International Journal of Emerging

Electric Power Systems

Volume 11, Issue 5 2010 Article 3

A New Protection Scheme for Distribution

Network with Distributed Generations Using
Radial Basis Function Neural Network

Hadi Zayandehroodi, Universiti Kebangsaan Malaysia

Azah Mohamed, Universiti Kebangsaan Malaysia
Hussain Shareef, Universiti Kebangsaan Malaysia
Marjan Mohammadjafari, Islamic Azad University of
Kerman, Iran

Recommended Citation:
Zayandehroodi, Hadi; Mohamed, Azah; Shareef, Hussain; and Mohammadjafari, Marjan (2010)
"A New Protection Scheme for Distribution Network with Distributed Generations Using Radial
Basis Function Neural Network," International Journal of Emerging Electric Power Systems:
Vol. 11 : Iss. 5, Article 3.
Available at: http://www.bepress.com/ijeeps/vol11/iss5/art3
DOI: 10.2202/1553-779X.2611
©2010 Berkeley Electronic Press. All rights reserved.
 A New Protection Scheme for Distribution
Network with Distributed Generations Using
Radial Basis Function Neural Network
Hadi Zayandehroodi, Azah Mohamed, Hussain Shareef, and Marjan
Mohammadjafari

Abstract
Distributed generations (DGs) have been increasingly connected on the distribution networks
that will have the unfavorable impact on the traditional protection methods because the
distribution system is no longer radial in nature and is not supplied by a single main power source.
This paper presents a new automated protection method using radial basis function neural network
(RBFNN) for a distribution system with DG units. In the proposed method, the fault type is
determined first by normalizing the fault currents of the main source. For implementing fault
location considering various types of faults, two staged RBFNNs have been developed. The first
RBFNN is used for determining the fault distance from each power source and the second RBFNN
is used for identifying the faulty line. To isolate the fault, another RBFNN has been developed for
determining which circuit breakers (CBs) that must be open or close. Several case studies have
been made to verify the accuracy of the method to specify the fault location and protection of the
system in distribution networks with DGs. The predicted results showed that the proposed
RBFNN based protection method can accurately determine the location of faults and isolate the
faulted line in the test power system.

KEYWORDS: distributed generation (DG), protection, fault location, distribution network (DN),
radial basis function neural network (RBFNN)
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

1. Introduction
Deregulation of the electric power industry, advancements in technology and the
need for reliable electric power has led to an increased interest in distributed
generation (DG). Compared with large generators and power plants, DG unit has
smaller generation capacity and operates at lower operational costs in distribution
networks. Besides, the application of DGs with renewable energy resources has
many advantages such as reduced environmental pollution, high efficiency,
distribution loss reduction, improvement of voltage profile and network capacity
enhancement (Jenkins, 1995; Barker and De Mello, 2000).
There are many benefits from the installation of DG, but a high
penetration of DGs in distribution systems would lead to conflicts with the current
protection procedures operated in the present networks. This is due to the fact that
the present distribution system is designed as a passive and radial network
configuration. A typical distribution protection system consists of fuses, relays
and reclosers. An inverse overcurrent relay is usually placed at the substation
where a feeder originates and reclosers are usually installed on the main feeders
with fuses on laterals. Reclosers are necessary in a distribution system since 80%
of all faults taking place in distribution systems are temporary. It gives a
temporary fault a chance to clear before setting a fuse to blow. The coordination
between fuses, reclosers and relays is well established for radial systems but
however, when DG units are connected to a distribution network, the system will
no longer be radial and this means losing coordination among the network
protection devices (Brahma and Girgis, 2004; Zayandehroodi et al., 2009). The
extent in which a DG affects protection coordination depends on the DG capacity,
type and location (Burke, 1994; Doyle, 2002; Kumpulainen and Kauhaniemi,
2004).
Hadjsaid et al. suggested that protection coordination must be checked
after connecting each DG to a distribution network (Hadjsaid et al., 1999).
However, this approach is applicable only in the presence of low penetration of
DG into the system. A study has been made on the fuse coordination problem and
suggested a solution to disconnect all DGs when a fault occurs (Brahma and
Girgis, 2002). The disadvantage of this solution is that it leads to disconnection of
all DGs even when temporary faults occur. In addition, the recloser-fuse
coordination in presence of DGs using microprocessor-based reclosers is also
addressed. It is suggested that all DGs located in the recloser downstream network
must be disconnected first before reclosure takes place to avoid asynchronous
connection. This solution is not appropriate if DGs are widely applied in a
distribution network. An approach to avoid major changes in the protective
devices so that the fuse based protection is not disturbed by DG was studied by
(Tran-Quoc et al., 2003). Theoretical studies on the impact of DG on the current

Published by Berkeley Electronic Press, 2010 1

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

seen by protective devices indicate that DGs may be blind in over current
protection (Kauhaniemi and Kumpulainen, 2004). For example, short-circuit
current fed by a synchronous generator can decrease the current seen by a feeder
relay or fuse, hence preventing its correct operation.
To overcome the above-mentioned problems, an adaptive protection
scheme for optimal coordination of over current relays has been introduced to
allow adaptation to changes in load, generation or system topology (Abdelaziz et
al., 2002). Another adaptive protection scheme for distribution networks with
high penetration of DGs is introduced using online fault diagnosis (Brahma and
Girgis, 2004). A protection scheme based on zoning for distribution network with
DG has been developed by (Javadian and Haghifam, 2008a). Here, zoning is
made by considering locations, generation capacities of DGs and loads, in which
one zone is allocated for each DG. To implement the protection scheme, a
computer-based relay with high processing power and large storage capacity is
installed at the sub-transmission substation of a distribution network (Javadian
and Haghifam, 2008a). The computer-based relay holds the main responsibility of
system protection and operates through monitoring of the currents flowing
through some specific points of the network.
A more recent protection scheme for a distribution network with DGs
considers the application of multi layer perceptron neural network (MLPNN)
(Rezaei and Haghifam, 2008; Javadian et al., 2009a; Javadian et al., 2009b).
However, considering the structure and training algorithm of the MLPNN, the
speed of this method is not suitable for fast and accurate protection. In this paper,
a new protection scheme for a distribution network with DGs using the radial
basis function neural network (RBFNN) is presented. RBFNN is considered to be
a better neural network model than the MLPNN for solving engineering problems
(Zayandehroodi et al., 2010b). The proposed protection scheme incorporates
determination of the fault type, fault location, identification of the circuit breakers
operating states during faults and determination of circuit breakers operating
states for network restoration. The fault type is determined based on the
normalized fault current of the main source while the location of faults in terms of
faulty line and fault distances from each DG and the main source is determined by
using RBFNN. To isolate a fault, RBFNN is also developed for determining
which circuit breakers that must be open or close.

2. Radial Basis Function Neural Network

The RBFNN is a feed-forward neural network consisting of three layers, namely,
an input layer which feeds the values to each of the neurons in the hidden layer, a
hidden layer which consists of neurons with radial basis activation functions and
an output layer which consists of neurons with linear activation function. A

http://www.bepress.com/ijeeps/vol11/iss5/art3 2
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

generic architecture of an RBFNN with k input and m hidden neurons is shown in

Figure 1.

W0
X1

C1 W1

X2
W2 Y
C2 ∑

Wm
Cm
Xk

Input Layer Hidden Layer Ouyput Layer

Figure 1: A generic architecture of the RBFNN

In the training of the RBFNN, the following computations are considered.

When the network receives a k dimensional input vector X, the network computes
a scalar value using,

m
Y f(X) w  ¦ w M (D )
0 i i (1)
i 1

where w0 is the bias, wi is the weight parameter, m is the number of nodes in the
hidden layer and (Di) is the RBF.
In this study, the Gaussian function is used as the RBF and it is given by

 Di2
I ( Di ) exp( ) (2)
V2

where σ is the radius of the cluster represented by the center node, Di is the
distance between the input vector X and all the data centers.

Published by Berkeley Electronic Press, 2010 3

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

The Euclidean norm is normally used to calculate the distance, Di which is

given as

¦ X
k
Di j  C ji ) 2 (3)
j 1

where C is a cluster center for any of the given nodes in the hidden layer (Yu et
al., 2008).
The implementation procedures in the training of the RBFNN are
presented as follows:

Step 1: Obtain input data and target data from the simulation.
Step 2: Assemble and preprocess the training data for the RBFNN.

Step 3: Create the network object and train the network until condition of
network setting parameters are reached.
Step 4: Test and regression analysis.
Step 5: Stored the trained network. Steps (1–5) are offline processes.
Next, the network is ready to test with the new input, which is an
online process.
Step 6: The new input is preprocessed before presented to the trained
RBFNN.

3. Proposed Protection Scheme Using RBFNN

Figure 2 shows the outline of the proposed protection scheme for a distribution
network with multiple DG units. The scheme considers three main steps in which
the first step is modeling and simulation of a distribution network by dividing the
network into several zones and generating suitable data for the RBFNN. The
second step considers implementing several functions which are identifying the
fault type by using the normalized currents, determining the exact fault location
and determining the circuit breakers (CBs) open/close states using RBFNN to
isolate the faulted zone. The final step is for restoring the network to its normal
operation for the case of temporary faults.

http://www.bepress.com/ijeeps/vol11/iss5/art3 4
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

Step 1 System modeling

Power flow and short circuit simulation

Dividing network into several zones

The actuating input

Step 2 Identify fault type based

on current normalization

Identify distance of
fault from all source
by RBFNN
Determine open or close
state of CBs for fault
isolation by RBFNN
Determine faulty line
by RBFNN

Determine CBs for reclosing

Step 3

Reclose with relevant CB

First time: fast mode
Second time: slow mode

No Is it a
clearing permanent
Temporary
fault
fault

Yes

Restoring network
With synchronize function

Figure 2: Outline of the proposed protection scheme

Published by Berkeley Electronic Press, 2010 5

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

Step 1: Data generation

Before executing the protection procedure, it is necessary to perform power flow

and short circuit analyses so as to generate training data for the RBFNNs. From
the simulation outputs, the currents flowing through the main source and all DGs
at the time of fault occurrence are selected as input variables for the RBFNN. The
fault distance from each source and the open/close states of the CBs are regarded
as outputs of the RBFNNs. Prior to the RBFNN implementation, the distribution
system is divided into several zones as shown in Figure 3. Network zoning is
done by considering locations and generation capacities of DGs as well as
network loads. A zone is said to have one DG source in which it starts from the
beginning of a feeder and extends to the end of the feeder such that the DG have
sufficient capacity to supply average power to the loads within the zone. For the
remaining network with no DG, zones are formed based on the feeder
configuration and the main source.

Supply

CB2

Zone 1 CB1 Zone 2 CB3 Zone 3

CB5

Zone 4 CB4 Zone 5 CB6 Zone 6

CB7 CB8 CB9

Zone 7 Zone 8 Zone 9

Figure 3: Distribution network divided into several zones

http://www.bepress.com/ijeeps/vol11/iss5/art3 6
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

As shown in Figure 3, these zones should be separated by breakers which

are able to repeatedly open or close on receiving a signal from the main relay
located in the substation. The breakers must also be equipped with check-
synchronization function. As for the main relay, it has to be computer-based so
that it is capable of storing and analyzing large data and able to communicate with
other devices like zone breakers and DG relays (Brahma and Girgis, 2004;
Javadian and Haghifam, 2008b).

Step 2: Determine fault type, location of fault and operating state of CBs

Step 2.1 Identifying the fault type based on current normalization

At normal operating conditions, the sum of current contribution from all sources
is equal to the total load current. When a fault occurs at any point, fault current
will be significantly larger than the total load current. Thus, a comparison
between the total currents of generators and loads can be used for the detection of
fault conditions. To identify the various fault types, the 3 phase currents of the
main source from the feeding substation are used. The three phase output fault
currents at the main source or the feeding substation are normalized using,

I
I (4)
normal I
max

where I is the fault current and Imax is the maximum fault current for each type of
fault.
Based on the normalized three phase fault currents and rounding the
obtained values to the nearest one, the types of faults can be classified as shown in
Table 1 (Gers and Holmes, 2005). From the table, “1”, “-1” and “0”, indicate
that a fault occurs in the phase, a fault occurs in the phase but the short circuit
current is in the opposite direction and no fault, respectively. The symbols Ag, Bg
and Cg indicate the single phase to ground faults for phase A, B and C,
respectively while symbols AB, AC and BC indicate the phase to phase faults for
the respective phases. Consequently, symbols ABg, ACg and BCg indicate 2
phase to ground faults for the respective phases.

Published by Berkeley Electronic Press, 2010 7

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

Table 1: Fault Type Classification Data

Fault type Ia Ib Ic
Ag 1 0 0
1-phase to ground Bg 0 1 0
Cg 0 0 1
AB 1 -1 0
phase to phase AC 1 0 -1
BC 0 1 -1
ABg 1 1 0
2-phase to ground ACg 1 0 1
BCg 0 1 1
3 -phase ABC 1 1 1

Step 2.2 Identifying fault location using RBFNN

After identifying the fault type, its location should be determined. In this study,
two stages of RBFNN have been developed in which the first RBFNN is for
determining fault distances from the main source and the all DG units (RBFNN1,
4, 7, 10) and the second RBFNN (RBFNN 2, 5, 8, 11) is for determining the
faulty line for the respective fault types. Figure 4 shows the procedures in
determining the fault location. From the figure, for each fault type, the 3 phase
currents of the main source and all the DGs are used as inputs to the first RBFNN.
The outputs of the first RBFNN which are the distances of fault from the main
source and the DGs are then used as inputs to the second RBFNN. Hence, the
output of the second RBFNN is the exact faulty line.

Step 2.3 Determine open or close states of CBs for fault isolation by RBFNN

After identifying the fault type and fault location, we need to determine which
CBs must operate for isolating this fault. In this work, a RBFNN has been
developed to determine the open and close states of CBs for isolating the faulty
zone. Therefore, fault currents of the main source and all DGs are taken as inputs
for training RBFNN 3, 6, 9and 12 as shown in Fig. 4. When a fault occurs in a
distribution network, after recognizing the fault type, the RBFNN corresponding
to a particular fault type is activated to determine the open/close states of CBs for
isolating the faulty zone. Hence, the disconnection signal is sent to those CBs
placed in faulty zone. In addition, to cut off supply of faulty zone in the case that
there are DGs inside that zone, relay sends a disconnection command to all DGs

http://www.bepress.com/ijeeps/vol11/iss5/art3 8
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

inside that zone. The RBFNN implementation procedures are as illustrated in Fig.
4.

Start

IS=SC-Current of main source

IDGs=SC-Current of all DGs

IS IS
IDGs

Normalization by Eq.(4)

No
1 phase to
ground fault

Yes

RBFNN 1 No
2 phase fault
RBFNN 3

Yes
RBFNN 2

RBFNN 4 2 phase to No
ground fault
RBFNN 6
Yes
RBFNN 5

RBFNN 7 3 phase fault

RBFNN 9
Yes

RBFNN 8

RBFNN 10

RBFNN 12

RBFNN 11

Faulty line Open / close CBs

End

Figure 4: RBFNN implementations for fault location and fault isolation

Published by Berkeley Electronic Press, 2010 9

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

Step 3: Restoration for the case of temporary faults

Once the faulty line and the DG connected to it are isolated, the next step is to
implement restoration in case of temporary faults. The main relay at the
substation control performs reclosing action on the isolating CB which is in the
faulted zone. After each reclosing action of the isolating CB, the relay which
monitors current contribution from the main source continuously would detect
whether a fault persists. In case a temporary fault disappears, the relay would send
closing signals to all the CBs which then perform synchronized closing using the
check-synchronizing function incorporated with each breaker. Relay would then
sense the closing of all the circuit breakers connected to the faulted zone. Once all
the circuit breakers are closed, the DG circuit breaker would also be closed and
this allows the system to be restored to normal operation. In case of a permanent
fault, the fault would have to be cleared by maintenance personnel and the relay
will adapt to the new network configuration. There is a possibility here that some
parts of the system may be islanded after disconnection of one zone.
To illustrate the reclosing procedure, Figure 3 is referred to in which if a
temporary fault occurs in zone 5, relay first sends disconnecting signal to CB4,
CB5, CB6, CB8 and all DGs located in zone 5 and then sends reclosing signal to
CB5 to diagnose temporary fault. In the end, after clearance of fault, relay sends
closing signals along with synchronization to CB4, CB6, CB8 and the all DG
circuit breakers located in zone 5 to restore the network to normal.

4. Simulation Results and Discussion

To verify the performance and accuracy of the proposed protection method using
the RBFNN, the 22 bus, 20 kV distribution network with 2 DG units is selected as
the test system as shown in Figure 5. The test system data can be found in
(Zayandehroodi et al., 2010a).
Fig. 6 shows the single line diagram of the test system. The network
zoning is applied by dividing the network into several protection zones. The
system has four zones after performing the zoning procedure in which two of the
zones have a DG unit as shown in Figure 6. The zones are isolated from each
other by CB1, CB2, CB3 and CB4.

http://www.bepress.com/ijeeps/vol11/iss5/art3 10
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

Figure 5: Single line diagram of the test system

Published by Berkeley Electronic Press, 2010 11

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

Figure 6: Single line diagram of the test system after performing network zoning

The RBFNNs for the fault location and protection scheme were
implemented using the MATLAB software and the training data for the RBFNNs
were generated using the DIgSILENT Power Factory 14.0.516 software by
simulating various types of faults created at every 100 meter of each line. The
target (output) vector of the RBFNNs which is the fault distance from each source
and number of CBs which must open for isolating fault zone are obtained from

http://www.bepress.com/ijeeps/vol11/iss5/art3 12
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

the simulations. Table 2 summarizes the description of the inputs and outputs of
the training data for the developed RBFNNs.
Table 2: Description of inputs and outputs of the RBFNN

About 3360 training and testing data sets have been generated with 80% of
the data sets used for training the RBFNNs, and 20% are used for testing to
evaluate the performance of the RBFNNs. Table 3 shows the accuracy of the
trained RBFNNs.

Table 3: Training performances of the RBFNNs

Mean square error
Fault type RBFNN Goal
(MSE)
RBFNN 1 0.0001 5.58e-005

1-phase to ground RBFNN 2 0.0001 3.93e-005

RBFNN 3 0.01 5.79e-002

RBFNN 4 0.0001 7.83e-005

Phase to phase RBFNN 5 0.0001 3.93e-005

RBFNN 6 0.01 7.42e-002

RBFNN 7 0.0001 9.07e-005

2 phase to ground RBFNN 8 0.0001 6.58e-005

RBFNN 9 0.01 7.52e-002

RBFNN 10 0.0001 7.85e-005

3 phase RBFNN 11 0.0001 4.78e-005

RBFNN 12 0.01 4.81e-002

Published by Berkeley Electronic Press, 2010 13

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

The trained RBFNNs are then tested to evaluate its performance in

locating various fault types, including single phase to ground fault (1Ph-G), phase
to phase fault (2Ph), two phases to ground fault (2Ph-G) and three phase fault
(3Ph). These faults have been generated at various lengths of the distribution lines
of the test system. Table 4 shows some samples of the RBFNN testing results in
which RBFNN 1, 4, 7 and 10 predicts fault locations in terms of distances (in km)
from the main power source (DS) and the two DG units (DDG1 and DDG2). Then
RBFNN 2, 5, 8 and 11 predicts the faulty line while RBFNN 3, 6, 9 and 12
predicts the number of CBs which must be opened for isolating the faulted zone.
From the table, ‘1’ and ‘0’, indicate that a CB is close and open, respectively.
The actual values for distances in km, faulty line and CB states which are
obtained from simulations are also given in the table.
From Table 4, it is shown that the RBFNNs give accurate results in which
the maximum error of the first stage RBFNN in terms of the difference between
the actual and estimated distances of fault from the main source and 2 DG units is
about 0.01 km. Since each distribution line section is 1 km in length in the
studied network, a deviation of 0.01 km is considered acceptable. The second
RBFNN outputs after rounding to the nearest number shows the exact number of
the faulty line. For instance, when a single phase to ground fault occurs at 950
meter of line 1, the estimated output of the second RBFNN is 0.98 as shown on
the 1st row and 4th column of Table 4. After rounding to the nearest number, the
detected faulty line is line 1. The third RBFNN output also after rounding to the
nearest number shows the exact number of CB in terms of ‘0’ and ‘1’ for open or
close CB state.
The results in Table 4 also show the results of restoration for the case of
temporary faults by identifying the isolating CB that performs reclosing operation
and CBs that must be closed. For instance, when a single phase to ground fault
occurs at 950 meter of line 1 (the 1st row of Table 4), the first and second RBFNN
outputs are the faulty line no. 1 and all the CBs are in the open state, respectively.
In this situation, the relay sends disconnection commands to all the CBs and
zones, Z1 and Z4 will face power cut and zones, Z2 and Z3 will operate is
islanded mode. If the fault on line is permanent, all the CBs remain open.
However, if the fault on line 1 is temporary, the fault can be cleared due to
reclosing operation of the CB1. The network can be restored to its normal
operation, by sending connection signal to CB1, which then causes CB1 to close
(1). Due to network synchronization operation, CB2, CB3 and CB4 will be in
close states and this restores the network to normal operation.

http://www.bepress.com/ijeeps/vol11/iss5/art3 14
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

Table 4: Testing Performance of the RBFNNs

Identify fault location Isolation of faulted zone Restoration

Identify
type
Testing data

RBFNN
RBFNN 1, 4, 7, 10 RBFNN 3, 6, 9, 12 For temporary fault
2, 5, 8, 11

Fault DS DDG1 DDG2 Faulty Recloser Close

CB1 CB2 CB3 CB4
type km km km line no. ‘1’ ‘1’

CB2
1 Ph-G 0.960 2.060 7.055 0.98 0 0 0 0 CB1 CB3
CB4
CB2
950 meter of line 1

2 Ph 0.960 2.061 7.059 0.91 0 0 0 0 CB1 CB3

CB4
CB2
2 Ph-G 0.950 2.053 7.060 1.02 0 0 0 0 CB1 CB3
CB4
CB2
3 Ph 0.961 2.057 7.057 1.07 0 0 0 0 CB1 CB3
CB4
Actual 0.950 2.050 7.050 1 0 0 0 0
DG1
1 Ph-G 2.347 0.643 8.351 3.03 1 0 1 0 CB2
CB4
350 meter of line 3

DG1
2 Ph 2.352 0.657 8.344 3.01 1 0 1 0 CB2
CB4
DG1
2 Ph-G 2.361 0.656 8.346 3.09 1 0 1 0 CB2
CB4
DG1
3 Ph 2.355 0.647 8.351 3.05 1 0 1 0 CB2
CB4
Actual 2.350 0.650 8.350 3 1 0 1 0

1 Ph-G 4.560 1.562 10.554 4.96 1 1 1 0 CB4 CB4

560 meter of line 5

2 Ph 4.559 1.555 10.553 4.92 1 1 1 0 CB4 CB4

2 Ph-G 4.560 1.559 10.561 5.02 1 1 1 0 CB4 CB4

3 Ph 4.558 1.556 10.556 5.08 1 1 1 0 CB4 CB4

Actual 4.560 1.560 10.560 5 1 1 1 0

Published by Berkeley Electronic Press, 2010 15

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

670 meter of line 7 1 Ph-G 6.670 3.671 12.666 7.02 1 1 1 0 CB4 CB4

2 Ph 6.669 3.668 12.666 6.98 1 1 1 0 CB4 CB4

2 Ph-G 6.670 3.675 12.661 7.00 1 1 1 0 CB4 CB4

3 Ph 6.668 3.671 12.666 6.97 1 1 1 0 CB4 CB4

Actual 6.670 3.670 12.670 7 1 1 1 0

CB2
1 Ph-G 2.851 3.851 5.154 9.01 0 0 0 0 CB1 CB3
CB4
CB2
850 meter of line 9

2 Ph 2.839 3.841 5.149 8.91 0 0 0 0 CB1 CB3

CB4
CB2
2 Ph-G 2.848 3.845 5.152 8.98 0 0 0 0 CB1 CB3
CB4
CB2
3 Ph 2.842 3.843 5.145 8.93 0 0 0 0 CB1 CB3
CB4
Actual 2.850 3.850 5.150 9 0 0 0 0
DG1
1 Ph-G 4.302 1.300 10.305 15.04 1 0 1 0 CB2
CB4
300 meter of line15

DG1
2 Ph 4.301 1.295 10.299 15.02 1 0 1 0 CB2
CB4
DG1
2 Ph-G 4.301 1.303 10.299 15.01 1 0 1 0 CB2
CB4
DG1
3 Ph 4.299 1.296 10.301 14.98 1 0 1 0 CB2
CB4
Actual 4.300 1.300 10.300 15 1 0 1 0

1 Ph-G 5.415 6.414 2.582 19.05 1 1 0 1 CB3 DG2

420 meter of line19

2 Ph 5.412 6.412 2.581 18.91 1 1 0 1 CB3 DG2

2 Ph-G 5.419 6.419 2.578 19.04 1 1 0 1 CB3 DG2

3 Ph 5.413 6.414 2.579 18.98 1 1 0 1 CB3 DG2

Actual 5.420 6.420 2.580 19 1 1 0 1

http://www.bepress.com/ijeeps/vol11/iss5/art3 16
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

570 meter of line21 1 Ph-G 7.555 8.541 0.431 21.03 1 1 0 1 CB3 DG2

2 Ph 7.558 8.551 0.431 20.98 1 1 0 1 CB3 DG2

2 Ph-G 7.560 8.561 0.430 21.09 1 1 0 1 CB3 DG2

3 Ph 7.562 8.553 0.427 21.01 1 1 0 1 CB3 DG2

Actual 7.550 8.550 0.430 21 1 1 0 1

5. Conclusion
An automated protection scheme for a distribution system with DG units using
RBFNN has been presented. In the proposed method, the fault type is first
determined by normalizing the fault currents of the main source. To determine the
fault location, two staged RBFNNs have been developed for various fault types.
The first RBFNN is used for determining the fault distance from each power
source and the second RBFNN is used for identifying the exact faulty line. For
fault isolation, the third stage RBFNN is also developed for determining the open
and close states of the circuit breakers. A test distribution system with two DG
units has been selected to verify the proposed protection method. The test results
showed that the proposed procedure can accurately determine the location of fault
and the CBs states for isolating the faulty zone in the system. Such a protection
method would be useful for assisting power engineers in performing service
restoration quickly and automatically. Hence, the proposed intelligent protection
method for a system with DGs can increase network reliability; and decrease the
total down time of the system.

References

Abdelaziz, A. Y., et al. (2002). "An adaptive protection scheme for optimal
coordination of overcurrent relays." Electric Power Systems Research
61(1): 1-9.
Barker, P. P. and R. W. De Mello (2000). Determining the impact of distributed
generation on power systems. I. Radial distribution systems. Power
Engineering Society Summer Meeting, 2000. IEEE.

Published by Berkeley Electronic Press, 2010 17

 International Journal of Emerging Electric Power Systems, Vol. 11 [2010], Iss. 5, Art. 3

Brahma, S. M. and A. A. Girgis (2002). Microprocessor-based reclosing to

coordinate fuse and recloser in a system with high penetration of
distributed generation. Power Engineering Society Winter Meeting, 2002.
IEEE.
Brahma, S. M. and A. A. Girgis (2004). "Development of adaptive protection
scheme for distribution systems with high penetration of distributed
generation." IEEE Transactions on Power Delivery 19(1): 56-63.
Burke, J. J. (1994). Power distribution engineering: fundamentals and
applications, CRC.
Doyle, M. T. (2002). Reviewing the impacts of distributed generation on
distribution system protection. Power Engineering Society Summer
Meeting, 2002 IEEE.
Gers, J. and T. Holmes (2005). Protection of electricity distribution networks,
Peter Peregrinus Ltd.
Hadjsaid, N., et al. (1999). "Dispersed generation impact on distribution
networks." Computer Applications in Power, IEEE 12(2): 22-28.
Javadian, S. A. M. and M. R. Haghifam (2008a). Implementation of a New
Protection Scheme on a Real Distribution System in Presence of DG.
Power System Technology and IEEE Power India Conference, 2008.
POWERCON 2008. Joint International Conference on.
Javadian, S. A. M. and M. R. Haghifam (2008b). Protection of distribution
networks in presence of DG using Distribution Automation System
capabilities. Power and Energy Society General Meeting - Conversion and
Delivery of Electrical Energy in the 21st Century, 2008 IEEE.
Javadian, S. A. M., et al. (2009a). A fault location and protection scheme for
distribution systems in presence of dg using MLP neural networks. Power
& Energy Society General Meeting, 2009. PES '09. IEEE.
Javadian, S. A. M., et al. (2009b). Determining fault's type and accurate location
in distribution systems with DG using MLP Neural networks. Clean
Electrical Power, 2009 International Conference on.
Jenkins, N. (1995). "Embedded generation." Power Engineering Journal 9(3):
145-150.
Kauhaniemi, K. and L. Kumpulainen (2004). Impact of distributed generation on
the protection of distribution networks. Developments in Power System
Protection, 2004. Eighth IEE International Conference on.
Kumpulainen, L. K. and K. T. Kauhaniemi (2004). Analysis of the impact of
distributed generation on automatic reclosing. Power Systems Conference
and Exposition, 2004. IEEE PES.
Rezaei, N. and M. R. Haghifam (2008). "Protection scheme for a distribution
system with distributed generation using neural networks." International
Journal of Electrical Power & Energy Systems 30(4): 235-241.

http://www.bepress.com/ijeeps/vol11/iss5/art3 18
DOI: 10.2202/1553-779X.2611
Zayandehroodi et al.: A New Protection Scheme for Distribution Network with DG Using RBFNN

Tran-Quoc, T., et al. (2003). Technical impacts of small distributed generation

units on LV networks. Power Engineering Society General Meeting, 2003,
IEEE.
Yu, L., et al. (2008). "Multistage RBF neural network ensemble learning for
exchange rates forecasting." Neurocomputing 71(16-18): 3295-3302.
Zayandehroodi, H., et al. (2009). An Overview of Protection Coordination
Methods in Distribution Network with DGs. International conference
electrical energy and industrial electronic systems (EEIES 2009). Penang,
Malaysia.
Zayandehroodi, H., et al. (2010a). "Automated Fault Location in a Power System
with Distributed Generations Using Radial Basis Function Neural
Networks." International Journal of Applied Sciences 10(23): 3032-3041.
Zayandehroodi, H., et al. (2010b). Performance Comparison of MLP and RBF
Neural Networks for Fault Location in Distribution Networks with DGs.
IEEE International Conference on Power and Energy (PECon 2010),
Kuala Lumpur, Malaysia.

Published by Berkeley Electronic Press, 2010 19