With increasing reliance on electricity, customers demand reliable power supply
with reduced outage time and operating costs. When a fault occurs in a distribution
network, it is important to quickly locate the fault by identifying either a
faulty bus or a faulty line section in the network. Without locating the faulty
section, no attempt can be made to remove the faults and restore the power supply.
Fault location in electric power distribution systems still presents many challenges
due to its varied topological and operational characteristics. The traditional
methods used for locating faults in a distribution network include visual inspection,
traveling wave method and the use of utility outage management system to identify
circuit outages. However, these traditional fault location methods are unable
to locate faults quickly. To locate fault location, voltage and current values
are generally measured using intelligent electronic devices installed at substations.
Although, these data loggers can capture the fault events precisely, they are
not equipped with fault diagnostic algorithms, which can quickly identify fault
location. To solve this problem, an automated fault location using intelligent
data interpretation system is required (Mohamed and Mazumder,
Several artificial intelligence techniques have been developed for automated
fault location in distribution systems (Jun et al.,
1997; Wen and Chang, 1998; Mohamed
and Mazumder, 1999; Chien et al., 2002; Fei
and Ying, 2003; Brahma, 2005; Senger
et al., 2005; Borghetti et al., 2006).
The intelligent based fault location methods locate faults by calculating the
fault distances, identifying the faulted phases and locating the faulty protective
devices. These methods, however, do not consider distribution networks with
Distributed Generation (DG). From a technical viewpoint, the presence of distributed
generators in a distribution network would result in some conflicts in the operation
of the present network because distribution network configuration is no longer
radial in structure. The conventional fault location scheme is suitable for
locating faults in a system with a single source radial supply line or with
multi-source open loop operation. With the increasing penetration of DG, the
distribution system becomes a multi source system and the system configuration
is not radial. In such network, determining the exact location of faults is
becoming complicated, as faults are fed by multi-sources. Hence, the existence
of DGs in a distribution network poses a difficulty in locating faults in the
Recently, fault location methods have been developed by considering DGs in
a distribution network (Ma et al., 2008). A fault
location algorithm for a distribution system with DGs has been developed by
using current measurements (Guo-Fang and Yu-Ping, 2008a,
b). In this method, after a faulted segment is located,
islands are formed involving groups of DGs and a load shedding scheme is implemented
to match the loads with the DGs generating capability in the island. This method
requires a mechanism to reconnect the disconnected loads after faults are removed.
A method for finding the exact location of faults in a network with DG has been
developed based on software procedures which require a telecommunication control
system (Conti and Nicotra, 2009). Another fault location
method is based on the estimates of the fault impedance by measuring current
and voltage at a substation (El-Fouly and Abbey, 2009).
In this method, the fault location performance is inaccurate when a DG is located
upstream of the fault section where the impact is more severe for synchronous
machine based DG. An asymmetric fault location method using a communication
system has been developed by identifying the direction of an asymmetrical fault
based on negative sequence current scalar product (Du et
al., 2009). Here, an asymmetrical fault line searching and locating
scheme is implemented by combining the fault direction distinguishing method
with its communication system. A more recent fault location method for a distribution
network with DGs considers the application of Multi Layer Perceptron Neural
Network (MLPNN) (Rezaei and Haghifam, 2008; Javadian
et al., 2009a, b). However, considering the
structure and training algorithm of the MLPNN, the speed of this method is not
suitable for fast and accurate fault location.
The aim of the research is to develop an accurate and automated fault location method for a distribution network with distributed generators by identifying the faulty line. This study presents an automated fault location scheme for a distribution network with DGs using the Radial Basis Function Neural Network (RBFNN). RBFNN is considered to be a better neural network model for solving engineering problems. The proposed scheme determines the fault type by normalizing the fault current of the main source whereas the location of faults is determined by using two RBFNNs. The first RBFNN determines the fault distance from each DG and the main source while the second RBFNN identifies the exact faulty line.
RBF NEURAL NETWORK
The RBF Neural Network (RBFNN) is a feed-forward neural network having three
layers, namely, an input layer, hidden layer and output layer.
|| A generic architecture of the RBFNN
The input layer feeds the values to each of the neurons in the hidden layer.
The hidden layer consists of neurons with radial basis activation functions
and an output layer consists of neurons with linear activation function. A generic
architecture of an RBFNN with k input and m hidden neurons is shown in Fig.
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:
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:
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.
The Euclidean norm is normally used to calculate the distance, Di which is given as:
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:
||Obtain input data and target data from the simulation
||Assemble and preprocess the training data for the RBFNN
||Create the network object and train the network until condition of network
setting parameters are reached
||Test and conduct regression analysis
||Stored the trained network. Steps (1-5) are offline processes
||Preprocess the new input before they are subjected to the trained network
to obtain required data
IMPLEMENTATION OF FAULT LOCATION IN A DISTRIBUTION NETWORK WITH DG USING RBFNN
An important consideration in the protection of distribution networks is the determination of the type and location of faults occurring in its protection zone. In this work, normalized fault current of the main source is used for determining the various types of faults. On the other hand, the fault location in a distribution system with DGs has been developed using RBFNN.
Identification of fault type: 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:
where, I is the fault current and Imax is the maximum fault currents 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.
|| Fault type classification data
Consequently, symbols ABg, ACg and BCg indicate 2 phase to ground faults for
the respective phases.
Determination of fault location: 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 determinings fault distances from the main source and the two DG units (RBFNN1, 3, 5, 7) and the second RBFNN (RBFNN 2, 4, 6, 8) is for determining the faulty line for the respective fault types. Figure 2 shows the procedures in determining the fault location. From the figure, for each fault type first, the 3 phase currents of 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.
SIMULATION RESULTS AND DISCUSSION
To verify the performance and accuracy of the proposed fault location method
using the RBFNN, the 22 bus, 20 kV distribution network with 2 DG units is selected
as the test system as shown in Fig. 3. The system consists
of a 4.5 MVA diesel generator as DG1 connected to bus 4 and a 3.5 MVA diesel
generator as DG2 connected to bus 22 (Javadian et al.,
2009a). The test system data are given in the Appendix.
The RBFNN for fault location has been implemented using the MATLAB software and the training data for the RBFNNs have been generated using the DIgSILENT Power Factory 14.0.516 software by simulating various types of faults created at any 100 m of each line. The target (output) vector of the RBFNNs which is the faulty line is obtained from the simulations. Table 2 summarizes the description of the inputs and outputs of the training data for the developed RBFNNs.
|| Flow chart of the proposed fault location method
|| Description of inputs and outputs of the RBFNN
About 756 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 in which the mean square error (MSE) is less than 10x10-5
and the maximum training epoch is 38. During the RBFNN training, the target
MSE value is set to 0.0001.
The trained RBFNNs are then tested to evaluate its performance in locating
various fault types including single phase to ground fault, phase to phase fault,
two phase to ground fault and three phase fault. These faults have been generated
at different lengths of the distribution lines of the test system.
|| Training performances of the RBFNNs
Table 4 shows some samples of the RBFNN testing results in
which RBFNN 1, 3, 5 and 7 predicts fault locations in terms of distances from
the main power source and the two DG units while RBFNN 2, 4, 6 and 8 predicts
the faulty line.
|| Single line diagram of the test system
|| Testing performance of the RBFNNs for locating faults
Thus, the location of faults is predicted by identifying the line in the system
where a fault occurs.
From Table 4, it is shown that the RBFNNs give accurate results in which the maximum error of the first RBFNN which is the difference between the actual and estimated distances of fault from the main source and all DGs is about 0.001 km. Since, each distribution line section is 1 km in length in the studied network, a deviation of 0.001 km is considered acceptable. The second RBFNN outputs after rounding to the nearest one shows the exact number of faulty lines. For instance, when a single phase to ground fault occurs at 950 m 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 one, the detected faulty line is line 1. The results in Table 4 also show that the second RBFNN outputs give accurate prediction of the faulty lines when compared to the actual faulty lines.
Considering more recent fault location method for a distribution network with
DGs considers the application of MLPNN (Rezaei and Haghifam,
2008; Javadian et al., 2009a, b).
Hence, to further evaluate the effectiveness of the fault location method using
RBFNN, it is compared with the results of using the MLPNN. Table
5 shows the training performance of the MLPNN in which the maximum MSE value
and training epoch of the MLPNN for estimating the fault distances are 0.0001
and 120, respectively. Comparing the MLPNN and RBFNN training performances,
in which the maximum training epoch for RBFNN and MLPNN are 38 and 120, respectively,
it can be said that the RBFNN takes shorter time to achieve the required training
Figure 4 to 15 show the computed fault
distances from each source using RBFNN and MLPNN in comparison with the actual
values for single phase to ground, phase to phase, two phase to ground and three
phase faults, respectively. From the figures, it is clear that the RBFNNs give
outputs closer to the actual values compared to that of the MLPNN for all the
fault types. Hence, the results confirm the accuracy of the RBFNN based fault
|| Training performances of the MLPNN
||The fault distance from main source for 1- phase to ground
||The fault distance from DG1 for 1- phase to ground fault
||The fault distance from DG2 for 1- phase to ground fault
||The fault distance from main source for phase to phase fault
||The fault distance from DG1 for phase to phase fault
||The fault distance from DG2 for phase to phase fault
||The fault distance from main source for 2-phase to ground
||The fault distance from DG1 for 2-phase to ground fault
||The fault distance from DG1 with 2-phase to ground fault
||The fault distance from main source for 3-phase fault
||The fault distance from DG1 for 3-phase fault
||The fault distance from DG2 for 3-phase fault
An automated fault location method using RBFNN for a distribution system with DG units has been presented. In the proposed approach, the normalized fault currents of the main source are used for determining the fault type and two RBFNNs have been developed for determining the fault location in a test distribution system with DGs. A test distribution system with two DG units has been selected to verify the proposed fault location method. The test results showed that the proposed RBFNN based fault location method give accurate predictions of fault distances from the main source and DG units as well the faulty lines. Furthermore, the developed RBFNNs perform better than the MLPNN in terms of accuracy and training time. The main advantage of the proposed fault location method is that it can identify the exact faulty line. Such a fault location method would be useful for assisting power engineers in performing service restoration quickly and automatically. Hence, the proposed intelligent fault location method for a system with DGs can increase network reliability; and decrease the total down time of the system.
For each load a three-step hourly load curve is considered as shown in Fig. 16. The peak load for all loads is 1 MW and the power factor for all of them at each time is assumed 0.92 lagging.
|| Hourly load curve of the simulated feeder's loads
|| Technical data of distribution lines
|| Technical data of DGs
All the distribution conductors are of HYENA type with 1 km length and the technical information of the conductors is given in Table 6. The technical data of all the DGs is presented in Table 7.