Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/304526301

Fault detection and classification technique for HVDC transmission lines


using KNN

Conference Paper · July 2016

CITATIONS READS

27 5,563

2 authors:

Jenifer Johnson Anamika Yadav


T.K.M. College of Engineering National Institute of Technology Raipur
5 PUBLICATIONS   163 CITATIONS    188 PUBLICATIONS   3,145 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Design and Development of Computational Intelligence based Fault Classification and Location Scheme View project

Analysis of electrical power systems View project

All content following this page was uploaded by Anamika Yadav on 23 August 2016.

The user has requested enhancement of the downloaded file.


Fault detection and classification technique for HVDC
transmission lines using KNN

Jenifer Mariam Johnson1 and Anamika Yadav1

1
Department of Electrical Engineering, National Institute of Technology, Raipur, 492010,
C.G., India
jeniferjohns91@yahoo.com, ayadav.ele@nitrr.ac.in

Abstract. In this paper, we have introduced a novel fault detection and


classification technique for High Voltage DC transmission lines using K-
Nearest Neighbours. The algorithm makes use of rectifier-end AC RMS
voltage, DC line voltage and current measured at both poles. These signals are
generated using PSCAD/EMTDC and are further analysed and processed using
MATLAB. For fault detection, the signals are continuously monitored to
identify the instant of occurrence of fault, wheras for fault classification, the
standard deviations of the data over a half cycle (with respect to AC signal)
before and after the fault inception instant are evaluated. The fact that only
single-end data is used makes the relay algorithm simple and cost-effective as
no signals need to be communicated to the other end as in the case of
techniques making use of two-end data. An added advantage of this technique
is that the sampling rate is as low as 1 kHz. This eliminates the need for
expensive high frequency samplers. The technique has proven to be 100%
accurate and hence reliable.
Keywords: HVDC transmission system, Rectifier-end signals,
PSCAD/EMTDC, Fault detection, Faulty pole identification, KNN.

1 Introduction

HVDC transmission systems, although are not yet capable of replacing the
conventional AC transmission systems, are extensively being used these days due to
its superiority over HVAC transmission systems based on various aspects. The most
important of these are controllability of power flow, economy while transmitting bulk
power over very long distances, lower transmission losses and absence of stability
issues. Thus the HVDC transmission system is preferred in almost all cases where
bulk power needs to be transmitted over 500 km and above [1-3].
The transmission line is that entity of a power system that enables the
interconnection of the generating station to the end user, in case of HVDC
transmission system, this being the receiving end AC system. The protection of this
entity is of utmost importance as any maloperation of the concerned protective relay
would lead to prolonged disruption in the power flow and associated economic losses.
Most of the works that have been reported for detection and classification of faults
in HVDC transmission lines are based on travelling wave algorithms [4-6]. These
algorithms tend to be accurate, yet cannot be considered completely reliable since
their accuracy depends on the accuracy with which the wave head is detected. An
evolutionary algorithm was proposed in [7] for detection and localization of faults. In
[8], a fault localization algorithm has been proposed that is based on radial basis
function (RBF) Neural Network with wavelet packet decomposition (WPD). Another
paper, [9], has proposed the application of RBF Neural Network for fault diagnosis.
The fault analysis using KNN approach has been proposed in [10] for three phase
transmission system.

2 HVDC transmission system

The monopolar HVDC CIGRE benchmark model provided in PSCAD/EMTDC was


modified to make it bipolar. Fig.1. shows the schematic diagram of the bipolar HVDC
model that was used to carry out the fault analysis. Table 1 shows the details of the
system used.

Smoothing reactor Smoothing reactor

Rectifier side 936 km Inverter side

AC System (inverter side)


AC System (rectifier side)

936 km

Smoothing reactor Smoothing reactor

Fig. 1. Schematic diagram of HVDC system

Table 1. Details of HVDC system modelled

Module Details
Rectifier end AC system 345.0 kV, SCR = 2.5 @ 84.0 deg, 50Hz
Inverter end AC system 230.0 kV, SCR = 2.5 @ 75.0 deg, 50Hz
Transmission line ±500 kV, 936 km
Smoothing reactor 5 mH on both ends of DC line
2.1 Simulation of fault on DC line

Pole 1 to ground, pole 2 to ground and pole to pole faults were simulated at intervals
of 10 km from the rectifier end to the inverter end of the transmission line. The
sampling frequency used is 1 kHz. The faults are simulated at 0.3 s and are permanent
faults. The signals that need to be recorded are the AC RMS voltage, DC line voltage
and current on both poles, at the rectifier-end only. Consideration of the inverter-end
readings too would make the technique more reliable, but would require the two-end
data to be synchronised at one end where the rest of the manipulations are done. This
would increase the complexity and cost. Single-end data gives sufficiently accurate
results as shown in this paper.

2.2 Waveforms

Fig. 2. (a)-(e) shows the waveforms of the signals that have been used as inputs for
the KNN based detection and classification algorithm. The classifier needs to
correctly identify as to which class each waveform belongs.

(a)

(b)

(c)
(d)

(e)
Fig.2.(a)-(e) Waveforms used for relay algorithm

3 Proposed Methodology

As seen from the waveforms, the constant current controllers at the converters will
bring down the current from the actual high faulty values to near to normal values
within a very less time, depending on the speed of response of the controllers. Hence,
once a DC fault has occurred, it needs to be detected as soon as possible, for the fault
clearance to be effectively be carried out. The speed of the detection algorithm is
hence of utmost importance. Moreover, the faulty pole too needs to be identified so
that the maintenance can be carried out as early as possible to ensure continuity of
power flow.

3.1 K-Nearest Neighbours Classifier

KNN is among the simplest of all machine learning algorithms. In this algorithm, the
input vector consists of the k closest training samples in the feature space, where k is
an interger. Classification of data is done by identifying the most common class
among the k nearest neighbours. These neighbours belong to a data set with whichthe
algorithm is first trained and are determined by the distance from the test sample. In
other words, while testing, the class which occurs most frequently among the
neighboring classes of the test sample under consideration becomes the class to which
this individual test sample belongs.
There are 6 different KNN Classifiers available in MATLAB that can be used to
classify our data [11]. The Fine, Medium and Coarse KNN algorithms make use of
Eucledian distance to determine the nearest neighbours. Their details, as given in
MATLAB are:
(i) Fine KNN: A nearest neighbor classifier that makes finely-detailed
distinctions between classes with the number of neighbours set to 1.
(ii) Medium KNN: A nearest neighbor classifier that makes fewer
distinctions than a Fine KNN with the number of neighbours set to 10.
(iii) Coarse KNN: A nearest neighbor classifier that makes coarse distinction
between classes, with the number of neighbours set to 100.
(iv) Cosine KNN: A nearest neighbor classifier that uses the cosine distance
metric.
(v) Cubic KNN: A nearest neighbor classifier that uses the cubic distance
metric.
(vi) Weighted KNN: A nearest neighbor classifier that uses distance
weighting.
For implementing the fault detector, the entire time domain signals are used
whereas for fault classification, the standard deviation of the signals over a time
extending from half cycle before the insant of fault to half cycle after the instant of
fault is used to train the KNN based algorithm. The Eucledian distances are evaluated
to find the nearest neighbours and the number of nearest neighbours is chosen as 1.
For implementing the fault classification, the KNN model is trained for all the 6
techniques available in MATLAB.

4 Results and discussion

Fig.3 shows the Detector output obtained for fault on pole 1 under various fault
instances under consideration. Table 1 tabulates the delay with which the fault is
detected under various conditions. The maximum delay is 4 ms.

(a)

(b)
(c)
Fig.3 Detector output obtained using KNN for fault at (a) 0.3s, at 500 km on pole 1 (b) 0.4s,
at 500 km on pole 1 (c) 0.5s, at 500 km on pole 1

Table 2. Delay in detecting fault at various conditions


Fault type Fault location Fault instant Detection instant Delay
0.3s 0.301 s 1 ms
100 km 0.4 s 0.401 s 1 ms
0.5 s 0.502 s 2 ms
0.3s 0.302 s 2 ms
Pole 1 to ground 500 km 0.4 s 0.403 s 3 ms
0.5 s 0.502 s 2 ms
0.3s 0.303 s 3 ms
800 km 0.4 s 0.403 s 3 ms
0.5 s 0.504 s 4 ms
0.3s 0.302 s 2 ms
100 km 0.4 s 0.401 s 1 ms
0.5 s 0.503 s 3 ms
0.3s 0.301 s 1 ms
Pole 2 to ground 500 km 0.4 s 0.403 s 3 ms
0.5 s 0.502 s 2 ms
0.3s 0.303 s 3 ms
800 km 0.4 s 0.403 s 3 ms
0.5 s 0.503 s 3 ms
0.3s 0.303 s 3 ms
100 km 0.4 s 0.404 s 4 ms
0.5 s 0.504 s 4 ms
0.3s 0.302 s 2 ms
Pole 1 to pole 2 500 km 0.4 s 0.403 s 3 ms
0.5 s 0.503 s 3 ms
0.3s 0.301 s 1 ms
800 km 0.4 s 0.403 s 3 ms
0.5 s 0.502 s 2 ms

Fig.4. (a) –(f) show the confusion matrices obtained using Fine KNN, Medium
KNN, Coarse KNN, Cosine KNN, Cubic KNN and Weighted KNN respectively. The
diagonal elements represent the test samples that have been correctly classified by the
classifier, whereas the off-diagonal elements represent the number of test samples that
have been wrongly classified. TPR stands for True Positive Rate and FNR stands for
False Negative Rate.
(a) (b)

(c) (d)

(e) (f)

Fig.4. Confusion matrix for (a) Fine KNN (b) Medium KNN (c) Coarse KNN (d) Cosine
KNN (e) Cubic KNN (f) Weighted KNN

A total of 94 cases (93 faulty cases, measured at every 10 km, and 1 no-fault case)
have been used to detect and classify each of the three types of DC faults. Except
Coarse KNN (mean accuracy 84.03%), all the various KNN techniques available in
the MATLAB software give 100% accurate results. Thus, the Fine, Medium, Cosine,
Cubic and Weighted KNN techniques give accurate results and can be used as
Classifiers for faults in the HVDC transmission line.
5 Conclusion

KNN, though not a new technique, has not yet been reported, to the best of our
knowledge, for detection and classification of faults in HVDC transmission systems.
The paper thus proposes a new methodology that gives accurate results and can thus
be efficiently used as a relay algorithm for HVDC lines. Further work needs to be
done to implement fault location estimation using KNN.

6 References

1. Arrillaga, Y.H.L.J., Watson, N.R.: Flexible power transmission: The HVDC option, Wiley
(2007)
2. John, H.G.: Economic aspects of D-C power transmission. Power App. Syst., Part III. Trans.
Amer. Inst.Elect.Eng. 76, pp. 849-855 (1957)
3. J.F. Perrin: High-voltage DC power transmission, J. Inst. Elect. Eng. 7, pp. 559-31, (1961)
4. 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, pp. 1-4 (2001)
5. Crossley P.A., Gale P.F, Aurangzeb M. :Fault location using high frequency travelling wave
measured at a single location on a transmission line. Developments in Power System
Protection, Amsterdam, Holland, pp. 403-406 (2001)
6. 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)
7. Burek, A., Rezmer, J.: Evolutionary algorithm for detection and localization of faults in
HVDC systems. Int. Conf. on Environment and Electrical Engineering, pp. 1317-1322
(2015)
8. Cui, H., Tu, N.: HVDC transmission line fault localization base on RBF neural network with
wavelet packet decomposition. Int. Conf. on Service Systems and Service Management, pp.
1-4 (2015)
9. Narendra, K. G., Sood, V. K., Khorasani, K., Patel, R.: Application of a Radial Basis
Function (RBF) for fault diagnosis in an HVDC system. Proc. of the 1996 Int. Conf. on
Power Electronics, Drives and Energy Systems for Industrial Growth, vol. 1, pp. 140-146
(1996)
10. Yadav, A., Swetapadma, A.: Fault Analysis in Three Phase Transmission Lines Using k-
Nearest Neighbours. Int. Conf. on Advances in Electronics, Computers and
Communications, pp.1-5 (2014)
11. Matlab R2015a software

View publication stats

You might also like