POWERAFRICA2012_adewole_ID25

Embed Size (px)

Citation preview

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    1/8

    IEEE PES PowerAfrica 2012 Conference and Exhibition

    Johannesburg, South Africa, 9-13 July 2012

    Fault Detection and Classification in a Distribution Network Integrated with

    Distributed Generators

    A.C. Adewole and R. TzonevaCentre for Substation Automation and Energy Management Systems

    Cape Peninsula University of Technology, Cape Town, South Africa

    Phone: +27 021-959-6459, Email: [email protected]

    Abstract This paper develops a methodology for application indistribution network fault detection and classification. The

    proposed methodology is based on wavelet energy spectrumentropy decomposition of disturbance waveforms to extractcharacteristic features by using level-4 db4 wavelet coefficients.Thus, few input features are required for the implementation.

    Different simulation scenarios encompassing various fault types atseveral locations with different load angles, fault resistances, faultinception angles, and load switching are applied to the IEEE 34

    Node Test Feeder. In particular, the effects of system changeswere investigated by integrating various Distributed Generators

    (DGs) into the distribution feeder. Extensive studies, verification,and analysis made from the application of this technique validatethe approach. Comparison with statistical methods based onstandard deviation and mean absolute deviation has shown thatthe method based on log energy entropy is very reliable, accurate,and robust.

    Index Terms Discrete wavelet transform, distribution network,fault detection and classification, wavelet energy spectrum.

    1.INTRODUCTION

    The recent restructuring in electric power utilities over the

    last decade has brought about the need for efficientgeneration and transfer (transmission and distribution) ofelectric power to load centers. The mode of power evacuation

    is usually via overhead lines. Overhead lines are subject to

    the forces of nature and other uncontrollable factors, thus

    liable to faults.

    An essential aspect of Abnormal Event Management

    (AEM) is fault detection and diagnosis. In the past, most

    research and development in power system faults detection

    and diagnosis focused on transmission systems, and it is notuntil recently with the introduction of stringent fault indices

    by regulatory bodies that research on power system faults hasbegun on the unique aspects of distribution networks. The

    application of algorithms designed for transmission networkswhen used for distribution lines are prone to errors because of

    the non-homogeneity, presence of laterals/tap-offs, radial

    operation, and load taps along distribution lines. Therefore,

    there is the need for contingency plans to troubleshoot faults

    and expedite service restoration in order to reduce downtime.

    Many diagnostic methods have been developed andproposed, but a perfect, dependable, and secure method is

    still the objective of continuous research. Methods based on

    Wavelet Transform (WT) for fault diagnosis were proposed

    by [1]-[6]. Reference [7] proposed a method for fault

    detection and classification in transmission systems using

    wavelet and fuzzy logic. Similarly, [5], [8], [10] suggested

    techniques using WT and Artificial Neural Network (ANN)

    for transmission line fault detection and classification.

    Another technique based on WT and Support Vector Machine(SVM) was proposed by [11] for power system disturbance

    classifier in transmission systems. Reference [12] presented a

    methodology for the classification of Power Quality (PQ)

    disturbances using Wavelet Packet Transform (WPT) andfuzzy k-nearest neighbor classifier. A method by [13] for PQ

    disturbances was based on Discrete Wavelet Transform

    (DWT) and wavelet network. Reference [14] also presented a

    WT and rule based method for power quality classification in

    a transmission network. A method based on Hubbard-

    Stratonovich (HS) transform and radial basis function neuralnetwork was suggested by [15]. Reference [16] described a

    method for fault detection and classification based on WT

    decomposition of the transformed current values. The method

    suggested the use of wavelet entropies for multi-agent fault

    diagnosis in distribution networks.

    Although, the method by [15] is fast because of the

    reduction in the computational requirements, the use of level-1 coefficients may fail to provide the appropriate transientcharacteristics that truly represent the fault type/phase(s)

    especially where there is mutual coupling between the

    phases. Also, the technique described by [14] did not cover

    the effect of noise disturbance on the model. The method

    proposed by [16] made use of Clarks Transform to convert

    the three phase current measurements to modal domain. The

    disadvantage of this is the added computation that would be

    required during implementation. In addition, the effects ofload angle, load switching, and capacitor switching were not

    considered in the various literature reviewed.In this paper, wavelet energy spectrum entropy based on

    log energy is employed to detect and classify faults in atypical distribution network. This is implemented by taking

    into account the distinct nature of distribution networks and

    network changes that are likely to occur. To validate the

    proposed approach, extensive simulation studies are carried

    out on the IEEE 34 Node Test Feeder Benchmark model at

    different fault locations, fault resistances, fault inceptionangles, load angle variations, load and capacitor switching,

    and network topology changes.

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    2/8

    The rest of this paper is organized as follows: Section II

    explains the principles of Wavelet Transform. Section III

    describes the Power System Model. The implementation of

    the fault detection and classification algorithm is outlined in

    section IV. Section V provides the results and discussion ofthis approach. Section VI summarizes the conclusion.

    II. WAVELET TRANSFORM ANALYSIS

    A. Wavelet Transform

    The classical Fourier Transform (FT) is a frequency

    domain method. That is, it transforms a signal from time-

    based to frequency-based one. Thus, time information is lost

    and it is impossible to tell when an event took place. Short

    Time (STFT) was introduced to correct the shortcoming of

    the FT. However, a fixed time window is used. Many signalsrequire a more flexible approach where the window size can

    be varied to determine the frequency or time more accurately.

    A method such as WT capable of multiple resolutions in time

    and frequency and with a flexible window size is thereby

    required. The windowing in WT automatically uses short

    time intervals for high frequency components and long time

    intervals for low frequency components by using scale and

    shift techniques.WT can be implemented using the Continuous Wavelet

    Transform (CWT). The CWT of a signal )(tx is the integral of

    the product between )(tx and the daughter-wavelets, which

    are the time translated and scale expanded/compressed

    versions of a function having finite energy, called mother-

    wavelet.

    The CWT of a signal )(tx is defined as [17], [18]:

    dta

    bttx

    abaC

    +

    = )(

    1),( (1)

    where )(t is the mother wavelet, a is the scale factor, b is

    the translation factor (position along the time axis), a2/1 is

    the normalization value of )(,t

    ba so that if )(t has a unit

    length, then its scaled version )(,t

    ba would also have a unit

    length.

    Another variant of WT is DWT. One area in which the

    DWT has been particularly successful is transient analysis in

    power systems [1], [2], [19]. This is because it acquires the

    transient features and accurately analyzes them in both thetime and frequency contexts at different frequency bands with

    different resolutions.

    The mathematical expression for DWT is given by [17-20]:

    =

    km

    m

    m

    knkfnmDWT

    2

    2)(

    2

    1),( (2)

    where )(kf is a discrete signal,)(n is the mother wavelet

    (window function), m and n are time scale parameters, k is

    the number of coefficients, 2

    m is the variable for scale, 2km

    is the variable for shift, and 21m is the energy normalization

    component to ensure the same scale as the mother wavelet.

    In implementing Multi-Resolution Analysis (MRA) for

    DWT, the scaling and wavelet functions are obtained from

    [21], [22].

    )22)( (

    )2/(

    ,ntt

    mm

    nm

    = (3)

    )22)( (

    )2/(

    ,ntt

    mm

    nm

    = (4)

    where )(, tnm is the scale function, and )(, tnm is the wavelet

    function.

    Wavelets are localized in both time (through translation)

    and frequency (through dilation). The first scale covers a

    broad frequency range at the high frequency end of the

    spectrum and the higher scales cover the lower end of the

    frequency spectrum. Signal decomposition starts by passing

    the signal through a set of filters. Approximations are the

    high-scale, low-frequency components of the signal producedby filtering with a low-pass filter with coefficient vector )(h

    .The details are the low-scale, high-frequency components of

    the signal produced by a high-pass filter with coefficient

    vector )(g .

    The filters are given by [22]:

    )2(2)()( ntngt

    n

    = (5)

    )2(2)()( ntnht

    n= (6)

    After each level of decomposition, the sampling frequency is

    reduced by half. Then, the lowpass filter output

    (approximation) is decomposed to produce the components ofthe next level. The original signal sequence )(kf can also be

    represented by the sum of all components i.e the sum of all

    the details and the approximation at the last level of

    decomposition. For example, for two levels of

    decomposition, the representation is:

    )()()()()()(22111kcAkcDkcDkcAkcDkf ++=+=

    )()()(1

    kcAkcDkf l

    l

    jj

    +==

    (7)

    where cDj

    is the detail at scale j and cAl

    is the approximation

    at scalej, and l= 2.

    B. Feature Extraction

    Fault signals are known to contain transients and

    harmonics. These high-frequency components carry essential

    information that could be used to identify fault orabnormalities in power system network. The energy of

    wavelet coefficient varies over different scales as per the

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    3/8

    energy distribution in the signal. Wavelet energy is the sum

    of the square of WT coefficients.

    The wavelet energy of a signal at scale jand instant k is

    given as [23]-[25]:

    )(

    2

    kDjEjk=

    (8)

    At scale j , the instants = 1, 2, 3, ..., N

    The log energy entropy of the signal at scale j is:

    =k

    jkEEj EW log(9)

    Standard Deviation is a statistical measure of distribution

    or spread in a data set and it is derived from the square root of

    the variance in a data set. Mean Absolute Deviation (MAD)is the mean of the absolute deviations of the data set from the

    mean of the data. It shows the statistical dispersion of a dataset.

    The standard deviation of the signal at scale j , instants k is:

    ( )= =N

    k

    jkj jDN 1

    221

    )(1

    1

    (10)

    Similarly, the Mean Absolute Deviation of a signal is givenas:

    =

    =N

    k

    jjkDN

    MAD1

    1 (11)

    whereDjk is the detail coefficient at scale j , instant k , j is

    the mean at scale j , and N is the number of instants.

    III. POWER SYSTEM MODEL

    A. Base Case IEEE 34 Node Test Benchmark Feeder

    The distribution network used is the IEEE 34 Node Test

    Feeder. It is a long feeder operated at 60Hz with unbalanced

    loading and nominal voltage of 24.9 kV. Fig. 1 shows the

    IEEE 34 Node Test Feeder.

    Fig. 1. IEEE 34 Node Test Feeder.

    Simulation of the power system was carried out using

    DIgSILENT PowerFactory and the steady state load flow

    results were validated with the results from IEEE 34 nodebenchmark system in [26]. The relative error of the node

    phase voltages is shown in Table 1.

    Table 1. Node Voltage Relative Error vs. [26]

    Dynamic electromagnetic transient simulation of differentfault types involving Single Phase-to-ground (1 Ph.-g), two

    Phase (2 Ph.), two phase-to-ground (2 Ph.-g), and three phase

    (3 Ph.) faults were performed. These simulations were carried

    out at different locations at an interval of 10% along the main

    feeder, at 95% of the main feeder, and on the laterals. Faultresistances )(Rf of 0, 2.5, 5, 10, 20, and 100, and

    fault inception angles (fa) of 0o, 30o, 45o, 60o, and 90o were

    used in the simulations. The fault inception angle fa is thephase angle of phase A voltage at the fault inception time.

    Simulations were done to discriminate between transients due

    to switching conditions from load and capacitor switching.Load angle variations of 0o, 60o, 90o were also carried out.

    The waveforms were generated with a sampling rate of 128

    samples per cycle.

    B. Modified IEEE 34 Node Test Benchmark Feeder

    Three cases which involved the integration of Distributed

    Generators (DGs) into the benchmark model were studied inthis paper. These include:

    DG1 case study: maximum load + 20% of DG

    installed at node 840

    DG2 case study: maximum load + 20% of DG

    installed at node 844

    DG3 case study: maximum load + 10% of DGs

    installed at node 840 and 844 respectively.

    Distributed generation refers to the electric power

    generation (usually between 5kW and 10MW) at theconsumption end of a distribution network. The generated

    power is integrated to the distribution network at the

    substation, feeder, or customer load levels [27]. DGs can be

    implemented with wind turbine, hydro, PV, fuel cells, etc.

    The integration of DGs into a distribution network often

    causes protection coordination issues [28].

    Case studies involving the integration of DGs into the IEEE

    34 node test feeder were carried out to investigate the abilityof WT based log energy entropy to correctly detect and

    classify faults even after DGs were integrated.

    Relative Error

    (%)

    Phase A Phase B PHASE C

    Minimum -5.2032 -1.5586 -0.2325

    Maximum

    Average

    -0.0307

    3.1509

    1.2722

    0.7719

    2.3210

    0.8746

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    4/8

    The studies carried out in this paper did not assign any

    specific energy source to the DG. Also, the parameters of the

    synchronous generators were based on previous work carried

    out by [28]. The connection of the generator to the grid was

    via a 500kVA step-up transformer. The transformerimpedances were also set equal to the transformer at node

    832 (XFM-1).However, the transformer winding was changed to delta-

    star type based on the recommendations by [27] on optimal

    transformer winding types for DGs. The placement and sizing

    of the DGs were based on [28]-[31]. Thus, node 840 (along

    the main line) and node 844 (one of the laterals) were used

    with a 20% penetration level one at a time. Furthermore,another test case was simulated with smaller DGs co-located

    in the network at nodes 840 and 844 respectively. A plot of

    the voltage profile is given in Fig. 2. Similarly, a plot of the

    short circuit currents at various nodes are shown in Fig. 3.

    Nodes 836-1 refers to lateral 836-862, while Node 836-2

    refers to lateral 836-840. The voltage profile plot shows the

    impact of the integration of DG into the feeder. Also, there

    was an increase in the short circuit current at various nodes inthe feeder.

    Fig. 3. Voltage Profile of the various Case Studies

    IV. ALGORITHM FOR FAULT DETECTION AND

    CLASSIFICATION

    A. Feature Extraction

    Various simulations were carried out in DIgSILENT

    PowerFactory. The waveform plots of the three phase andzero sequence currents were exported to MATLAB as ASCII

    files. These files are decomposed into coefficients using db4

    level-6.

    Daubechies 4 (db4) is one of the most used wavelet in

    power system disturbance analysis and it was chosen for this

    research because of its orthogonality, compact support in the

    time domain, and for its good performance in power system

    studies as reported by [11], [12], [32], [33].

    The lowpass filter )(g and highpass filter )(h of the db4

    have four coefficients. These coefficients are:4830.0,8365.0,2241.0,1294.0

    4321==== gggg

    1294.0,2241.0,8365.0,4830.04321

    ==== hhhh

    The particular level of decomposition to use is based on the

    wavelet spectra. The log energy entropy, standard deviation,and mean absolute deviation at levels-1 to -6 were computed

    using (9) - (11). Level-4 was chosen as the level of interest

    for both fault detection and classification because the best

    results for log energy entropy, standard deviation, and mean

    absolute deviation were obtained at that level. Level-4

    corresponds to the frequency range of 240Hz to 480Hz.

    Fig. 4. Short Circuit Current for the various Case Studies

    B. Design of Rule Based Detector and Classifier

    The proposed algorithm in this paper is implemented with

    software subroutines written in MATLAB. The fault

    detection module is activated first and on detection of a faultcondition, the fault type and faulted phase(s) module is

    triggered to perform the classification tasks. Each fault has its

    characteristic feature or signature by which its faulted

    phase(s) can be identified.

    The fault detection module compares the computed level-4

    entropy values with a predetermined threshold )(d for each

    of the phases. The predetermined threshold )(d is carefully

    800 810 820 830 840 850 860 870 880 8900.95

    0.96

    0.97

    0.98

    0.99

    1

    1.01Voltage Profile

    Node

    P

    h.

    Vab(p

    .u)

    DG Case 1

    DG Case 2

    DG Case 3

    Base Case

    1 2 3 4 5 6 7 8 9 100

    100

    200

    300

    400

    500

    600

    700

    800

    900

    1000

    N-800

    N-8

    08

    N-8

    16

    N-8

    24

    N-8

    54

    N-832

    N-8

    58

    N-8

    34

    N-8

    36-1

    N-8

    36-2

    N = Node

    Short Circuit Current for Case Studies

    Nodes

    MaxShortCircuitCurrent/(A)

    Base Case

    DG Case 1

    DG Case 2

    DG Case 3

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    5/8

    chosen to ensure that the algorithm would be able to

    accurately discriminate between faults and normal switching

    events. In this particular case, da = 100, db = 100, and

    dc= 100. Fault is detected when any of the computed

    wavelet entropy values )(),(),( cWEEbWEEaWEE for the

    three phases is greater than dp . ),,( CBAp . whereda ,

    db , anddcare the entropy value thresholds for phase A, B,

    and C respectively.

    When fault is detected, the fault classification module is

    triggered for fault type classification and faulted phase(s)

    identification. The patterns observed through exhaustive

    simulations were used to draw up the rules for the algorithm.

    Fig. 5. is a flow chart for the implementation of this

    algorithm.

    The criteria for fault classification are:

    R1: if )(aWEE > )(ca , & )(bWEE < )(cb & )(cWEE < )(cc

    A-g FaultR2: if )(aWEE < )(ca , & )(bWEE > )(cb & )(cWEE < )(cc

    B-g Fault

    R3: if )(aWEE < )(ca , & )(bWEE < )(cb & )(cWEE > )(cc

    C-g Fault

    R4: if )(aWEE > )(ca , & )(bWEE > )(cb & )(cWEE < )(cc

    AB Fault

    R5: if )(aWEE < )(ca , & )(bWEE > )(cb & )(cWEE > )(cc

    BC Fault

    R6: if )(aWEE > )(ca , & )(bWEE < )(cb & )(cWEE > )(cc

    CA Fault

    R7: if )(aWEE > )(ca , & ( )(bWEE ) > )(cb & )(cWEE < )(cc

    AB-g FaultR8: if )(aWEE < )(ca , & ( )(bWEE ) > )(cb & )(cWEE > )(cc

    BC-g Fault

    R9: if )(aWEE > )(ca , & )(bWEE < )(cb & )(cWEE > )(cc

    CA-g Fault

    R10: if )(aWEE > )(ca , & )(bWEE > )(cb & )(cWEE > )(cc

    3Ph. Fault.

    where )(),(),( cWEEbWEEaWEE are the computed level-4

    log energy entropy values for phases A, B, and C. )(ca ,

    )(cb , and)(cc are the fault classification thresholds for

    phases A, B, C respectively and was set to 100.

    2Ph. and 2Ph.-g faults were classified using the values of0IWEE . Line-to-ground faults exhibited higher zero sequence

    entropy ( 0IWEE ), thus, this formed the basis for 2Ph. and

    2Ph.-g classification.

    Therefore, faults with values of 0IWEE > -250 will be

    classified as 2 Ph.-g faults.

    V. RESULTS AND DISCUSSIONS

    A. Results

    The proposed method was tested using several fault cases

    comprising of various fault types, fault conditions, and

    system parameters. In particular, the line segments at thebeginning and at the extreme end of the feeder were studied.

    Initialization

    Select the nextevent

    3 Ph. Fault

    Select level -4 detail coefficients

    DWT level -6 decompositionusing db4 mother wavelet

    PrintResult

    No

    Yes

    Select 3Ph. & zerosequence currents

    waveforms

    dpWEEp >

    WEE IWEE p 0, CBAp ,,Compute log energy entropy

    Fault?

    Single PhaseFault?

    2 Ph. Phase Fault ?

    2 Ph.-g Phase

    Fault?

    A-g, B-g, C-g

    AB, BC, CA

    AB-g, BC-g, CA-g

    Yes

    Fig. 5. Flowchart of the Proposed Algorithm

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    6/8

    Fig. 6. Distribution Plot of 2Ph. Fault for DG1 Case Study

    The effects of the following were considered: Faultresistance, fault distance, fault inception angle, and the

    integration of DGs. Fig. 6. is a visualization of the

    distribution or spread of 2 Ph. faults for 10% to 95% of the

    main feeder, laterals 820-822, and 846-848 respectively for

    DG1 case study using log energy entropy. This shows that the

    fault types are quite distinguishable from one another. From

    the results obtained through several simulation cases, the

    faulted phase was seen to have the highest log energy

    entropy. The text in bold signify the faulted phase(s).

    Statistical methods have been reported to show good

    performance in power system analysis [34-39]. The proposed

    method based on wavelet log energy entropy is comparedwith that based on features from Standard Deviation and

    Mean Absolute Deviation (MAD) of the WT decomposition.Tables 2-4 show some of the results obtained for log energy

    entropy, standard deviation and mean absolute deviation

    respectively.

    Table 2. DG1 Case Study at 10% of the Main Feeder (Rf= 0 , f = 0o)

    Method

    No Fault

    (0o) Load

    Angle

    No Fault

    (60o) Load

    Angle

    No Fault

    (90o) Load

    Angle

    Load

    Switching

    Capacitor

    844

    Switching

    1 Ph.

    A-g

    2Ph.

    A-B

    2Ph.

    A-B-g

    3Ph.

    )(aWEE 62.16 26.74 64.64 59.2 77.73 219.22 188.20 213.95 217.055

    )(bWEE 9.76 53.38 12.03 8.42 32.71 36.65 126.39 159.20 165.31

    )(cWEE

    )0(IWEE

    )(a)(b)(c)0(I

    )(aMAD

    )(bMAD

    )(cMAD

    )0(IMAD

    47.94

    -451.03

    3.68

    3.00

    2.75

    0.20

    2.12

    1.82

    1.80

    0.12

    10.59

    -414.94

    3.42

    2.87

    3.12

    0.23

    2.08

    1.87

    1.77

    0.14

    42.45

    -414.94

    3.16

    3.11

    3.03

    1.83

    1.94

    1.90

    1.83

    0.24

    45.45

    -474.87

    3.67

    2.99

    2.75

    0.20

    2.12

    1.82

    2.57

    0.13

    38.67

    -440.38

    4.34

    3.45

    2.81

    0.21

    2.53

    2.11

    1.88

    0.13

    83.38

    -136.23

    25.86

    6.54

    5.87

    6.83

    13.55

    3.52

    3.17

    3.62

    38.16

    -506.73

    33.23

    32.69

    3.27

    0.13

    13.30

    12.95

    2.25

    0.091

    73.36

    -132.63

    27.21

    51.15

    11.57

    20.69

    13.54

    19.51

    5.44

    8.91

    229.449

    -820.41

    26.33

    34.50

    42.89

    17.84

    13.48

    14.29

    18.33

    4.28

    Table 3. B-C Fault at Line 846-848 (Rf= 2.5 , f=30o)

    Case

    Study)(aWEE )(bWEE )(cWEE )0(IWEE )(a )(b )(c )0(I )(aMAD )(bMAD )(cMAD )0(IMAD

    Base

    Case

    19.94 154.22 170.03 -279.02 1.79 6.11 6.56 0.45 1.54 3.95 4.10 0.25

    DG1 46.34 138.03 146.82 -282.16 3.52 4.85 5.16 0.13 2.17 4.88 5.48 1.93

    DG2

    DG3

    61.89

    20.01

    137.41

    165.35

    122.88

    127.53

    -255.06

    -282.67

    3.52

    3.84

    4.63

    4.66

    4.72

    4.56

    0.11

    0.09

    2.51

    2.73

    3.105

    3.09

    2.99

    2.87

    0.07

    0.06

    Table 4. C-A-G Fault at Line 820-822 (Rf= 5 , f = 60o)

    Case

    Study )(aWEE )(bWEE )(cWEE )0(IWEE )(a )(b )(c )0(I )(aMAD )(bMAD )(cMAD )0(IMAD

    Base

    Case

    188.90 21.41 194.09 -69.81 13.04 3.27 13.74 4.85 6.61 2.14 6.64 2.01

    DG1 185.5 30.55 180.87 -82.18 9.76 4.11 8.86 4.24 5.46 3.00 5.04 1.78

    DG2

    DG3

    136.9

    144.4

    22.69

    12.99

    159.07

    154.47

    -142.2

    -165.48

    9.56

    8.91

    4.01

    2.73

    8.93

    9.35

    3.92

    0.18

    5.22

    4.63

    2.39

    1.88

    4.92

    5.04

    1.40

    0.09

    B. Discussion

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    7/8

    The algorithm was able to differentiate between fault events

    and no fault conditions like load switching, capacitor

    switching, and steady-state system operation when log energy

    entropy, standard deviation, and mean absolute deviation

    were used as inputs. Tables 2-4 show the results obtained atfault locations close to the upstream substation, a lateral, and

    at a location 189,205 ft. away from the upstream substationrespectively. Table 2 present some of the values obtained for

    fault detection at 10% of the main feeder (Line 806-808).

    Tables 3 and 4 show the results for faults at various fault

    inception angles and fault resistances. For all the methods

    presented, the faulted phase is associated with values many

    times greater than the healthy/unfaulted phase(s). For faultdetection and classification using log energy entropy, the

    various fault types were quite distinguishable for all the case

    studies. Furthermore, the thresholds used for the fault

    detection and classification for the base case performed well

    even for the DG cases without the need to review these

    thresholds. Simulation plots and entropy results showed the

    existence of mutual coupling in the phases especially for

    faults in close proximity to the DG location. However, the

    algorithm was able to accurately distinguish between the

    healthy phase(s) and the faulted phase(s). Although, different

    entropy values were obtained for different combination of

    fault resistances and fault inception angles for the same

    location, that did not affect the detection and classificationperformance since the results obtained per fault types were

    apparently above the pre-defined thresholds.The values obtained for and MAD after the integration

    of DGs were not completely useful for fault classification

    because there were similarities between the values obtained

    for faulted phase(s) in one location and the values for healthy

    phase(s) at another location. Table 5 illustrates some of the

    errors obtained when using and MAD. Entry 1 of Table 5was misclassified as no fault. Entry 2 was also misclassified

    as A-B fault. Entries 3 and 4 were wrongly denoted as no

    fault respectively. Also, the values of and MAD showed a

    corresponding decrease in value for faults with high

    resistances and for faults located far away from the

    substation. This implies that and MAD are influenced by

    fault resistance and fault location.

    Table 5. Features Using Standard Deviation and Mean Absolute Deviation

    CaseStudy

    Location Fault

    Type

    Fault Parameters )(a )(b )(c )0(I )(aMAD )(bMAD )(cMAD )0(IMAD

    DG1 Line 834-842 A-g Rf= 0, f= 0o 5.12 3.85 3.55 1.99 3.43 2.57 2.30 1.16

    DG1 Line 834-842 B-g Rf= 0, f= 0o 9.25 10.25 4.05 4.39 5.31 4.53 2.30 1.97

    Base Case Line 828-830 A-g Rf= 100, f= 0o 5.43 2.98 2.88 2.08 3.46 1.93 2.00 0.86

    Base Case Line 834-842 A-g Rf= 100, f= 0o 5.35 3.39 3.35 2.67 3.34 2.17 2.21 1.02

    VI. CONCLUSION

    This paper proposes an accurate approach for fault

    detection and classification of fault types and faulted phase(s)

    in distribution networks. Various scenarios were simulatedusing DIgSILENT PowerFactory. DWT was implemented in

    MATLAB to decompose the three phase and zero sequence

    current waveforms using db4 level-4 detail coefficients. A

    rule based method was used afterwards for fault detection and

    classification tasks respectively. It was observed that the

    proposed method based on wavelet log energy entropy

    accurately detects and classify the fault type. Comparisons

    with statistical feature extraction methods based on

    computation of standard deviation and mean absolutedeviation of the WT decomposition show that the method

    based on log energy entropy is very reliable, accurate, robust,

    and is independent of system conditions/changes. That is, it

    provided accurate results irrespective of load angle variation,

    load and capacitor switching, and network topology changes.

    It is also immune to varying fault location, fault resistance,

    and fault inception angles. It has been shown to be suitable

    for conventional distribution network as well as modified

    network with DGs. As a result of its simplicity, accuracy andspeed of operation, it can be used to aid existing protection

    equipment in fault diagnosis.

    ACKNOWLEDGEMENT

    This research work is funded by the South African

    National Research Foundation (NRF) UID62364 Substation

    Automation and Energy Management Systems.The authors

    are grateful for the financial support.

    REFERENCES

    1 Z.G. Bo, A.T. Johns, and R.K. Aggarwal, R.K., A Novel Fault Locator basedon the Detection of Fault Generated High Frequency Transients.Developments in Power System Protection, IEE Conference Publication, No.

    434. pp. 197-200. 25-27th Mar. 1997.

    2 Z.Q. Bo, G. Weller, and M.A. Redfern, Accurate fault location technique fordistribution system using fault-generated high-frequency transient voltagesignals.IEE Proceedings of Generation, Transmission, and Distribution, Vol.

    146, No. 1, pp. 73-79, Jan. 1999.

    3 A.M. Gaouda, M.M.A. Salaina M.K. Sultan, and A.Y. Cliildiaoi, PowerQuality Detection and Classification using Wavelet-Multiresolution Signal

    Decomposition. IEEE Transactions on Power Delivery, Vol. 14, No. 4, pp.1469-1476, Oct.1999.

  • 7/31/2019 POWERAFRICA2012_adewole_ID25

    8/8

    4 A. Abur, and F.H. Magnago, Use of time delays between modalcomponents in wavelet based fault location Electrical Power and

    Energy Systems,No. 22, pp. 397403, 2000.

    5 K. Gayathri, N. Kumarappan, and C. Devi, An Apt method for FaultIdentification and Classification on EHV Lines using Discrete Wavelet

    Transform. Power Engineering Conference IPEC, Singapore, pp. 217-

    222, 3-6 Dec., 2007.

    6 O.A.S. Youssef, Combined Fuzzy-Logic Wavelet-Based FaultClassification Technique for Power System Relaying.IEEE Transactions

    on Power Delivery, Vol. 19, No. 2, pp. 582-589. 2004.

    7 R.N. Mahanty, and P.B. Dutta Gupta, A fuzzy logic based faultclassification approach using current samples only. Electric Power

    Systems Research, No. 77, pp. 501507, 2007.

    8 K.M. Silva, B. A. Souza, and N.S.D. Brito, Fault Detection andClassification in Transmission Lines based on Wavelet Transform andANN.IEEE Transactions on Power Delivery, Vol. 21, No. 4, pp. 2058-

    2063, 2006.

    9 V.S. Kale, S.R. Bhide, P.P. Bedekar and G.V.K. Mohan, Detection andClassification of Faults on Parallel Transmission Lines using WaveletTransform and Neural Network. World Academy of Science, Engineering

    and Technology, No. 46, pp. 927-931, 2008.

    10 M. Geethanjali, M. and K. Sathiya Priya, Detection and Classification inTransmission Lines. International Conference on Control, Automation,

    Communication and Energy Conservation, pp. 1-7, 4-6 June, 2009.11 S. Ekici, S. Yildirim, M. Poyraz, Energy and entropy-based feature

    extraction for locating fault on transmission lines by using neural network

    and wavelet packet decomposition, Expert Systems with Applications,No.34, pp. 29372944, 2008.

    12 B.K. Panigrahi and V.R. Pandi, Optimal feature selection forclassification of power quality disturbances using wavelet packet-based

    fuzzy k-nearest neighbor algorithm IET Generation, Transmission,Distribution, Vol. 3, Iss. 3, pp. 296306, 2009.

    13 M.A.S. Masoum, S. Jamali, N. Ghaffarzadeh, Detection andclassification of power quality disturbances using discrete wavelet

    transform and wavelet networks. IET Science, Measurement,

    Technolology. Vol. 4, Iss. 4, pp. 193205, 2010.

    14 H.K. Chuah, P. Nallagownden, and K.S. Rama Rao, Power QualityProblem Classification Based on Wavelet Transform and a Rule-Based

    method. IEEE International Conference on Power and Energy (PECon2010), Kuala Lumpur, Malaysia, pp. 1-6, Nov 29 - Dec 1, 2010.

    15 S.R. Samantaray, P.K. Dash, G. Panda, Fault classification and locationusing HS Transform and radial basis function neural network, Electric

    Power Systems Research, No. 76, pp. 897905, 2006.

    16 El-Zonkoly A.M. Fault Diagnosis in Distribution Networks withDistributed Generation, Smart Grid and Renewable Energy, pp. 1-11,

    February 2011.

    17 M. Vetterli and C. Herley. Wavelets and Filter Banks: Theory andDesign.IEEE Transactions on Signal Processing. Vol. 40, No. 9, pp.

    2207-2232, Sep. 1992.

    18 A. Borghetti, S. Corsi, C.A. Nucci, M. Paolone, L. Peretto, and R.Tinarelli, On the use of continuous-wavelet transform for fault location

    in distribution power systems. Electrical Power and Energy Systems,

    No. 28, pp. 608617, 2006.

    19 A.S. Yilmaz, A. Subasi, M. Bayrak, V. M. Karsli, E. Ercelebi,

    Application of lifting based wavelet transforms to characterize powerquality events. Energy Conversion and Management, No. 48, pp. 112

    123, 2007.

    20 S. Ekici, S. Yildirim,M. Poyraz, A transmission line fault locator basedon Elman recurrent networks. Applied Soft Computing, pp.341347,

    2009.

    21 S. Burrus, R.A. Gopinath, and H. Guo, Introduction to Wavelet andWavelet Transform: A Primer.Prentice-Hall, 1998.

    22 J.C. Goswami, and A.K. Chan, Fundamental of Wavelets: Theory,Algorithms and Application.John Wiley & Sons, 1999.

    23 H. Zhengyou, F. Ling, L. Sheng, and B. Zhiqian, Fault Detection andClassification in EHV Transmission Line Based on Wavelet Singular

    Entropy. IEEE Transactions on Power Delivery, Vol. 25, No. 4, pp.2156-2163, 2010.

    24 H. Zhengyou, H.E. Shibin, G. Xiaoqin, C., Jun, Z., Zhiqian, and Q.Qingquan, Study of a new method for power system transients

    classification based on wavelet entropy and neural network. Electrical

    Power and Energy Systems, Vol. 33, pp. 402410, 2011.

    25 A.M. El-Zonkoly, Fault diagnosis in distribution networks withdistributed generation. Electric Power Systems Research Vol. 81, pp.

    14821490, 2011.

    26 IEEE Distribution System Analysis Subcommittee. Radial Test Feeders[Online]. Available:

    http://www.ewh.ieee.org/soc/pes/dsacom/testfeeders.html .

    27 P. Barker, R.W. de Mello, Determining the impact of distributedgeneration on power systems: part 1 Radial power systems. In Proc.

    IEEE Power Eng. Soc. Summer Meeting, Vol. 1, pp. 16451658, 2000.

    28 J.A. Silva, H.B. Funmilayo, and K.L. Butler-Purry, Impact of DistributedGeneration on the IEEE Node Radial Test Feeder with OvercurrentProtection.Proceedings of 39thNorth American Power Symposium. pp.

    49-57, 2007.

    29 R.C. Dugan and W.H. Kersting, Induction machine test case for the 34-bus test feeder-description. IEEE Power Engineering Society General

    Meeting, pp. 1-4, 2006.30 S. Santoso, and Z. Zhou, Induction machine test case for the 34-bus test

    feeder: a wind turbine time domain model. IEEE Power Engineering

    Society General Meeting, pp. 1-4, 2006.

    31 N. Samaan, T. McDermott, B. Zavadil, and J. Li, Induction machine testcase for the 34-bus test feeder-steady state and dynamic solutions.IEEE

    Power Engineering Society General Meeting, pp. 1-5, 2006.

    32 N. Perera, and A.D. Rajapakse, Power system transient classification forprotection relaying. 13th International Conference on Harmonics andQuality of Power, ICHQP, pp. 1 6, 2008.

    33 F. B. Costa, B. A. Souza, and N. S. D. Brito, Real-time detection of fault-induced transients in transmission lines,IET Electronics Letters, pp. 753-

    755, May 2010.

    34 P.K. Dash, M.V. Chilukuri, B.K. Panigrahi, Power Quality Analysis and

    Classification using a Generalized Phase Corrected Wavelet Transform.Power Electronnics, Machines and Drives, Conf Pub No. 487, pp.610-

    615, 16-18 April, 2002.

    35 P. K. Dash, B. K. Panigrahi, and G. Panda, Power Quality AnalysisUsing STransform.IEEE Transactions on Power Delivery, vol. 18, No.

    2pp. 406-411, April 2003.

    36 W. Kanitpanyacharoean, S. Premrudeepreechacharn. Power QualityProblem Classification using Wavelet Transformation and Artificial

    neural Network IEEE PES Power Systems Conference and Exposition,

    vol. 3, pp. 1496-1501. 2004.

    37 G.G. Pozzebon, G.G. Pena, A.F.Q Goncalves, R.Q. Machado, NeuralClassification of Power Quality Disturbances: An Application of the

    Wavelet Transform and Principal Component Analysis 9 th IEEE/IAS

    International Conference on Industry Applications (INDUSCON), pp. 1-6.

    2010.

    38 C. Venkatesh, D.V.S.S Siva Sarma, M. Sydulu, Classification of voltagesag, swell and harmonics using S-transform based modular neuralnetwork 14thInternational Conference on Harmonics and Quality of

    Power (ICHQP),pp. 1-7, 2010.

    39 J. Christy X. Jeno Vedamani S. Karthikeyan , Wavelet Based Detectionof Power Quality Disturbance - A Case Study. Proceedings of 2011

    International Conference on Signal Processing, Communication,

    Computing and Networking Technologies (ICSCCN 2011) pp. 157-162,2011.

    http://www.ewh.ieee.org/soc/pes/dsacom/testfeeders.htmlhttp://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=Authors:.QT.Perera,%20N..QT.&newsearch=partialPrefhttp://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=Authors:.QT.Rajapakse,%20A.D..QT.&newsearch=partialPrefhttp://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=4662486http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=4662486http://www.ewh.ieee.org/soc/pes/dsacom/testfeeders.htmlhttp://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=Authors:.QT.Perera,%20N..QT.&newsearch=partialPrefhttp://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=Authors:.QT.Rajapakse,%20A.D..QT.&newsearch=partialPrefhttp://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=4662486