A Proposed Technique to Alleviate Fuse-Recloser MisCoordination for Power Systems Penetrated with DG
Using Probabilistic Neural Network
Amr AbouGhazala
Asmaa Mousa
I. INTRODUCTION
The presence of distributed generation units connected to
power distribution networks is greatly increasing. Thus,
investment in distributed generation manufacturing and
integration with power systems has been increasing
exponentially in the last few years. Unfortunately, the
integration of DGs into current power distribution networks is
not yet problem free, as these networks were designed based
on the assumption that only loads, not generators, will be
connected to it. These problems may be classified into four
main categories: impacts on system protection, impacts on
voltage regulation, impacts on power quality, and
ferroresonance.
Before evaluation of these system impacts, data that can
properly describe both the DG and the utility system need to
be collected. Data needed to evaluate DG impacts are [1]:
Size rating of the proposed DG
Type of DG power converter (static or rotating)
Type of DG prime energy source (photovoltaic, wind or
fuel cell)
Operating cycles
17th International Middle East Power Systems Conference, Mansoura University, Egypt, December 15-17, 2015
1. Limit the first peak of the fault current.
2. Exhibits low impedance and low energy losses in normal
state.
3. Does not generate unacceptable harmonics in the normal
state.
4. Eliminate sensors and control devices if their reliability
compromises the overall reliability of the system.
5. Exhibits a smooth and gradual change of impedance from
the normal mode to fault mode and vice-versa.
6. Compactness.
7. Fail-safe operation.
8. Zero reset time.
Figure (2) shows that the fuse coordinates with the recloser
for temporary fault if the fault current is less than 508 A.
The short circuit simulation of this radial system shows that
fuse current (Ifuse=307 A) is nearly equal to recloser current
(Irec.=319 A), which is less than 508 A. Hence the
coordination between fuse and recloser is well achieved.
17th International Middle East Power Systems Conference, Mansoura University, Egypt, December 15-17, 2015
the previous equations, the DG current equals 106 A. hence
the DG size which maintain the fuse recloser coordination is
less than 2 MVA. It is important to note that this method
assumes that the substation current still constant, while the
substation current becomes lower due to DG sharing. This
means the situation is a little worse in real application.
If the previous system is imbedded with 5MVA or
10MVA, the simulation results can be summarized in table(1),
and coordination curves shown in figure(4)
TABLE I
SIMULATION RESULTS OF 5 MVA& 10 MVA DG PENETRATED SYSTEM
5MW DG
10MW DG
Recloser current
(Irec)
320A
315A
Fuse current
(Ifuse)
427A
504A
17th International Middle East Power Systems Conference, Mansoura University, Egypt, December 15-17, 2015
A Probabilistic Neural Network (PNN) consists of four
layers. One is the input layer where the actual input vector is
given. The second one is the pattern layer with one neuron for
every training example. Input layer and pattern layer are fully
connected. The next one is the summation layer, where the
results of the pattern layer will be added. In this layer we have
one neuron for each class we want to distinguish. The neurons
of the pattern layer are connected to the neurons of the
summation layer based on the class of the neuron in the pattern
layer. So every neuron in the pattern layer of the same class is
connected to the same neuron in the summation layer. The last
layer is the output layer where the estimated class of the input
data is shown. PNNs use exponential function instead of
sigmoid function which is used in back propagation networks.
There are several advantages for using PNN instead of back
propagation networks.PNNs are much faster than back
propagation networks. PNNs can be more accurate than back
propagation networks. On the other hand, PNNs are slower
than back propagation networks at classifying new cases and
PNNs require more memory space to store the model [6], [7].
The previous results show that only 2 of 38 tested data
wrongly detected by the neural network. The neural network
succeeded to detect the true decision for about 95% of the
tested data. Although results show high level of accuracy for
the usage ANN as a classifier, yet the level of generalization of
such network is doubtful. This is due to the limited number of
parameters which used in the training set. For the sake of
better generalization of ANN, more parameters are needed to
be included. This is discussed in the next section.
VII. ADVANCED ANN FOR COORDINATION
MAINTAINABILITY
For better generalization of the obtained results, more
advanced artificial neural network is constructed to cope with
other electrical power systems with different parameters. Many
varying parameters should be used to train the artificial neural
network. These parameters include the short circuit capacity of
the power system at the point of study, main transformer rated
power, distribution transformer rated power, distributed
generator rated power, distributed generator relative location
and distributed generator type, See figure (6).
TABLE II
RECLOSER AND FUSE CURRENTS
case
Fig. 6 Advanced ANN for coordination maintainability
1
2
3
DG
unit
DG1
DG1
DG2
Fault
location
Fault 1
Fault 2
Fault 1
Recloser
current(Ir)
Is
IDG
Is+IDG
Fuse
current(If)
Is+IDG
Is+IDG
Is+IDG
17th International Middle East Power Systems Conference, Mansoura University, Egypt, December 15-17, 2015
Case1: In this case, the recloser time (fast curve) should be
less than fuse time by time margin (> 0.35 second) [8].
Case 2: In this case, to maintain the original
coordination
scheme, the recloser time (fast curve) should be less than fuse
time by time margin (> 0.35 second), where the recloser
current is smaller than fuse current.
Case 3: In this case before using the DG, the recloser
current nearly equals zero if the fault occurs in the lateral
feeder. To maintain the original coordination scheme, the
recloser time should be greater than the fuse time with time
margin (> 0.35 second).
6.
b. MATLAB./SIMULINK Program
MATLAB/SIMULINK is used to simulate the system with
varying parameters. Therefore the overall a set of training data
consists of 3(S.C level)*3(main transformer)*3(distribution
transformer)*
5(DG
size)*3(DG
location)
+3(S.C
level)*3(main transformer size)*3(distribution transformer
size)*3(DG size)*3(DG location) = 405 (for rotating machine)
+ 243 (for static source) = 648 cases
Each case has been simulated on MATLAB./SIMULINK
to calculate fuse and recloser currents. These currents have
been allocated on the relevant coordination curves to
determine which action is needed to be taken to preserve
system coordination. As mentioned before, three different
actions may be expected. These are; no action, recloser resetting or using fault current limiter. These three outcomes
have been assigned values of 3, 1 and 2 respectively. The
MATLAB code which used to train the ANN follows the same
steps of the program used in the previous section.
C. RESULTS
The ANN failed to detect the true decision for only 5 cases
out of 129 cases. The ANN succeeded in 96% of the examined
cases. The cases which the ANN failed to detect the true
corrective decision are as shown in table (3).
TABLE III
COMPARISON BETWEEN THE TRUE CORRECTIVE DECISION AND ANN DECISION
Case 1
Case 2
Case 3
Case 4
Case 5
ANN decision
(output)
1
3
2
2
3
Case 1
Case 2
Case 3
Case 4
Case 5
ANN decision
(output)
2
2
2
2
2
17th International Middle East Power Systems Conference, Mansoura University, Egypt, December 15-17, 2015
X. CONCLUSION
The penetration of DGs to distribution networks greatly
impacts the distribution networks performance and reliability.
Among the various problems that associate the DG integration
with general utilities, is fuse-recloser mis-coordination. To
overcome this problem, if occurred, this study suggested two
remedial actions which are reset recloser or use fault current
limiter. For better utilization of the available remedal actions,
a particular selected class of neural networks has been chosen
to help the system operator to maintain the reliable operation
of the power systems penetrated with distributed generators.
This technique will decide with upto 96% success rate the
suitable method to restore system total coordination.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]