Simulation resultes demonstrates that the proposed approach can be examinate accuracy of the proposed method. Simulation
used for online fault detection in high voltage DC systems. results indicate that the proposed method is effective, reliable,
accurate and fast in fault detection.
Keywords—High Voltage Direct Current Transmission, Fault This paper is organized as follows. The system under study is
Detection, Power Electronic Converters, ADAptive Linear Neuron introduced in section II. DC system faults are analyzed in
(ADALINE).
section III. Section IV gives an overview of ADALINE neural
network architecture. Simulation results and proposed
I. INTRODUCTION
criterion for fault detection are given in section V. Finally,

R ECENTLY, Hybrid AC and DC transmission systems are


increasingly used in power systems. As a result of this
technology, reliability and security of power systems are
conclusions are summarized in section VI.

II. SYSTEM UNDER STUDY


improved. Because of the controllability of converters, high
In order to analyze and diagnosis of fault in HVDC system,
voltage direct current (HVDC) transmission technology has
several different cases are considered. In practical HVDC
been used to enhance stability margins of interconnected
system, usually 12-pulses converters system is used. Fig. 1
power systems [1-2]. Nonetheless, and reliable operation of
demonstrates this configuration in MATLAB/Simulink
HVDC transmission systems depends on fast detection and
environment. In this system a 1000MW (2kA, 500kV) DC
clearing of fault. In recent years, revelation and detection of
transmission line is used to transmit the electrical power
fault based on a database has been a matter of grave challenge
through a 500kV, 5000 MVA, 60Hz AC network to a 345kV,
for researchers. In this method, fault is detected by searching
10000MVA, 50Hz one. Rectifier and inverter is connected
in database. This approach has significant drawback;
through a 300kV DC transmission line. Transformer’s tap
searching in database, which is time-taking and moreover,
changer is not considered in modeling process and in fact
illegal operation is probable with the presence of noise.
transformer’s tap is assumed to be constant. Transformer’s tap
Another approach for fault detection in HVDC systems
is 0.9 in rectifier side and 0.96 in inverter side. Note that
considering conductive state of thyristors has been presented
usually transformer’s tap is assumed to be constant in
in [3]. In [4] operation of a HVDC converter under various
transient and dynamic stability studies. Required reactive
normal and abnormal situations has been analyzed and then
power for rectifier is produced by a capacitor bank. Moreover,
fault detection method is presented. In [5] an artificial neural
13th and 11th harmonic filters are used in each side for
network (ANN) has been used for fault detection in HVDC
generating required reactive power. MATLAB control
system.
systems block set elements used HVDC control system. In the
Only four types of faults including, short circuit in DC
case that a fault is happend in rectifier, protection system
transmission system, single line to ground (SLG), double line
detects the fault and puts the rectifier in the inverter mode of
operation by changing firing angle in order to clear the fault.
The First Author is with the Electrical Engineering Department in University Also, in the inverter, a preventing system for commutation
of Kurdistan, Kurdistan-Iran (phone: 098-9188784368,
faults, detects AC faults and decreases the maximum firing
moshtagh79@yahoo.com. The Second, third and fourth authors are with the
Electrical Engineering Department in Amirkabir University of Technology, angle limit to reduce the effect of commutation error. This
Hafez Ave, Tehran-Iran (phone: 0098-21-66466009; fax: 0098-21-66406469; system detects the AC under voltage. After that, protection
e-mails:, mohsen.jannati@aut.ac.ir, hrbaghaee@aut.ac.ir and system is disabled. Overal control system generates required
ehsan.nasr333@gmail.com )

978-1-4244-1933-3/08/$25.00 ©2008 IEEE 307


Fig. 1 Schematic of 12-pulse HVDC system in Simulink

Fig. 2 DC line voltage in Rectifier terminal: a) Normal operation condition b) Commutation error for Th3 of inverter c) DC line fault d) Single
line to ground fault in AC side of inverter

control signals for starting or stopping the operation of


converter as soon as increase or decrease in refrence current is N
detected. v (t ) = Adc e − β t + ∑ Vn sin(nωt + φn ) (1)
n =1
III. DC SYSTEM FAULTS In Eq. 1, Adc e−βt is the transient dc component, β is
DC transmission line’s voltage in rectifier terminal in the decaying coefficient, Vn and φn respectively is amplitude and
different situations (fault in DC transmission line,
the phase of the nth harmonic, N is total number of
commutation error in one of thyristors of inverter, single
phase fault in AC system in inverter side and normal harmonics and ω is the fundamental frequency and assumed
operation) is depicted in Fig. 2. This Figure shows that to be constant. So, signal v (t ) can be expressed in discrete
different faults cause quasi-similar transient. Using the form as below:
conventional methods, accurate Fault detection and authentic N N

decision making for protection is not possible. This is more v(k ) = Adc (1 − β kTs ) + ∑ An sin nωt (k ) + ∑ Bn cos nωt (k ) (2)
n =1 n =1
difficult with the presence of noise [5].
In the Eq. (2), the term Adc (1 − β kTs ) represents the first
IV. ADALINE NEURAL NETWORK ARCHITECTURE two term of the Taylor series expansion of the decaying dc
An ADALINE is a n-input single output neural network component, Ts is 2π / ω N s , N s is sampling period, An is
whose output is a linear combination of its inputs [6]. An Vn cos φn , Bn is Vn sin φn and t (k ) is kth sampling time.
ADALINE can be used for online following of harmonic
content of a signal. To illustrate the problem, a signal which
has some harmonics is considered as below:

308
1.2
To extract the fundamental and harmonic components
from v ( k ) , the ADALINE input vector, X ( k ) , is chosen to 1

be:
0.8
X (k ) = [sin ωt (k ) cos ωt (k ) sin 2ωt (k ) cos 2ωt (k ) ...

Voltage(pu)
0.6
... sin nωt (k ) cos nωt (k ) 1 − kTs ]T (3)
(3)
In the training process, if desire output yd (k ) assumed to be 0.4

equal to the actual signal, therefore, the W ( k ) vector, weight 0.2

vector of ADALINE, is furrier transform coefficients of input


signal. Weighting factors are selected to minimize the 0

difference between output of ADALINE and the reference


-0.2
signal. Widrow-hoff learning rule [7] which is equation based 0 0.5 1 1.5
Time(sec)
on Least Square Error (LSE) minimization is used for
training. Weighting factors correction is based on the equation Fig. 3 Voltage waveform at the beginning of transmission line with
(4): trained signal
method. Moreover, sampling rate is 32 samples in each cycle.
e( k ) X ( k ) This rate is suitable for the practical implementation using a
W (k + 1) = W (k ) + α (4)
X T (k ) X (k ) very cheap A/D chip. But, if
advance A/D chip wouldn’t be available, higher sampling rate
In this equation, W (k ) is weighting factor’s vector in kth must be chosen.
sampling time, W ( k + 1) is weighting factor’s vector in
V. FAULT DETECTION USING PROPOSED INDEX
(k+1)th sampling time, X ( k ) is the input vector in kth
Referring to Fig .2.d, it is obvious that after voltage gets its
sampling time, e( k ) is training rate and α is error correction nominal value, voltage waveform have small oscillations
coefficient. Perfect tracking is attained when the tracking environed its nominal value. In this situation, sinosuidal’s
error e( k ) is brought to zero. So, we have: coefficients in Eq (1) are close to zero. After happening a
(5) fault in the system, voltage waveform of faulted line changes
y (k ) = y (k ) = v(k ) = W T X (k )
d 0 obviously compare to pre-fault condition. This changes
resulting in non-zero coefficients in Eq (1). This is a key point
where Wo is weighting factor’s vector after the convergence for successful calculation of fault period. Thus, considering
of error to zero and equals to: the type of fault, it is clear that the transient caused by each
fault is different compare to another one. This causes
Wo = [ A1 B1 ... AN BN Adc βAdc ] (6) difference in coefficients in Eq (1). So, different types of fault
are considered as different resources of harmonics that disturb
After obtaining weighting factor’s vector, harmonic content of the voltage waveform. Voltage harmonics are achieved by the
the signal is achieved by the following equations: equation below:
N −1
2 k 2πh k 2πh
2
Vn = w( 2n ) + w( 2n − 1 ) 2 (7) Vh =
N ∑( v
k =0
k cos
N
− jvk sin
N
) (9)

w( 2n − 1 )2 (8) h is harmonic order and N is number of samples in each cycle


φn = cos −1( ) , ( n = 1, 2 ,3....N )
Vn and j is imaginary univalent. Magnitude of hth harmonic is
obtained as below:
Voltage at the beginning of DC transmission line and trained
signal is depicted in Fig .3. This Figure demonstrates that Vh = V h = Re 2 (V h ) + Im 2 (V h ) (10)
neural network response is very accurate and authentic in the So, proposed index is introduced as below:
following of refrence signal. So, this method is an online 5
method. THI = ∑V n (11)
In this paper, ADALINE neural network is used for the n =1
estimation of harmonics in the voltage waveform of rectifier
where, THI is Total Harmonic Index. In pre-fault and post-
side. This method is more accurate that furrier analysis
fault condition, this index takes a small value close to zero.
method presented in [7] and requires less calculation.
During the each types of fault because of high amount of
Moreover, implementation of the presented method is more
harmonics in voltage waveform during the fault, this index
convenient. The results presented in [8] show that
takes an considerable value. After that the fault is cleared, the
convergence time of harmonic estimation process using
value of index start decreasing.
ADALINE neural network is less than half of a cycle. But, at
least one complete period is required in furrier transform

309
Simulations performed for standard 12-pulse HVDC system HVDC system faults and protection against abnormal
show that the transients caused by various types of faults lead conditions.
to different maximum values of proposed index. So, type of
fault can be obtained by measuring the peak of THI. Fig .4
depicts the flow chart of the proposed approach. Table .1 VII. REFERENCES
shows the peak of THI for different types of faults. [1] W. Gang, Wu Min, L. Haifeng, H. Chao "Transient based Protection for
HVDC Lines Using Wavelet Multi-resolution Signal Decomposition",
VI. CONCLUSION IEEE/PES Transmission and Distribution Conference & Exhibition, 2005, pp.
1-4.
Analysis of HVDC signals during the fault is a matter of [2] L. Shang, G. Herold; J. Jaeger , R. Krebs and A. Chao "High-Speed Fault
grave concern for reliable and authentic operation and Identification and Protection for HVDC System Line Using Wavelet
protection of HVDC transmission system. Proposed approach Technique", Porto Power Tech conference, 2001, Portugal, pp. 10.-13.
[3] K.S. Swarup, and H. S. Chndrasekharaiah, “PDES: Fault Diagnosis Expert
is a novel and simple method for accurate and fast fault System for HVDC System”, second symposium on Expert system for HVDC
detection. In this paper, different types of faults in HVDC system, USA, 1989, pp. 296-302
system are analyzed and new Index for fault diagnosis and [4] N. Bawane, A.G. Kothari, “Artificial Neural Network based Fault
Identification of HVDC Converter”, 4th IEEE International Conference on
Diagnostics for Electric Machines, Power Electronics and Drives, 2003, pp.
152 – 157.
[5] L.L. lai, F. Ndeh-che, “Fault Identification of HVDC System with Neural
Network”, Proceeding of 5th European conference on power Electronics and
Application, 1994, pp. 231-263.
[6] M.T. Hagan, H.B. Demuth, M. Beal, "Neural Network Design", PWS
publishing company, 20 park, Boston, Ma 02116-4324
[7] Zi-Qin Wang, Michaael T. Manry and Jeffrey L. Schiano "LMS Learning
Algorithms: Misconceptions and New Results on Convergence", IEEE
Transactions on Power Delivery, 2000, Vol. 11, pp. 47-56.
[8] Mostafa I. Marei, Ehab F. El-Saadany and M. A. Salama, "A Processing
Unit for Symmetrical Components and Harmonics Estimation Based on a New
Adaptive Linear Combiner Structure", IEEE Transactions on Power Delivery,
2004, Vol. 19, pp. 1245 – 1252.

VIII.BIOGRAPHIES

Jamal Moshtagh: received the B.S. degree in


Electrical Power Engineering from Sharif University
of Technology (Iran) in 1989 and the M.S. degree in
the same field from Kh. Nasir University of
Technology (Iran) in 1995. He received the Ph.D
degree from the University of Bath (UK) in 2006. At
Fig. 4 Flow chart of proposed method for diagnosis of fault type in the present, he is the assistant Professor of Electrical
HVDC system engineering Department in University of Kurdistan. His areas of interest
include power distribution systems analysis, application of Intelligent systems
protection of HVDC transmission is presented. Simulation in Power Distribution Systems. He has published more than 20 papers in
results demonstrates that proposed yardstick with the aforementioned field.
TABLE I Mohsen Jannati was born in Shahreza, Iran in 1982.
PEAK VALUE OF HARMONIC INDEX FOR VARIOUS TYPES OF He received B.S. in electrical engineering from Isfahan
FAULTS University of Technology (IUT), Isfahan, Iran in 2006.
Fault type Proposed Index Currently, he is a M.S student at the department of
value electrical engineering of a Amirkabir University of
Single Line to Ground (SLG) fault for phase A 0.801 Technology (AUT), Tehran, Iran. His main fields of
Double line fault for phase A&C 0.1692 research are Power System Transient, Digital
Double line fault for phase A&B 0.16 Protection of Power System and Power System
Single Line to Ground (SLG) fault for phase C 0.1345 Modeling.
3 phase line fault 0.124
Hamid Reza Baghaee received the BSc degree in
Fault in DC transmission line 0.1189
Electrical Engineering from Kashan University in
Single Line to Ground (SLG) fault for phase B 0.094 2006. Currently he is graduate student of Power
Double line fault for phase B&C 0.084 Engineering in Amirkabir University of
Commutation error for Th3 0.053 Technology. His research interests are power
Misfire error for Th3 0.038 system dynamic and control, HVDC & FACTS
Fire through error for Th3 0.024 devices, Distributed Generation (DG), power
market and application of Artificial Intelligence in
application of ADAptive LINear NErone (ADALINE) leads power systems.
to authentic and reliable diagnosis of HVDC line’s faults,
commutation error and AC system faults. With the application
of this novel approach, simple, accurate and valid solution
with a high degree of reliability is achieved for detection

310
Ehsan Nasr Azadani was born in Isfahan, Iran. He
received his B.Sc degree in Electrical Engineering from
Zanjan University in 2006. He is currently an M.Sc
student in Department of Electrical Engineering at
Amirkabir University of technology (AUT). His
research interests include deregulation in power
systems, Artificial Intelligence Applications in Power
System Studies, HVDC and Renewable energy.

311

You might also like