Fault Location Estimation in HVDC transmission line

using ANN

Jenifer Mariam Johnson1 and Anamika Yadav1

1
Department of Electrical Engineering, National Institute of Technology, Raipur, 492010,
C.G., India
jeniferjohns91@[Link], [Link]@[Link]

Abstract. This paper presents a simple, yet accurate method for fault location
in HVDC transmission lines using Artificial Neural Networks. The ±500𝐾𝑉
HVDC system has been modelled using PSCAD/EMTDC and further analysis
has been done using MATLAB. Single-end AC RMS voltage and DC voltage
and current have been used to identify the location of the fault. This method is
relatively simple because the standard deviation of the fault data alone gives
acceptably accurate results while using ANN.
Keywords: ANN, fault location, HVDC transmission line, PSCAD/EMTDC.

1 Introduction

High Voltage DC transmission systems are being extensively used now owing to
the ever increasing need for electric power and hence their bulk transmission over
often very long distances [1]. Such transmissions may even involve transfer of power
between asynchronous systems. The HVDC technology comes as an economic
solution, in comparison to HVAC, to both these issues for transmissions beyond a
certain break-even distance [2-4].
Faults in the transmission line are frequent and at the same time need to be cleared
as soon as possible to ensure the continuity of power transfer and to avoid economic
losses. Since HVDC technology is preferably used for transmission line lengths above
the break-even distance, which is as high as 800km, the accurate and fast location of
faults, once they have occurred, is very important for the reliability and efficiency of
the system.
The present-day primary protection schemes for HVDC transmission lines employ
travelling wave based methods, whereas the back-up protection is based on dc
minimum voltage and dc line differential protection methods. These schemes,
however, are not sufficiently accurate or sensitive and hence are less reliable. The
present scenario hence demands better line protection techniques that would ensure
better reliability that is a key issue due to the extensive usage of the HVDC
technology these days.
The travelling wave algorithms estimate the fault location based on the time taken
by the fault generated traveling wave to propagate along the transmission line [5-8].
In methods requiring two-terminal data, the global positioning system (GPS) is
usually employed to keep the measurements synchronized. The traveling wave theory
follows that the transients that are generated as a result of faults or switching
procedures are composed of traveling waves that continue to bounce back and forth
between the fault point and the terminals until a post-fault steady state is reached. The
signals intended to be used need to be synchronized before the traveling wave
algorithm is applied. Once the signals have been synchronized, the traveling wave
algorithm is used to estimate the travel time of forward and backward fault transients
along the concerned transmission line between the fault point and the location of the
relay. The fault location is then estimated using these travel times. Although these
methods have fast response and high accuracy, the accuracy is affected by the
accurate detection of the surge arrival time. However, in methods requiring only
single terminal data, there is no need of GPS and these methods are hence more
economical [9-12]. But they require the detection of secondary reflection wave as
well.
A single ended fault detection and location method for HVDC transmission
systems using DWT was first proposed in [9]. A two-ended traveling wave-based
fault location method for an HVDC transmission system was put forth and
implemented in [10]. A two-ended fault location method for overhead HVDC
transmission line using steady-state voltages and currents was proposed in [12].
In this paper, a relatively simple fault location method has been proposed that uses
the AC RMS voltage and DC current and voltage at a single end of the HVDC
transmission line. A ±500𝑘𝑉 HVDC system was modelled using PSCAD/EMTDC
and a single line to ground fault was simulated on the positive line at different
locations from the rectifier end to the inverter end and the data was collected and
recorded. This data was analyzed in MATLAB and it was found out that the rectifier
end AC RMS voltage, DC current and DC voltage of the faulted line alone were
sufficient to determine the location at which the fault occurred. The standard
deviations of these measurements for each location were then used to train the
Artificial Neural Network. The accuracy of this technique was then verified by testing
the network with data obtained with different fault locations than those given for
training. The accuracy obtained was ±2 km (0.21%).

2 HVDC Transmission

A 936km long ±500kV transmission system has been modelled in PSCAD/EMTDC

for further analysis. The CIGRE Benchmark HVDC model which is monopolar was
modified to make it bipolar as in Fig.1. The tower structure of the ±500 kV HVDC
transmission line is also depicted in Fig.2, which has been adopted from [13].
 Smoothing reactor Smoothing reactor

Rectifier side Inverter side

936 km

AC System (inverter side)

AC System (rectifier side)

936 km

Smoo thing reactor Smoothing reactor

Fig.1. Bipolar HVDC system model

Fig.2. Tower structure for ±500 𝑘𝑉 HVDC lines

2.1 DC Transmission line fault

The common types of faults that are likely to occur in HVDC transmission lines
are the single line to ground and line to line fault. Moreover, the single line to ground
fault can occur either on the positive or on the negative polarity transmission line. For
our analysis, a single line to ground fault has been simulated on the positive polarity
transmission line, assuming the remaining two fault characteristics to be of similar
nature.
A permanent fault is simulated on the HVDC transmission like at every kilometer
from 1km to 935km. A half of this data is utilized for training the neural network and
the remaining to test the same network.
3 Artificial Neural Network

An ANN is an artificial (man-made) system motivated by research into the human

brain. It mimics the operation of the brain in performing a particular task [14]. The
brain has many features that are desirable in Artificial Intelligence systems:
 it is robust and fault tolerant; nerve cells in the brain die every day without
affecting its performance significantly
 it can deal with information that is fuzzy, probabilistic, noisy or inconsistent
 it is highly parallel
 it is small, compact and dissipates very little power

The building block of any ANN is a very simple model of the fundamental cell of
the brain -the neuron. Neurons are connected to form networks with different
structures to perform certain functions. The manner in which the neurons of a neural
network are structured depends on the learning abilities and functions that we want to
achieve.

Fig.3. Neural network architecture used

The data extracted from the PSCAD/EMTDC model of the HVDC transmission
system is used as the input and target data for training the neural network. In
particular, the rectifier end AC RMS voltage and the DC voltage and current values,
also on the rectifier end, on the faulted line are used here. The standard deviation of
these fault data over a sufficient pre-fault, during-fault and post-fault duration are fed
as the input data to the neural network in our method. The corresponding distances act
as the target for the network. The neural network is trained using the Levenberg-
Marquardt backpropagation algorithm.
The main goal of designing the neural network is to accurately estimate the fault
location along the transmission line under consideration. For this, the network is
initially trained with a set of training data set consisting of extracted samples of pre-
fault and post-fault AC RMS voltage, DC voltage and current measurements taken at
the rectifier end of the concerned transmission line from different fault case
simulation, along with the corresponding targets, being the targeted fault location.
The neural network architecture used in this paper consists of two layers (one hidden
and one output) of 20 and 1 neuron respectively, as shown in Fig. 3. The activation
functions used for the layers are tangent sigmoid and linear function respectively. The
three inputs given to the network are the AC RMS voltage, DC voltage and current at
the rectifier end. The output is the fault location corresponding to the input data given.
Once trained, the network is now ready to predict the fault location for any data
including those for which it has not been trained with.

3 Results

The fault data that had earlier not been used for training the neural network is used
to test the same network to determine how well the neural network used can identify
the fault location. The test data consist of total 468 fault cases simulated at every 2km
from 2-934km length. It is found out that the network that predicts the fault location
with an accuracy of ±2 km (0.21%), which is quite acceptable with respect to the
length of the transmission line under consideration.

Table 1 illustrates few test results along with the error statistics of proposed fault
location algorithm implemented using ANN. The error plot has been depicted in Fig.4
which shows the different between the estimated and target fault location for 468 test
fault cases. It can be observed that the error is more at the sending end from where
voltage and current measurements have been made. The minimum error obtained is
4.633e-5(4.95e-6%) for fault at 164 km whereas the maximum error obtained is -
1.9908(0.213%) for fault at 54 km.

𝐴𝑐𝑡𝑢𝑎𝑙 𝑓𝑎𝑢𝑙𝑡 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛−𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛

𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑒𝑟𝑟𝑜𝑟 = 𝑥100 (1)
𝐿𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑟𝑎𝑛𝑠𝑚𝑖𝑠𝑠𝑖𝑜𝑛 𝑙𝑖𝑛𝑒

1.5

0.5

-0.5

-1

-1.5

-2
0 50 100 150 200 250 300 350 400 450 500

Fig.4. Error plot obtained using ANN

 Table 1. Error statistics obtained using ANN

Actual Fault Fault location estimated Error (km) Percentage Error (%)
Location (km) with ANN (km)

2 1.883 0.1170 0.0125

8 8.4543 -0.4543 -0.0485
10 9.6167 0.3833 0.0409
20 19.8986 0.1014 0.0108
42 43.1579 -1.1579 0.1237
50 50.2751 -0.2751 -0.2939
54 55.9907 -1.9907 0.2126
62 61.1674 0.8326 0.0889
100 100.0509 -0.0509 -0.0054
150 149.9481 0.0519 0.0055
200 200.0856 -0.0856 -0.0091
300 299.927 0.0730 0.0078
400 400.0856 -0.0856 -0.0091
500 500.0427 -0.0427 -0.0045
600 599.927 0.0730 0.0077
700 699.8462 0.1538 0.0164
800 800.0639 -0.0639 -0.0068
900 900.0382 -0.0382 -0.0041
910 909.8746 0.1254 0.0134
920 920.1917 -0.1917 -0.0204
930 929.8417 0.1583 0.0169
934 933.9065 0.0935 0.0099

4 Conclusion

The proposed method is thus a relatively simple method for the prediction of fault
location in a bipolar HVDC transmission system. The simplicity lies in the fact that
the voltages and current data obtained can be used in the neural network by just taking
the standard deviation of the fault data for fault at various locations. The accuracy is
±1 km which can be further improved if we take the inverter end data along with the
rectifier end data. Although this option provides better accuracy, it has been avoided
owing to the additional communication channel requirement that would be involved if
the receiving end and sending end data were both required for the fault location
algorithm.
References

1. Arrillaga, Y.H.L.J., Watson, N.R.: Flexible power transmission: The HVDC option, Wiley
(2007)
2. Willis, L., Stig, N.: HVDC transmission: Yesterday and today. IEEE Power Energy Mag. 5,
22-31 (2007)
3. John, H.G.: Economic aspects of D-C power transmission, Power App. Syst., Part III. Trans.
Amer. [Link]. 76, 849-855 (1957
4. Perrin, J.F.: High-voltage DC power transmission, J. Inst. Elect. Eng. 7, 559-31, (1961)
5. Dewe, M.B., Sankar, S. and Arrillaga, J.: The application of satellite time references to
HVDC fault location, IEEE Trans. Power Del. 8, 1295-1302 (1993)
6. Nnanyakkara, K., Rajapakse, A.D., Wachal, R.: Fault location in extra HVDC transmission
line using continuous wavelet transform, IPST Conf., Delft, the Netherlands (2011
7. Magnago, F.H., Abur, A.: Fault location using wavelets, IEEE Trans. Power Del. 13, 1475-
1480 (1998)
8. Evrenosoglu, C.Y., Abur, A.: Travelling wave based fault location for teed circuits, IEEE
Trans., Power Del. 20, 1115-1121 (2005)
9. Shang, L., Herold, G., Jaeger, J., Krebs, R., Kumar, A.: High-speed fault identification and
protection for HVDC line using wavelet technique, in Proc. IEEE Porto Power Tech Conf.
Porto, Portugal, 1-4 (2001)
10. Chen, P., Xu, B. , Li, J.: A traveling wave based fault locating system for HVDC
transmission lines, in Proc. Int. Conf. PowerCon, Chongqing, China, 1-4 (2006)
11. Ancell, G.B., Pahalawaththa, N. C.: Maximum likelihood estimation of fault location on
transmission lines using travelling waves, IEEE Trans. Power Del., vol. 9, no. 2, 680-689
(2004)
12. Suonan, J. ,Gao, S. , Song, G. , Jiao, Z. , Kang, X.: A novel fault-location method for
HVDC transmission lines, IEEE Trans. Power Del. 25, 1203-1209 (2010)
13. Yuangsheng, L. , Gang, W. , Haifeng, L.: Time-domain fault-location method on HVDC
transmission lines under unsynchronized two-end measurement and uncertain line
parameters, IEEE Trans. Power Del., vol. 30, no. 3, 1031-1038 (2015)
14. Aggarwal, R., Song, Y.: Artificial neural networks in power systems. Part I. General
introduction to Neural Computing, IEEE Trans., Power Eng. 11, 129-134