Beruflich Dokumente
Kultur Dokumente
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
1
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
2
at the substation. Fig. 1. Integrated intelligent algorithm for fault location and PQ analysis.
• Fast and accurate diagnosis of the fault type and power
A. Module 1: Fault Detection Algorithm
quality events through the combined use of the techniques
such as Wavelet Transform (WT), Multiresolution The wavelet transform (WT) is a mathematical tool similar
Analysis (MRA) and Fuzzy-ARTMAP Artificial Neural to the Fourier Transform (FT) used for signal analysis, which
Network (ANN). can be used to extract important characteristics of different
• Geographic estimation of the faulty point based on analysis types of disturbances. WT is a modern signal processing
of electric circuits and voltage deviations, overcoming technique that overcomes the limitations of methods based on
multiple estimation problem presented by most techniques. FT. Usually, the WT of transient signals is expressed by a
• The integration of proposed algorithms in a modular design multiresolution decomposition fast algorithm to decompose
of the diagnostic system, which provides greater flexibility, the signal to components under different scales. It is equal to
precision and high reliability for the obtained results. successively filtering the signal with a high-pass and low-pass
Section II is devoted to intelligent algorithms for fault filter pair.
detection, feature extraction of behavioral indices, power The Fault Detection Algorithm is designed taking into
quality analysis, short circuit analysis, and fault location. The account the continuous monitoring of the current signal
diagnosis of fault type and power quality events present in the recorded from a measuring system present at the substation.
This module analyzes the input signal using a sampling rate of
waveforms is accomplished through the combined use of
wavelet transform concepts [21], multiresolution analysis [22] 7.68 kHz, having a fixed window of 128 samples/cycle with
and artificial neural networks [23]. The combination of such step of 1/4 cycle. This data window is decomposed into MRA
using the fourth-order Daubechies mother wavelet (db4). The
concepts in the design of the diagnostic system provides
analysis is performed for the first four levels of detail. Fig. 2
greater flexibility, accuracy and high reliability of the obtained
shows the flowchart of the disturbances detection algorithm.
results. Section III provides tests results for all the algorithms
Begin
using a model of an actual distribution network in Brazil.
Conclusions and references are given at the end. Input Data Acquisition
(Vabc, Iabc)
𝑘
𝑖𝑖 max�𝑣𝑣𝑣𝑖𝑖 � ≤ 3𝜎𝑖𝑖 𝑜𝑜 max��𝑑𝑖𝑖 �� ≤ 𝜆𝑖𝑖 ⟶ 𝜆𝑖𝑖 = 0 (2)
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
3
deviation of the sampled signal, the algorithm seeks to know Pre-fault and During-fault Energy
Calculation (Epij, Edij)
the general behavior of the signal, comparing the maximum
variance of the detail coefficients (varij) to a percentage of the Diference Calculation ( Epij - Edij ) Ratio Calculation ( Edij / Epij )
standard deviation (σij), providing greater immunity to the Normalization ( Nij ) Normalization ( Xij )
coefficients (dijk), a punctual analysis is performed, facing the Axis Translation Axis Translation
highest detail coefficient with a predetermined threshold value and Codification and Codification
(γj), defined, and based on the operating characteristics of each ARTMAP-Fuzzy Neural ARTMAP-Fuzzy Neural
Network Network
system. Fig. 3 shows a voltage sag, followed by the first (D1),
Diagnostic: Diagnostic:
second (D2) and third (D3) levels of detail, decomposed by Power Quality Events Short-circuit
Short-circuit
MRA. It can be seen that the three first levels of detail indicate Power Quality
Analysis Module Analysis Module
disturbance through the illustrated peaks. Fig. 4. Flowchart of the disturbances classification algorithm.
B. Feature Extraction 𝑌1𝑗 = 𝑋𝐼𝑎 𝑗 − 𝑋𝐼𝑏 𝑗 , 𝑌2𝑗 = 𝑋𝐼𝑏 𝑗 − 𝑋𝐼𝑐 𝑗 , 𝑌3𝑗 = 𝑋𝐼𝑐 𝑗 − 𝑋𝐼𝑎𝑗 (4)
According to Parseval's theorem [24], the energy E of a 𝑌4𝑗 = 𝑋𝑉𝑎𝑗 − 𝑋𝑉𝑏 𝑗 , 𝑌5𝑗 = 𝑋𝑉𝑏 𝑗 − 𝑋𝑉𝑐𝑗 , 𝑌6𝑗 = 𝑋𝑉𝑐𝑗 − 𝑋𝑉𝑎𝑗 (5)
discrete signal x [n] can be calculated in both time domain and
Where: 𝑋𝑖𝑗 is the normalized value of the ratio between the
frequency domain, as in (3).
𝑁𝑗
post-fault 𝐸𝑖𝑖𝑑 and pre-fault 𝐸𝑖𝑖𝑃 energies of the oscillographic
record i in the level of decomposition j. These representative
𝐸𝑖𝑖 = �|𝑥𝑖 [𝑛]|2 (3)
indices lie between -1 and +1. In order to be used as input of
𝑛=1
the Fuzzy-ARTMAP neural networks they need to be
Where: Nj means the number of samples in the decomposition
translated to the domain Ω = [0, +1].
level j, 𝑥𝑖 [𝑛] represents the n-th sample coefficient of detail or
𝑇
𝑌𝑛𝑛 + 1
approximation and, i is the oscillographic record analyzed, i.e., 𝑌𝑛𝑛 = (6)
2
Ia, Ib, Ic, Va, Vb or Vc. Where: n represents the behavioral indices in analysis, i.e., n =
1, 2, 3, 4, 5 or 6. Thus, the Fuzzy-ARTMAP neural networks
responsible for the short-circuit classification have as input
some vectors composed by the behavioral indices YnjT , which
characterize accurately the phases involved in the disturbance.
2) Representation of the Voltage Wavelet Indices
The Wavelet indices Wij are represented by the normalized
and translated values of the difference between the pre-fault
𝐸𝑖𝑖𝑃 and post-fault 𝐸𝑖𝑖𝑑 energies of the oscillographic record i in
the level of decomposition j:
𝑃 𝑑
𝐸𝑖𝑖 − 𝐸𝑖𝑖
+1
∆𝑚𝑚𝑚
𝑉𝑥 𝑗 (7)
𝑊𝑖𝑖 =
2
Where: ∆𝑚𝑚𝑚𝑉𝑥𝑗 represents the maximum value related to the
analysis of the voltage signals in the decomposition level j, i.e,
Fig. 3. MRA decomposition of a voltage sag.
∆𝑚𝑚𝑚 𝑚𝑚𝑚 𝑚𝑚𝑚 𝑚𝑚𝑚
𝑉𝑥𝑗 = 𝑚𝑚𝑚�∆𝑉𝑎𝑗 , ∆𝑉𝑏 𝑗 , ∆𝑉𝑐𝑗 � and, x = a, b or c.
As signals with power quality disturbances have large Thus, the Fuzzy-ARTMAP neural networks responsible for
imbalances with different frequency components, the the classification of power quality events have as input some
calculation of signal energy can be used to extract important vectors composed by the wavelet indices, which characterize
characteristics of different types of disturbances. Thus, by satisfactorily the power quality events.
calculating the energy of the signals pre-fault and post-fault, it
C. Module 2: Power Quality Analysis
is possible to quantify the degree of imbalance in the
operational status of the system. This must be a good tool to The Fuzzy-ARTMAP neural networks are self-organizing
evaluate, discreetly, non-stationary signals. Fig. 4 presents the networks presenting supervised training, capable to
flowchart of the disturbances classification algorithm. identify/classify data in clusters. This architecture presents
two key characteristics, stability and plasticity, which enable
1) Representation of the Behavioral Indices. its application in modern electric power systems. These
A relative representation between phases is employed for attributes enable continuous training to be incorporated,
the representation of the behavioral indices [18]. provide fast diagnosis and stable learning [16].
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
4
The PQ analysis module is responsible for the evaluation minimum set of points placed along the feeder in order to
and classification of events recorded in the voltages calculate the voltage deviations. These waveforms are
waveforms. Its input vectors are the voltage wavelets indices obtained through the usage of smart meters, protective and
calculated in the previous subsection. The six levels of detail control equipment with intelligent sensors such as automatic
and the first and sixth levels of approximations of the wavelets switchgear, reclosers or devices equipped with remote
indices were used as input data to compose the input vector ����
𝐼𝑥𝚤𝚤 terminal units (RTUs).
for the ANNs. Thus, the input vector ���� 𝚤𝚤
𝐼𝑥 has dimension 16, Fig. 5 shows the flowchart of the fault location algorithm
being composed of these eight wavelet indices approaches developed based on the algorithms of backward/forward
plus its complement. sweep power flow [25] and short-circuits analysis [26], which
𝐼����
𝚤𝚤 𝑐
𝑥 = [𝐼𝑥 𝐼𝑥 ] (8) is proven to be efficient and appropriate for radial or weakly
Where: meshed distribution networks, even in the presence of
𝐼𝑥 = [𝑊𝑉𝑥𝑑1 𝑊𝑉𝑥𝑑2 𝑊𝑉𝑥𝑑3 𝑊𝑉𝑥𝑑4 𝑊𝑉𝑥𝑑5 𝑊𝑉𝑥𝑑6 𝑊𝑉𝑥𝑎1 𝑊𝑉𝑥𝑎6 ] distributed generation.
𝐼𝑥𝑐 = 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑜𝑜 𝐼𝑥 Besides the measurement data, the algorithm needs a
database containing information about the network topology,
The output vector I xout used in the training phase of Fuzzy- i.e., impedances of the lines sections and the power rating of
ARTMAP neural networks is encoded. This encoding both the distribution transformers. Reference [17] explains in a more
input vector as well as the output vector is used for the three detailed way the fault location methodology. The main aspects
voltage oscillographic records recorded, i.e., for the three of the FL algorithm are presented in the next subsection.
phases of the system: Va, Vb and Vc. Begin
D. Module 3: Short-circuit Analysis Digital Signal Processing Pre-fault and post-fault phasors of voltages and currents
Module measured at substation
After determining the disturbance starting time, detection
Pre-fault and post-fault phasors of voltages and currents
and identification, the short circuit diagnosis is carried out measured along the feeder
along with the power quality analysis. This is possible due to Upload feeder database
the modular structure employed to develop the detection and
classification algorithms. The independency of the power Pre-fault Loading Load flow for pre-fault condition
Estimation
quality analysis and short circuit analysis modules is achieved (load flow)
behavioral indices of the system with four levels of detail and Fault current estimation Fault current calculation (Ifaultabc)
the fourth level of approximation, plus its complement. (short-circuit analysis)
𝑇
𝐼𝑥 = [𝑌1𝑑 𝑌 𝑇 𝑌1𝑑
1 1𝑑2
𝑇
3
𝑇
𝑌1𝑑 4
𝑇
𝑌1𝑎 4
𝑇
𝑌2𝑑 1
𝑇
𝑌2𝑑 2
𝑇
𝑌2𝑑 3
𝑇
𝑌2𝑑 4
Load flow for post-fault condition
𝑌2𝑎𝑇
𝑌 𝑇 𝑌3𝑑
4 3𝑑1
𝑇
2
𝑇
𝑌3𝑑 3
𝑇
𝑌3𝑑 4
𝑇
𝑌3𝑎 4
] (9) N
ΔISS converged ?
The second ANN is designed using an input vector ����
𝐼𝑥𝚤𝚤 also Y
with dimension 30, composed by the last three behavioral Fault section
Voltage deviation calculation d
identification
indices, obtained from oscillographic records of voltages
recorded at the Substation.
𝑇
𝐼𝑥 = [𝑌4𝑑 𝑌 𝑇 𝑌4𝑑
1 4𝑑2
𝑇
3
𝑇
𝑌4𝑑4
𝑇
𝑌4𝑎 4
𝑇
𝑌5𝑑 1
𝑇
𝑌5𝑑 2
𝑇
𝑌5𝑑 3
𝑇
𝑌5𝑑 4
Faulted area definition
(Indication of probable faulted nodes)
𝑇 𝑇 𝑇 𝑇 𝑇 𝑇
𝑌5𝑎4 𝑌6𝑑1 𝑌6𝑑2 𝑌6𝑑3 𝑌6𝑑 4
𝑌6𝑎 4
] (10)
End
The third ANN was designed and trained using an input Fig. 5. Flowchart of the fault location algorithm.
vector ����
𝐼𝑥𝚤𝚤 with dimension 60, composed by the six behavioral
1) Data Acquisition and Digital Signal Processing
indices of the system, referring to the current and voltage
oscillographic records recorded at the substation. The digital signal processing of the currents and voltages in
After the execution of the detection algorithm and the order to obtain the pre-fault and post-fault phasors of voltages
classification of disturbances, which provides an overview of and currents, necessary for performing the fault location
the operational status of the network, the process of algorithm is carried out through the Fast Fourier Transform
geographical fault location is started, through the fault location (FFT). The sampling of the current signals is also used to
algorithm. identify the fault type by means of a Fuzzy-ARTMAP neural
network that identifies the type and the phases involved in the
E. Module 4: Fault Location fault. Dedicated devices installed along the feeder should
The fault location procedure consists of automatically make remote measurements of pre-fault and post-fault
determining the geographical location of a fault by performing voltages available. Communication channels should also be
digital signal processing of the waveforms of voltages and available for the transmission of the measured data points to
currents. They are measured at the substation level and a the remote processing site of fault location algorithm.
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
5
57 122
computation of the voltage deviations (deltai) through the 1
51 50 49 61 62 63 76 77
88
89 90
119
indication of likely faulty nodes, using the distances (distij), 31
48 78
3 30 28 27 26 32 33 47 64 79 120 118
calculated for each line i of the system with j measuring 2
29
34
65
80
91
121
25 93 92 117
81
devices. 4
24
46
66
83
82
96
94
95 97 116
68 69
|𝑎𝑉𝑖𝑚𝑚𝑚𝑚 − 𝜃𝑖𝑚𝑚𝑚𝑚 + 𝑐|
6 7 8 9 10 22 35 36 37 39 70 84
5 99
115
𝑑𝑑𝑑𝑑𝑖𝑖 =
67 85 98
(13) 23 38
128
√𝑎2
86 113
+1
100
14 13 12 43 42 40 41 71 102 114
11 87 101 112
129
72 103
Where: distij is the distance between the measured and 15 44
73 105 106
107
coefficient. So, the voltage deviations are calculated using the ~ 20 ~ Distributed Generator ~ 134
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
6
A. Fault Detection Algorithm devices, the bigger the amount of ANN required. Each ANN
The disturbance detection is based on the set of rules was trained using a set of 11,704 simulations and then tested
presented in (1) and (2), which are composed by (a) macro with the simulations reported in the other measuring devices
analysis and (b) micro analysis. Through (a), the signal placed along the network. The vigilance parameters ρa and ρb
general behavior is examined, comparing the maximum adopted for the training of the 12 neural networks were 0.85
variation of the detail coefficients with a standard deviation and 0.95, respectively. The inter-ART parameter used was
percentage, which provides major immunity to noise presence 0.99 and the training rate for both networks, ARTa and ARTb,
in oscillographies. The threshold plays an important role for was 1.0.
the sensitivity to a fault because he performs a punctual
analysis in (b), analyzing the maximum value of the detail
coefficients (dijk), facing the highest detail coefficient with a
predetermined threshold value (γj). The thresholds are defined
by the protection engineer, based on the operating
characteristics of each system.
The evolution of a signal containing a short circuit caused
by a phase-to-ground fault of 20.0 Ω at node 43, starting at
0.08333 seconds is considered to illustrate the detection of a
disturbance and its current location time. Table I presents the
signal detection window and its respective statistical analysis
of variance and standard deviation.
Fig. 9. Waveform of a voltage sag in phase A of node 25.
TABLE I
Detection of a Current Disturbance in Phase A
Table II shows the operational efficiency of ANNs trained
Window max varij 3σ ij max dijk γj to classify PQ events occurring in phase A. The input patterns
. . . . .
. . . . . of these networks are composed by energy coefficients of the
15 227,5 326,7 314,0 281,2 first four levels of details and approximations of current and
16 174,5 296,8 297,8 281,6 voltage signals and, the output network presents the fault type
17 227,6 326,7 314,3 281,2 code and the phases involved in the problem.
18 823,7 571,3 544,6 281,6 TABLE II
Operational Efficiency of ANNs Trained for PQ Events in Phase A.
In this situation, the fault was detected after 0.08372 Operation
seconds related to the sample number 323 of the recorded bus # 26 75 87 115 Mean
signal. This result was found during the analysis of the 18th 26 100.0% 88.9% 88.7% 87.4% 91.2%
window sampled and decomposed by MRA, considering the 75 88.9% 100.0% 97.5% 93.2% 94.9%
Training
set of rules presented in (1) and (2). This module reached 87 88.2% 97.3% 100.0% 94.5% 95.0%
100% accuracy with a high-computing performance. 115 85.3% 91.9% 93.2% 100.0% 92.6%
Following the detection of the disturbance in a sampled Mean 90.6% 94.5% 94.8% 93.8% 93.4%
signal, the algorithm records 10 complete cycles of this signal, The events most frequently reported by ANNs after
being four pre-fault cycles and six post-fault cycles and, then, evaluation of all networks are the ones that will prevail and
starts the power quality analysis stage. these events will appear in the output report. The high
B. Power Quality Analysis performance of these ANNs in the classification process of
power quality disturbances can be clearly verified.
The PQ events were generated by summing the steady-state
sinusoidal waveform generated at a sampling rate of 7.68 kHz, C. Short-circuit Analysis
through the ATP/EMTP software [28], with a vector The training and the testing of the 3 ANNs responsible for
containing the respective disorder. The disturbance amplitude the short-circuit classification were carried out using the
is estimated using the peak values for voltage level of the first waveforms of voltages and currents simulated using the ATP
approach (cA1), before and during the disturbance. The software, applying fault situations in various parts of the
difference between such values indicates the variation in system, inserting these faults with various incidence angles
signal amplitude that characterizes the disturbance. and, different values of fault resistances. Table III shows the
Fig. 9 shows the first version of approximation (cA1a) of a variables used in the short-circuit simulations.
signal containing a voltage sag in phase A of node 25. In this TABLE III
example, the peak value before the disturbance has a value Variables Used in Short-Circuit Simulations.
equal to 10,379.07 and, the value during the disturbance Variable Variation Total
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
7
After training these three ANNs with the 17,556 fault computational time since the repairing time and dispatching of
simulations previously performed, in order to verify the maintenance crews take a time much higher. Fig. 10 shows the
efficiency of the classification process, other 11,704 confidence intervals of the probabilities of faults occurrence
simulations were performed using the variables shown in and the positions classified by the fault location algorithm, for
Table III, but now considering the angles of incidence of each fault type simulated. The presence of distributed
failure as being 30 and 60 degrees. Table IV shows the generation is considered in the calculation of power flow and
efficiency of the classification process encountered during the short circuit analysis and the fault resistance can be estimated
tests with the three ANNs. at the end of the process by the currents and voltages
TABLE IV encountered after their convergence, since the faults occurring
ANNs Efficiency for Short-Circuit Analysis.
in the system are modeled as current injections.
ANN Behavioral Indices Efficiency Graphic of Probability Intervals Graphic of Probability Intervals
95% RI for mean 95% RI for mean
𝑌1𝑗𝑇 , 𝑌2𝑗
𝑇 𝑇 100 95
1 and 𝑌3𝑗 97.3 %
90
𝑇 𝑇 𝑇
2 𝑌4𝑗 , 𝑌5𝑗 and 𝑌6𝑗 99.2 % 90
85
3 𝑌1𝑗𝑇 , 𝑌2𝑗
𝑇 𝑇 𝑇 𝑇
, 𝑌3𝑗 , 𝑌4𝑗 , 𝑌5𝑗
𝑇
and 𝑌6𝑗 99.8 %
Probability
Probability
80
80
75
60
the presence of ground or not was pretty much the same. A AB ABC ABCG ABG B
Phase
BC BCG C CA CAG A AB ABC ABCG ABG B
Phase
BC BCG C CA CAG
Position
Position
oscillographic records of the currents and voltages from the 8 7
2 4
along the feeder. A sequence of random variables indexed in A AB ABC ABCG ABG B
Phase
BC BCG C CA CAG A AB ABC ABCG ABG B
Phase
BC BCG C CA CAG
time with a well-defined correlation structure was considered (c) Position without DGs (d) Position with DGs
to simulate the operating conditions of a real system as a Fig. 10. Confidence intervals of the probabilities of faults occurrence and the
positions classified by the fault location algorithm.
stochastic process. Table V shows the random variables
considered in the simulations, with their respective variation For each bus feeder was applied a line-to-ground fault
ranges. occurring in phase A, considering a line-to-ground resistance
TABLE V of 10 ohms and medium loading modeled like constant
Stochastic Variables Used in Simulations. impedance. Fig. 11 shows the classification of buses for a fault
Std. occurring in bus 70. The fault location system had no problem
Variable Distribution Mean Variation
Deviation
Faulted Bus Uniform - - 2 to 134
classifying the fault because the voltage deviations showed
Fault Phase Uniform - - A, B ou C
quite different values, which shows the good accuracy of fault
5% three-phase location system.
Fault type Normal 0 1 10% line-to-line 100,00 100,00 100,00 100,00
100,0
Probability (%)
85% single-line-to-ground
Line-to-ground 75,0
Normal 25 4.854 10 – 40
resistance
50,0
Line-to-line
Normal 0.5 0.129 0.1 – 0.9
resistance 25,0
Loading Uniform - - high, medium, low 5,03 2,09 2,05 1,47 1,17 1,03
,0
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TPWRD.2015.2464098, IEEE Transactions on Power Delivery
8
distribution network, either operating normally or under [6] IEEE guide for identifying and improving voltage quality in power
systems, IEEE Standard 1250-2011, Mar. 2011.
contingencies. [7] IEEE recommended practice for monitoring electric power quality,
60
IEEE Standard 11590-2009, Jun. 2009.
50 48 [8] Procedures for Electric Power Distribution in the National Electric
System, Normative Resolution N. 395, Dec. 2009, [Online] Available:
Number of buses
40
http://www.aneel.gov.br/cedoc/ren2009395.pdf.
30
[9] I. D. Suarez, G. C. Caicedo, and H. R. Vargas-Torres, “Review of stable
and transient fault location techniques for distribution systems part II:
20
15
13
steady-state knowledge based and transient analysis techniques”, Int.
10
10
8 8 7
Conf. on Renewable Energies and Power Quality, pp. 1-6, 2010 .
4 4 3 2 2 2 2
4
[10] J. G. M. S. Decanini, M. S. Tonelli-Neto, and C. R. Minussi, “Robust
0 fault diagnosis in power distribution systems based on fuzzy artmap
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15+
neural network-aided evidence theory”, IET Generation, Transmission
Position Located
& Distribution, vol. 6, n. 11, pp. 1112-1120, Jul. 2012.
Fig. 12. Number of buses classified by position. [11] A. A. Girgis, C. M. Fallon, and D. L. Lubkeman, “A fault location
IV. CONCLUSIONS technique for rural distribution feeders,” IEEE Trans. Industry
Applications, vol. 29, n. 6, pp. 1170 – 1175, Nov. 1993.
A methodology for integrated, automated diagnosis and [12] R. Das, M. S. Sachdev, and T. S. Sidhu, “A fault locator for radial
fault location through the combined usage of modern subtransmission and distribution lines,” in IEEE Proc. Power
Engineering Society Summer Meeting, pp. 443 – 448, 2000.
techniques of network analysis, signal processing and [13] J. Zhu, D. L. Lubkeman, and A. A. Girgis, “Automated fault location
intelligent systems was presented in this paper. The and diagnosis on electric power distribution feeders,” IEEE Trans.
continuous (online) process of fault detection is accomplished Power Delivery, vol. 12, n. 2, pp. 801 – 809, Apr. 1997.
[14] S. J. Lee, M. S. Choi, S. H. Kang, B. G. Jin, D. S. Lee, B. S. Ahn, N. S.
through statistical analysis and direct current multilevel signal Yoon, H. Y. Kim, and S. B. Wee, “An intelligent and efficient fault
analysis in wavelet domain, recorded at substation. The location and diagnosis scheme for radial distribution systems,” IEEE
sampling rate adopted in this work was 7.68 kHz, representing Trans. Power Delivery, vol. 19, no. 2, pp. 524 – 532, Apr. 2004.
[15] E. C. Senger, G. Manassero, C. Goldemberg, and E. L. Pellini,
128 samples/cycle. This isn’t a very high frequency and can be “Automated fault location system for primary distribution networks,”
easily applied. The disturbance detection process is IEEE Trans. Power Delivery, vol. 20, n. 2, pp. 1332 – 1340, Apr. 2005.
accomplished using the fourth-order Daubechies (db4) [16] H. R. Diaz, and M. T. Lopez, “Fault location techniques for electrical
distribution networks: a literature survey”, in Proc. 2005 Int. Power and
wavelet filter in MRA. The behavioral indices of the current Energy Systems Conf., pp. 311-318.
and voltage signals are extracted employing the discrete [17] R. A. F. Pereira, M. Kezunovic, and J. R. S. Mantovani, “Improved fault
wavelet transform, multiresolution analysis and the concept of location on distribution feeders based on matching during-fault voltage
sags”. IEEE Trans. Power Delivery, v. 24, n. 2, p. 852-862, Apr. 2009.
signal energy. These proposed indexes correspond to the input [18] J. Mora-Florez, J. Melendez, and G. Carrillo-Caicedo, “Comparison of
vectors of a number of independent Fuzzy-ARTMAP neural impedance based fault location methods for power distribution systems”,
networks, which aim to estimate the fault type and the power Electric Power System Research, vol. 78, n. 4, pp. 657-666, Apr. 2008.
quality events present in the sampled signal. [19] I. D. Suarez, G. C. Caicedo, and H. R. Vargas-Torres, “Review of stable
and transient fault location techniques for distribution systems part I:
The fault location is performed after the classification steady-state locators based on circuit analysis”, in Proc. 2010 Int. Conf.
process, and it is applicable to three-phase radial or weakly on Renewable Energies and Power Quality, pp. 1-8.
meshed feeders with unbalanced loads, three phases, and [20] F. C. L. Trindade, W. Freitas, and J. C. M. Vieira, “Fault location in
distribution systems based on smart feeder meters”, IEEE Trans. Power
double-line or, single-line lateral extensions and, it works with Delivery, vol. 29, n. 1, p. 251 – 260, Feb 2014.
presence of distributed generation. [21] C. H. Kim, R., and Aggarwal, “Wavelet transforms in power systems -
The integrated system diagnosis and fault location is able to part I - general introduction to the wavelet transforms”, Power
Engineering Journal, vol. 14, n. 2, pp. 81-87, Apr. 2000.
assist the electric utility personnel in their daily job providing [22] C. S. Burrus, R. A. Gopinath, and H. Guo, Introduction to wavelets and
important information for the operation and maintenance, wavelet transforms: a primer, Upper Saddle River: Prentice Hall, 1998,
minimizing the repair time and, consequently, experiences p. 268.
[23] G. A. Carpenter, S. Grossberg, N. Markuzon, J. H. Reynolds, and D. B.
lesser financial losses due to ability for rapid fault diagnosis Rosen, “Fuzzy ARTMAP: a neural network architecture for incremental
and restoration to the normal system operation. supervised learning of analog multidimensional maps”, IEEE Trans.
The proposed methodology can be applied in online Neural Networks, vol. 3, n. 5, pp. 698–713, Sept. 1992.
operation to identify power quality events and still valid even [24] A. V. Oppenheim, R. W. Schafer, R. W., and J. R. Buck. Discrete-time
signal processing. Upper Saddle River: Prentice Hall, 1998, p. 870.
after a network reconfiguration since the Fuzzy-ARTMAP [25] C. S. Cheng, D. Shirmohammadi, “A three-phase power flow method for
ANNs is able to perform pattern recognition, independent of real-time distribution system analysis”, IEEE Trans. Power Systems,
network configuration. vol. 10, n. 2, pp. 671 – 679, May 1995.
[26] X. F. Zhang, F. Soudi, D. Shirmohammadi, and C. S. Cheng, “A
V. REFERENCES distribution short circuit analysis approach using hybrid compensation
[1] M. Kezunovic, “Smart fault location for smart grids”, IEEE Trans. method”, IEEE Trans. Power System, vol. 10, n. 4, pp. 2053-2059, Nov.
Smart Grids, vol. 2, n. 1, pp. 11-22, Mar. 2011. 1995.
[2] M. H. J. Bollen, Understanding power quality problems: voltage sags [27] Power System Test Case Archive, 2014. [Online]. Available:
and interruptions. New York: Willey-IEEE, 2000, p. 543. http://www.feis.unesp.br/#!/departamentos/engenharia-eletrica/pesquisas
[3] M. H. J. Bollen, and I. Y. H. Gu, Signal processing of power quality -e-projetos/lapsee/downloads/materiais-de-cursos1193//
disturbances. Massachusetts: Willey-IEEE, 2006, p. 861. [28] Comite Argentino de Usuarios del EMTP/ATP. (2002, Dec.). ATP
[4] M.H.J. Bollen, “Voltage sag: effects, mitigation and prediction”, Power rulebook. Revista Iberoamericana del ATP. [Online] 3(4), p. 700.
Engineering Journal, vol. 10, n. 3, pp. 129-135, 1996. Available: https://sites.google.com/site/atpcaue/reviberoamericana/ anio
[5] G. Olguin, F. Vuinovich and M.H.J. Bollen, “An optimal monitoring 2001/vol3_num4.
program for obtaining Voltage sag system indexes”, IEEE Trans. Power
System, vol. 21, n. 1, pp. 378-384, 2006.
0885-8977 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.