Sie sind auf Seite 1von 6

2016 International Conference on Computational Science and Computational Intelligence

Modeling Characteristic Curves of Digital


Overcurrent Relay (DOCR) for User-defined
Characteristic Curve Using Artificial Neural Network
Anang Tjahjono, Ardyono Priyadi, Margo Pujiantara, Taufik Taufik, Ali Shaban, Xiao-Hua Yu
Mauridhi Hery Purnomo Electrical Engineering Department
Department of Electrical Engineering California Polytechnic State University
Institut Teknologi Sepuluh Nopember San Luis Obispo, California, USA
Surabaya, Indonesia

Abstract Intelligent-based protective relays in the power Artificial intelligence (AI) has brought the technological
system is currently growing very rapidly with the use of embedded advancement toward mimicking human intelligence. Examples
systems and artificial intelligence algorithms. Overcurrent relays such as fuzzy logic control, artificial neural networks and their
operated based on the standard characteristic curve will move combinations between fuzzy neural networks have been widely
toward digital overcurrent relay (DOCR) using a flexible utilized. In the power system especially for overcurrent
characteristic curve based on the user settings. DOCR with the protection relay AI has also been widely studied [14-15] such as
standard characteristic curve cannot be used if the user must make modeling of DOCR tested employing a digital signal processor
modifications to the curve for field implementation purpose. In [16], a computer or microcontroller [18-19].
this paper, a new DOCR based on modeling a characteristic curve
using Artificial Neural Network (ANN) will be presented. Four It has been known that when it comes to OCR, end users
types DOCR with the standard and user-defined characteristic often face difficulties to make modifications on its characteristic
curve were developed and tested. Results demonstrate their curve. Therefore, a new method must be sought to ease off the
successful operation with highly accurate results. The results also process and hence to provide convenience for end users to
prove that the proposed DOCR with user-defined characteristic implement OCR. In this paper, the OCR characteristic curves
curve may serve as a convenient alternative to the conventional using standard equations will be solved by performing relay
DOCR particularly on the implementation phase. modeling using artificial neural network (ANN). With the ability
to perform modeling, it would be simple to modify the
Keywordsdigital overcurrent relay; characteristic curve,
neural network; modeling; user defined.
characteristic curves as desired and needed by the user. This
paper will present the results of modeling using ANN in OCR
using the standard characteristic curve equation for each relay
I. INTRODUCTION that is classified as the standard inverse (SI), very inverse (VI),
Power system protection has a significant contribution in extremely inverse (EI) and long-time inverse (LTI). The inputs
maintaining a stable supply of electric power, therefore it to the model consist of Time multiplier setting (TMS), the
requires a robust protection devices to ensure high reliability of current pickup or current settings and load currents while the
a power system that should not be disturbed even for a short time output yields the operation time of OCR. In addition to modeling
[1-2]. Electrical interference that occurs in an electrical system the four different types of OCR using standard equations, the
is typically a current flow that exceeds the normal current called model also incorporates user defined characteristic curve.
overcurrent. If it continues to remain the high current would lead
to short circuit. To protect from such a disaster, an overcurrent II. DIGITAL OVERCURRENT RELAY
relay (OCR) plays a very important role to maintain the stability
as well as for coordination among power system equipment [3- Digital-based relaying was first contemplated during the late
4], 1960s. The first article written by Rockefeller [20] suggests that
all of the power system could be protected using digital
Development of relays has been shifted from its original computers, and from that moment the researchers has actively
mechanical type into digital devices such as using a simple been conducting research in the field of digital protection.
microprocessor to realize the distance relay has been done [5] According to the standard IEC-60255 OCR consists of a
while applying various types of overcurrent relay characteristic standard inverse, extremely inverse, very inverse and longtime
curves [6]. Other alternatives have been conducted by using inverse, the graph as shown in Fig. 1.
EPROM memory that is filled with the characteristics of relay
[7]. Others investigated the use of a simulation function for the The standard equation for the OCR characteristic curve
characteristic curve [8-10]. Other studies using FPGA and shown in Fig.1 using IEC-60255:
module board to realize more flexible overcurrent relay have t = TMSK/((Iload/Ipick)^-1) (1)
also been done [11-13].


978-1-5090-5510-4/16 $31.00 2016 IEEE 475


478
DOI 10.1109/CSCI.2016.96
 reaches a value determined by the setting delay short circuit
current (Trip_Hset).
TMS: Time Multiplier Setting
I_load: Load Current
I_pick: Pickup Current or Current Setting

Fig. 2. Function block diagram of DOCR

Parameters to determine the function of DOCR is the time


multiplier setting (TMS) and the relay type. Fig.3 illustrates
important input and output parameters of DOCR. Fig.4 is a
timing diagram working operations of DOCR.

Fig. 1. OCR characteristic curve

The constant values of K and are given in Table I.

TABLE I. LIST O CONSTANT VALUE

Type of Relay K
Standard Inverse 0.14 0.02
Very Inverse 13.5 1
Extremely Inverse 80 2
Long Time Inverse 120 1 Fig. 3. Parameter input and output of DOCR

DOCR is actually an OCR with mechanical parts using


microprocessor or microcontroller as its main processor, along
with a current sensor and a voltage sensor to obtain current flow
at any given time. The algorithms installed in the
microcontroller memory unit will enable us to control a relay
operation to disconnect the line in case the current exceeds the
current setting or pickup of the OCR.
Fig. 2 is one example of a functional block diagram of
DOCR using a microcontroller unit (MCU). To get the value of
current and voltage, it uses sensors and signal conditioners that
will be read by the analog to digital converter (ADC) of the
microcontroller. In the memory of the MCU, a program is
installed to function as a relay. In addition, DOCR has
communication peripherals that are used for setting and
monitoring of power system equipment.
In general, DOCR operates by reading the conditions of the Fig. 4. Timing diagram of DOCR
current flowing at a time. If the current exceeds the preset pickup
current and does not exceed the short-circuit current (I_Hset) Fig.5 shows DOCR using four types of OCR namely SI, VI,
then DOCR will calculate how much the delay operation is EI and LTI. By providing the TMS and current pickup or current
needed, based on the characteristic curve equation. When the setting (I_pickup) and types of DOCR, the current flow I_load
time delay value has been reached, then the MCU will instruct will feed into the model by calculating the ratio I_load /
the relay to trip. If the current exceeds the preset short-circuit I_pickup. The result will produce a delay value for the relays to
current value, then the timer will start counting until the current trip by issuing a trip signal as output.

479
476





where the superscripts indicate the position of the layer.

Fig. 5. DOCR with four types of relay

III. ARTIFICIAL NEURAL NETWORK


Artificial neural network (ANN) is a collection of simple
processors called nodes connected with each other, where each
node can only perform mathematical tasks in one way. The
number of nodes that are connected may be high, and so it may
generate very large jobs, as shown in Fig. 6. The input "i" will
be connected to all the nodes n, as well as between nodes. In Fig. 7. Node of Neural Network
each node is a mathematical function that processes an incoming
signal to the node and output the resulting signal to each of its
connected nodes. There are also weight and bias as parameters
which must be found through training with reference to the value
of the output of "o" given the input " i ".

Fig. 8. MLP network

The next step is learning to network by acquiring dynamic


parameters w and b until they reach the current output value
approximately equal to the desired target value. Alternative
Fig. 6. Diagram of artificial neural network learning algorithm may then be selected according to the
problem. The modeling is done with the following main
Fig. 7 shows one node of a neural network with input i for n parameters
and there is a bias b. The output may be written as:

Number of inputs and outputs
Number of layers.
  Feedforward backpropagation.
Function of the selected learning.
The output is calculated through activation function or Activation function for hidden layer and output layer
transfer function: Parameter learning and other functions
IV. RESULTS AND DISCUSSION

A. Modeling DOCR with the standard characteristic curve
equation
Fig. 8 depicts the multi-layer perceptron (MLP) network. In
this case there are two layers, a hidden layer and an output layer, Modeling is designed for the four types of OCR wherein
each layer has a different amount of nodes as needed. Each node each relay has the TMS ranging from 0.1 to 1.0 and IL/Ipick
consists of input, weight, bias and activation functions. In the ratio ranging from 1.1 to 20. Using the standard equation, the
MLP network can be calculated output in the forward direction value of TMS and value IL/Ipick produced a number of 390 rows
by: of data. The network was designed as follows and is shown in
Fig. 9.

 Two inputs TMS and IL/Ipick



A single output t_opr or delay time
Two-layers, 1 hidden layer with 30 neurons and an output
Assuming there are 3 neurons in the hidden layer and two layer
activation functions, then the total output is: Use of feedforward backpropagation

480
477
Training function with Levernberg-Marqurdt. 0.3 0.000107 0.001203 0.000559 0.004610
Transfer function or activation function in the hidden layer 0.4 0.000107 0.001037 0.000769 0.003955
uses hyperbolic tangent or linear to the output layer.
0.5 0.000114 0.001122 0.000531 0.004713
0.6 0.000106 0.001117 0.000621 0.004102
0.7 0.000117 0.000786 0.000761 0.004742
0.8 0.000107 0.001174 0.000533 0.004535
0.9 0.000152 0.001022 0.001005 0.004971
1.0 0.000167 0.000878 0.000686 0.006065

Fig. 9. Design layout for standard characteristic curve

After the learning process and reaching 1000 epochs, the


simulation yielded mse =3.968e-10 for SI relay, mse =7.4582e-
9 for VI relay, mse =3.2794e-9 for EI relay, and mse =1.9747e-
9 for LTI relay. Results are plotted in Fig. 10.

Fig. 11. Modeling results for the characteristic curve SI

(A) (B)

(C) (D)
Fig. 12. Modeling results for the characteristic curve VI
Fig. 10. Best of Performance mse (A) SI, (B) VI, (C) EI, (D) LTI

After the evaluation and calculation of RMSE, the


characteristic curves resulted with the ANN model using the
standard equations are shown in Table II. The results are also
plotted in Figs. 11 to 15. Fig. 16 depicts the characteristic curve
of the modified relay. These results further demonstrate the
ability of changing the characteristic curve in the OCR should a
user require to do so during its field implementation.

TABLE II. LIST O RMSE

SI VI EI LTI
TMS
RMSE RMSE RMSE RMSE
0.1 0.000143 0.001421 0.000858 0.008116
0.2 0.000142 0.001594 0.001004 0.005830 Fig. 13. Modeling results for the characteristic curve EI

481
478
The results of modeling are given in Table III, and the results
of the characteristic curve graph are shown in Fig. 18. When
compared with the characteristic curves that have been modified
with characteristic curve model with Neural Network, the results
show accurate performance with RMSE= 0.003933 as shown in
Fig. 17. It proves that the neural network has been able to model
the characteristic curve relay. Fig. 18 presents three curves: the
original characteristic curve before being modified, after being
modified, and the characteristic curve modeling results with
Neural Network.

TABLE III. LIST O CONSTANT VALUE


Epochs 2000 Maximum epoch reached
Fig. 14. Modeling results for the characteristic curve LTI Best Training performance mse at
Performance 2.4628e-05
epoch 2000
Gradient 5.89e-05 At epoch 2000
B. Modeling User-defined Characteristic Curve Relay Mu 1.00e-8 At epoch 2000
Regression value 1 r1
In Fig. 15 shows relay characteristic curve that has been
Slope of regression 0.7926 m1
modified from SI-type relay characteristic curve with TMS = Offset of regression 0.9884 b1
0.1. Modeling the characteristic curve using Neural Network 379 data, target curve and model
with planning parameters uses the following: RMSE 0.003933
curve

Feedforward Backpropagation
Two layers, a single hidden layer of 10 neurons, there is
one output layer
One input and one output
Training data as much as 379
Levernberg-Marquardt training algorithm
Algorithm performance Mean Squared Error (mse)
Hidden layer used hyperbolic tangent sigmoid transfer
function
The output layer is used linear transfer function.
2000 epochs

Fig. 17. Modeling results for user defined characteristic curve

Fig. 15. Characteristic curve modified by the user

NN planning image shown in Fig. 16 uses the input data that


is the value IL / I_pick and one output that is the time to trip the
relay.
Fig. 18. Comparison charactersitic curve

V. CONCLUSIONS
In this paper, modeling characteristic curves of digital
overcurrent relay (DOCR) for user-defined characteristic curve
using Artificial Neural Network has been presented. Simulation
Fig. 16. Design layout for user defined characteristic curve test of the proposed model demonstrates the accuracy of the

482
479
proposed model. This implies that the proposed DOCR offers a Relay, in 2014 6th IEEE Power India International Conference
significant opportunity to develop DOCR using characteristic (PIICON), 2014, pp. 15.
curves with rmse=0.0034, mse=2.463e-05 and regression r1=1, [14] D. N. Vishwakarma and Z. Moravej, ANN based directional overcurrent
relay, in Transmission and Distribution Conference and Exposition,
m1=0.7926, b1=0.9884. User defined characteristic curve is 2001 IEEE/PES, 2001, vol. 1, pp. 5964 vol.1.
done by entering the coordinate points between the current and [15] H. A. Abyaneh, H. K. Karegar, and M. Al-Dabbagh, Interpolation of
the operating time of the relay which will then be used as the measured data based on neural network to model overcurrent relays in
basis for modeling curves. The proposed model has also been power systems, in Control Conference, 2004. 5th Asian, 2004, vol. 3, p.
able to generate C source code that can be integrated with 13931397 Vol.3.
programming language used in a microcontroller unit to build [16] Y. L. Goh, A. K. Ramasamy, A. A. Z. Abidin, and F. H. Nagi, Modelling
the "IED" that in turn serves as a protection device in power of overcurrent relay using digital signal processor, in 2010 IEEE
Symposium on Industrial Electronics Applications (ISIEA), 2010, pp.
systems. 367370.
[17] A. Tjahjono, A. Priyadi, M. H. Purnomo, and M. Pujiantara, Overcurrent
ACKNOWLEDGMENTS relay curve modeling using adaptive neuro fuzzy inference system, in
2014 Makassar International Conference on Electrical Engineering and
This research is one of the outputs of the program Informatics (MICEEI), 2014, pp. 103108.
"Improvement of Writing International Journal" which was [18] A. Tjahjono, D. O. Anggriawan, A. Priyadi, M. Pujiantara, and M. H.
funded by the Ministry of research, technology and higher Purnomo, Digital overcurrent relay with conventional curve modeling
education of the Republic of Indonesia, authors expressed using Levenberg-Marquardt backpropagation, in 2015 International
gratitude to the ministry which has supported this program. Seminar on Intelligent Technology and Its Applications (ISITIA), 2015,
pp. 171176.
Authors also express gratitude for the support and assistance [19] G. D. Rockefeller, Fault Protection with a Digital Computer, IEEE
in the provision of collaborative research by the department of Trans. Power Appar. Syst., vol. PAS-88, no. 4, pp. 438464, Apr. 1969.
electrical engineering at California Polytechnic State University 
in San Luis Obispo California, USA. 

REFERENCES
[1] J. L. Blackburn and T. J. Domin, Protective Relaying: Principles and
Applications, Third Edition. CRC Press, 2007.
[2] Y. G. Paithankar and S. R. Bhide, Fundamentals of power system
protection. PHI Learning Pvt. Ltd., 2010.
[3] D. Birla, R. P. Maheshwari, and H. O. Gupta, Time-Overcurrent Relay
Coordination: A, 2005.
[4] M. Khederzadeh, Back-up protection of distance relay second zone by
directional overcurrent relays with combined curves, in 2006 IEEE
Power Engineering Society General Meeting, 2006, p. 6pp.
[5] A. L. St-jacques and G. Santerre, A Multiprocessor-Based Distance
Relay: Design Features And Test Results, IEEE Trans. Power Appar.
Syst., vol. PAS-102, no. 12, pp. 38423849, Dec. 1983.
[6] G. Benmouyal, Design of a digital multi-curve time-overcurrent relay,
IEEE Trans. Power Deliv., vol. 5, no. 4, pp. 17251731, Oct. 1990.
[7] K. Balasubramanian, EPROM-based overcurrent relays, in ,
Proceedings of the Thirty-Eighth IEEE Holm Conference on Electrical
Contacts, 1992, 1992, pp. 203207.
[8] H. A. Darwish, M. A. Rahman, A. I. Taalab, and H. Shaaban, Digital
model of overcurrent relay characteristics, in , Conference Record of the
1995 IEEE Industry Applications Conference, 1995. Thirtieth IAS
Annual Meeting, IAS 95, 1995, vol. 2, pp. 11871192 vol.2.
[9] H. A. Darwish, A. I. Taalab, and H. Assal, A novel overcurrent relay
with universal characteristics, in Transmission and Distribution
Conference and Exposition, 2001 IEEE/PES, 2001, vol. 1, pp. 5358
vol.1.
[10] J. C. Tan, P. G. McLaren, R. P. Jayasinghe, and P. L. Wilson, Software
model for inverse time overcurrent relays incorporating IEC and IEEE
standard curves, in Canadian Conference on Electrical and Computer
Engineering, 2002. IEEE CCECE 2002, 2002, vol. 1, pp. 3741 vol.1.
[11] K. Shehata, A. Bahaa, K. Morad, and A. Sharaf, Design and
implementation of FPGA based and microcontroller based current relay,
in The 16th International Conference on Microelectronics, 2004. ICM
2004 Proceedings, 2004, pp. 783786.
[12] C. J. Mozina, Advances in Multifunction Digital Transformer Relays,
in 2005/2006 IEEE/PES Transmission and Distribution Conference and
Exhibition, 2006, pp. 858862.
[13] G. V. Kshirsagar, G. N. Mulay, and S. Yeolekar, TMS320F28335 based
single phase overcurrent protection Implementation using Numerical

483
480

Das könnte Ihnen auch gefallen