IET Signal Processing

Research Article

Detection and classification of islanding and ISSN 1751-9675

Received on 28th June 2016
Revised 15th May 2017
power quality disturbances in microgrid using Accepted on 15th August 2017
E-First on 28th September 2017
hybrid signal processing and data mining doi: 10.1049/iet-spr.2016.0352
www.ietdl.org

techniques
Tatiana Chakravorti1, Rajesh Kumar Patnaik2, Pradipta Kishore Dash3
1Electronics and Communication Engineering, SOA University, Bhubaneswar, Odisha 751030, India
2Electrical and Electronics Engineering, SOA University, Bhubaneswar, Odisha 769030, India
3Multidiciplinary Research Cell, SOA University, Bhubaneswar, Odisha 751024, India

E-mail: pkdash.india@gmail.com

Abstract: This study presents multi-scale morphological gradient filter (MSMGF) and short-time modified Hilbert transform
(STMHT) techniques, respectively, to detect and classify multiclass power system disturbances in a distributed generation (DG)-
based microgrid environment. The non-stationary power signal samples measured near the target DG's are processed through
the proposed MSMGF and STMHT techniques, respectively, and some computations over them generates the target parameter
sets. Depending on the complexity of the overlapping in the target attribute values for different disturbance patterns, fuzzy
judgment tree structure is incorporated for multiclass event classification, which proves to be robust for most of the classes. In
this regard, an extensive simulation on the proposed microgrid models, subjected to a number of multiclass disturbances has
been performed in MATLAB/Simulink environment. The faster execution, lower computational burden, superior efficiency as well
as better accuracy in multiclass power system disturbance classification by the proposed judgment tree-based MSMGF and
STMHT techniques, respectively, as compared to some of the conventional techniques, is significantly illustrated in the
performance evaluation section. Further, as illustrated in this section, the real-time capability of the proposed techniques has
been verified in the hardware environment, where the results shown are satisfactory.

1 Introduction the SPT's, like WPT, HHT, WT and so on, become unreliable [1].
Adding to this fact, spectral leakage problem is the major drawback
In spite of rigorous expansion of transmission and distribution of associated with WT or WPT, which further degrades their
power supply, it is still inadequate to meet the scarcity of the power performance. In addition, some of the techniques such as HT or
demand. Consequently, there has been an elevation in the HHT [16] though illustrate a better performance but are
penetration of renewable energy-based distributed generations in computationally overburdened.
the existing power networks, most applicable in microgrid In this context, some of the SPT's, such as morphological filters
environment. Microgrid technologies [1, 2] are maturing rapidly (MFs) [17] and short-time HTs (STHT) [18] have been identified
day by day, as the renewable percentage in the generation mix as to be fast and excellent candidates for the processing of the non-
rises. Thus by generating clean power and minimising power stationary power signals, which are often contaminated with power
congestion on the grid during peak periods, micro-grids can help to quality problems. In contrast to some of the conventional SPT's,
reduce the requirement for additional utility generation, the involvement of a few number of additions and subtractions,
transmission and distribution. In addition to that, the cost of have made the formation of the multi-scale morphological gradient
generation of such power is affordable. As far as the protection of (MSMG) as well as STHT easier and subsequently reducing
microgrid is concerned, processing of the non-stationary power complexity [19]. Both the algorithms involve less computational
signals is considered to be very much significant [3–5]. Adding to burden in comparison with some of the integral transform methods
this fact, the gradual increase in the penetration levels of the ever like WT and so on.
fluctuating distributed generations (DGs), has consequently led to Therefore, in the context to the above discussions, and as a new
some of the issues, such as power quality and voltage instability contribution to the existing literature, two of the advanced and
and so on. Consequently, it demands for the development of some computationally efficient SPT's, i.e. MSMG filter (MSMGF) and
efficient signal processing techniques (SPTs) [6] for rapid and short-time modified HT (STMHT), are proposed in this paper.
accurate detection as well as classification of the disturbances to Conceptually, the proposed STMHT incorporates HT to little
ensure reliable operation of the DGs [7, 8]. Some of the widely windowed fragments of the signal while the Gaussian window
used SPTs include Fourier transform (FT), short-time FT, Hilbert moves along the span of the signal. In STMHT, each window
transform (HT), Hilbert–Huang transform (HHT), parallel diverges from the preceding by one sample point, where the middle
computing, wavelet transform (WT), wavelet packet transform value of the HT is being calculated in each window. The middle
(WPT), Gabor transform, S-transform [1, 2, 9–13] and so on, for value of the entire individual window is processed to achieve the
handling islanding and power quality problems in microgrid STMHT plot. The accuracy of the proposed method significantly
environments. In [14] the foremost inadequacies of the above SPTs depends on the selection of an accurate window. The voltage and
are discussed. FT is an analysing tool for extracting the frequency current signal samples at the target DG terminals of a standard IEC
contents of the signal. However, due to its constant bandwidth, it is microgrid model [20] in MATLAB/Simulink are collected for a
ineffective in capturing short-term transients like impulses and number of operating conditions, which are processed through the
oscillatory transient in power networks. In this context, some proposed MSMGF and STMHT techniques, respectively, in order
improved techniques are proposed for detection and classification to detect the occurrence of the events. Some simple computations
of islanding and power quality [15] problems in microgrids. over the outputs of the proposed MSMGF and STMHT techniques,
Whenever the signals are buried under noise or harmonics, some of respectively, generate the target parameter set for the microgrid

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 82

© The Institution of Engineering and Technology 2017
subjected to a large number of multiclass power system Y Nδ(i) = (α ⊕ N β)(i) = α ⊕ β ⊕ β ⊕ ⋯ ⊕ β(i) (6)
disturbances like islanding, switching faults and power quality
events such as voltage sag, voltage swell and capacitor bank Y Nε(i) = (α ΘN β)(i) = α Θβ Θβ Θ⋯Θβ(i) (7)
switching, harmonic load switching, balanced and unbalanced load
switching, flicker and so on. Further, based on the level of Multi-scale morphological gradient
overlapping in the target attribute values, the proposed MSMGF
and STMHT techniques combined with fuzzy judgement trees μmmg(i) = α ⊕ N β i − αΘN β i (8)
(FJT) are proposed as new multiclass disturbance detection and
classification techniques for multiple DG-based microgrid model. Modified multi-scale morphological gradient
It is shown in the performance appraisal section that the
proposed MSMGF and STMHT techniques are faster, μmmg, R(i) = (μmmg, R − 1 ⊕ β)(i) − (μmmg, R − 1Θβ)(i) (9)
computationally efficient, and exhibit the highest general accuracy
in classification (GAC in %) as compared to some of the
with the assumption
conventional SPT's. Finally, the robustness of the proposed
techniques has been validated in the hardware environment in the
μmmg, 0(i) = μg(i) (10)
same section, where the results shown are satisfactory.
and R ∈ I (I is an integer) is the number of recursive iterations
2 Signal processing techniques performed during individual ith instance of the algorithm and
This particular section briefly explains the proposed SPTs for μmmg, R is the multi-morphological gradient at the Rth instance of
detecting the occurrence of a number of multiclass power signal the iteration.
disturbances. Before proceeding, let us first discuss about the Finally, from the above iterative method (6)–(9) the value of the
conventional morphology. proposed MSMGF (μg MSMGF) can be expressed as

2.1 Conventional mathematical morphology μg MSMGF(i) = (μmmg, R ⊕ β)(i) − (μmmg, RΘβ)(i) (11)
The main function of mathematical morphology is to extract In this paper, the value of R is optimised simply with the trial and
specific features in the neighbourhood of every sample. It is hit optimisation. The optimised value of R is 4 for the above test
achieved by the interaction between the signal and structuring system. The shape of the SE is usually chosen according to some
element (SE). As a universal perception of morphology, dilation prior knowledge about the shape of the signal. In this paper, a
and erosion are considered as the two fundamental calculations, trapezoidal SE has been chosen for the detection of disturbances.
which form the foundation for the development of various
operations, especially applicable for power signal analysis [17]. In
this context, the application of closing and/or opening operations 2.3 Conventional HT
along with some of their combinations have been identified as to be HT is a widely used technique for generating an analytical signal
best suited for the single dimension power system signals. The from a real signal. The conventional HT of a signal x(t) is
basic operations of the mathematical morphology are defined in calculated as
(1)–(4) as follows [17, 21, 22]:

∫
∞
1 x(τ)
α(i − j) + β( j), x′(t) = dτ (12)
Y δ(i) = (α ⊕ β)(i) = max (1) π −∞
t−τ
0 ≤ (i − j) ≤ i, j ≥ 0
with a constraint that the integral subsists as a principal value.
α(i + j) + β( j), Alternatively, z is expanded into complex plane as
Y ε(i) = (αΘβ)(i) = min (2)
0 ≤ (i + j) ≤ i, j ≥ 0
z(t) = x(t) + jx′(t) = a(t) e jφt (13)
Y θ(i) = (α ∘ β)(i) = ((αΘβ) ⊕ β)(i) (3)
2.4 Short-time HT
Y ς(i) = (α ∙ β)(i) = ((α ⊕ β)Θβ)(i) (4)
A modified version of HT, i.e. a short-time modified HT (STMHT)
where the operators ⊕ , Θ, ∘ and ∙ refer to the dilation, erosion, [18] employs a window Wα in the interval [t − α/2, t + α/2] to
opening and closing operations between the quantities α(i) and β(j), determine the STHT of the signal A(t), defined as
respectively, which are denoted as ′Y δ′, ′Y ε′, ′Y θ′and′Y ς′,
∫
t + α /2
respectively. Here α(i) and β(j) are the input signal and the SE 1 W α(τ − t)x(τ)
STHT(A(t)) = dτ (14)
defined over the domains Dα = [0,1,2…, i − 1] and Dβ = [0,1,2…, j π t − α /2 t−τ
− 1], respectively, with i > j where ‘i’ and ‘j’ are integers. Actually,
opening operation (3) is simply the dilation of the eroded signal (2) where α is the window width.
and SE, whereas the closing operation (4) is simply the erosion of The value of A(t) is the centre value of the HT of Wα(τ − t)x(τ)
the dilated signal (1) and SE, respectively. in the interval [t − α/2, t + α/2].

2.2 Proposed MSMGF 2.5 Proposed discrete STMHT

A morphological gradient (μg) at a particular instance ‘i’, can be Let us consider a N-point discrete time series A[n] such that n = 0,
defined as the difference between the dilation and erosion 1, 2, 3,…, N − 1. Let each discrete window WY[l] consist of Y
operation, and is expressed as points where l = − (Y − 1)/2, −(Y − 1)/2 + 1,…, (Y − 1)/2 and Y is
odd. The Y-point discrete time series xY,y [i] is substituted by the
μg(i) = Y δ(i) − Y e(i) windowed discrete time series WY[n − y]A[n] and is expressed as
(5)
= (α ⊕ β)(i) − (αΘβ)(i)

Consequently, let us define the multi-scale morphological gradient

(μmmg) in terms of morphological gradient (μg) and the same SE (β)
as

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 83

© The Institution of Engineering and Technology 2017
 On the contrary, a discrete form representation of the STHT is
defined as [18]

xY , y[i] = W Y [n − y]A[n] y + (Y − 1)/2

∑
^
An = A n W Y n − y W E n − y − Y − 1 /2 (23)
n = y − (Y − 1)/2

Y −1 Y −1 Y −1 (15) It exhibits only one multiplication operation, so it must have a

y= , + 1, …, (N − 1) −
2 2 2 higher operation rate. The output signal of (23) is again operated
Y −1 Y −1 with the same Gaussian window function and finally the expression
n= y− :y + for the proposed STMHT for the detection of disturbances, is
2 2
proposed as
Y −1
i=n− y−
2 ~
y + (Y − 1)/2
∑
^
A(n) = A n W Y n − y W E n − y − Y − 1 /2 (24)
^ n = y − (Y − 1)/2
From (15), the discrete STHT of A[n], denoted by A(n) is derived
as
where y = (Y − 1)/2, (Y − 1)/2 + 1,…,(Y − 1)−(Y − 1)/2 which is the
^ Y −1 new contribution to the literature. In any short-time transform, the
A(n) = xY , y ^
(16) size of the data window plays an important role. Smaller data
2
windows are able to enhance the local properties of the signal in
more detail. Let us consider that the size of the windows used in
where x^ Y , y is the HT of the windowed signal W Y [n − y]A[n], with a
STMHT is so small that the section of signal within each data
Gaussian window function of Y is equal to 5, and is mathematically window may be approximated as a straight line with varying slope.
defined as The presence of any disturbance in a signal will result in sudden
changes in the local slope of the signal. Since at any given instant
x^ Y , y = HT(W Y [n − y]A[n]) (17) of time, the STMHT value of the signal is nearly proportional to
the slope, the corresponding value of the STMHT at that instant
Conventional HT consists of a discrete Fourier transform (DFT) will be large. In a microgrid, the frequency deviates occasionally
followed by elimination of the negative frequency components and from the nominal frequency. This deviation of frequency in the
one inverse DFT (IDFT). Conventional HT is computationally case of islanding or power quality disturbances is not large enough
complex. (−0.5 to +0.5 Hz) [23, 24] to cause major alterations in the peak
The Y point discrete time analytic signal of xY,y, and zY,y is amplitudes of voltage and current samples (only phase angle
computed by the Y point IDFT changes to some extent) and hence in the extracted features from
MSMGF and STMHT algorithms. The MSMGF in its most
Y −1 fundamental form is a gradient filter that obtains an approximate
1 j2πik
Y k∑
zY , y i = z~Y , y k exp (18) filtered representation of the signal in the time domain. The values
=0
Y
of the extracted features are determined by the SE chosen for the
transformations. It has been reported in the literature that in power
where i, k = 0, 1, …, Y−1, z~Y , y is DFT of the Y point discrete time system disturbance detection, the choice of SE is determined by
series zY , y and is given by various factors such type of signal, frequency spectrum and
~
sampling rate [25]. For power system applications, it has been
XY, y k for k=0 recommended that the optimal choice should be such that it retains
~ (Y − 1) the feature of interest [26]. On the other hand, STMHT of a signal
z~Y , y k = 2X Y , y k for 1 <= k <= at any instant of time has been shown to have a good correlation
2 (19)
with the instantaneous slope of a signal at that time, and is a
(Y + 1) function of the window size employed [27]. Consequently, the
0 for <= k <= (Y − 1)
2 window size can be optimally chosen so as to minimise the
~
influence of frequency deviations on the features extracted from
where X Y , y is DFT of the Y point discrete time series and given by the STMHT.

Y −1 2.6 Test of proposed STMHT and MSMGF techniques,

~ − j2πik
XY, y k = ∑ xY y i exp
,
Y
(20) respectively, on synthetic signals
i=0
In this section, the performance of the STMHT and MSMGF has
where i, k = 0, 1,…, Y − 1. been tested on 1000 samples of some synthetically generated
Based on (18)–(20), the imagery part [18] of zY , y at (Y − 1)/2, signals, considering 128 samples per cycle (Ns) with a fundamental
^
i.e. X Y , y[(Y − 1)/2] is the HT and computed by frequency (f0) of 60 Hz.
The pure synthetic signal ‘V’ is considered as
((Y − 3)/2) Y −1
1
x^ Y , y[(Y − 1)/2] =
Y ∑ xY , y i + ∑ xY , y i V(k) 0 ≤ k ≤ 1000 = a(k)sin(2 ∗ π ∗ f 0 ∗ k ∗ ts) (25)
i=0 i = ((Y + 1)/2)
(21)
Voltage sag was created between the sampling instance
∗ WE i 400 ≤ k ≤ 590 setting a1(k) = 0.5 by altering (25) as

Therefore, x^ Y , y (Y − 1)/2 is equivalent to V(k) 400 ≤ k ≤ 590 = a1(k)sin(2 ∗ π ∗ f 0 ∗ k ∗ ts) (26)

Y −1 Fig. 1 shows the performance comparison of the proposed STMHT

x^ Y , y[(Y − 1)/2] = ∑ xY y i WE i
, (22) and MSMGF as well as the conventional technique WT. From the
i=0 figure, it is clear that both the proposed methods are very much
promising. Adding to this, the formation of the proposed STMHT
where i = 0, 1, 2, …, Y − 1. and MSMGF technique is very simple with low computational
complexity (as compared to the conventional WT).

84 IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94

© The Institution of Engineering and Technology 2017
Fig. 1 Performance comparison of the proposed STMHT and MSMGF with conventional WT over synthetic signals subjected to voltage Sag with 128
samples/cycle

Event 7 (E7): Voltage swell + harmonics, by closing cb5.

Event 8 (E8): Unbalanced load switching, by opening cb7.
Event 9 (E9): Balanced load switching, by opening cb7.
Event 10 (E10): Flicker, by opening cb6.

The three-phase voltage and current samples collected from the

relay points r1, r2, r3 and r4 are processed through the proposed
and conventional algorithms to develop the disturbance detection
techniques, which are explained in details in the following sections.

3.1 Performance of proposed STMHT, MSMG and

conventional WT techniques, respectively, on signal from the
test system
In this paper, four operating conditions (Z) have been defined in
terms of wind speed (m/s) and solar irradiance (W/m2), i.e. Z1: 8
m/s and100 W/m2; Z2: 8 m/s and 1000 W/m2; Z3: 20 m/s and 100
W/m2; and Z4: 20 m/s and 1000 W/m2, respectively. Figs. 3a–d and
Fig. 2 Block diagram representation of the multiple DG-based standard 4a–d, respectively, illustrate the comparison of the proposed as
IEC microgrid test system well as conventional techniques for different disturbances (E1, E2,
E3, E6 and E8) in the microgrid model (Fig. 2) subjected to
3 Test system operating conditions (Z1 and Z4), respectively. It can be illustrated
from these figures, that the proposed MSMGF and STMHT
Original research papers submitted to the IET Research Journals techniques, respectively, are very much efficient in detecting the
should conform to the IET Research Journals Length Policy [2]. occurrence of the events. Combination of the disturbances and
The length guidelines include the abstract, references and operating conditions, as described above, and subjected to the
appendices but do not include figure captions or table content. noise-free condition yields to 1300 cases. Only a few simulation
Fig. 2 illustrates the block diagram representation of the proposed results have been shown in order to maintain the limit of the paper.
multiple DG-based standard IEC microgrid test system [28]. The When a power system disturbance, such as islanding occurs, the
model consists of multiple DG's, i.e. DFIG-based wind farm [29] grid frequency is not constant, but the proposed algorithm
and solar photovoltaic (PV) farm nomenclature as DG 1, 2 and DG accurately detects the occurrence of the event. Fig. 3b shows the
3, 4, respectively, which are connected in a messed structure at the grid frequency change when islanding occurs. The DFIG model
pcc (circuit breakers (cb), cb5 and cb6 off). To formulate a feasible has been developed with 128 samples/cycle. Some results are given
microgrid structure, a dc battery [30] has been installed at the 25 in Figs. 1a, b, 3b and 4b. In these four cases, the classification
kV pcc through a converter (C) C1. The pcc has been connected to accuracy has been tested and the results are given in Table 5b.
a 120 kV utility with the help of 25 kV/120 kV transformer (TR)
TR1. The detailed values of the network parameters and simulation 3.2 Target feature extraction
are mentioned in the Appendix. A number of events have been
created varying the wind speed (between 8 and 20 m/s with a In this particular section, the process of extracting the target feature
frequency of 1 m/s) and solar irradiance (100–1000 W/m2 with a sets for the test system using the proposed MSMGF and STMHT
techniques, respectively, is explained. The sampled values of the
frequency of 100 W/m2), by operating the cb's, which are
voltage and current signals near the target DG's are collected from
explained as under
the respective relays, for a large number of operating conditions
(by varying the wind speed and the solar irradiance) for the test
Event 1 (E1): Islanding by opening cb1.
system subjected to disturbances as explained in Section 3. These
Event 2 (E2): Three-phase fault, by closing cb2. signals are then passed through the proposed MSMGF and STMHT
Event 3 (E3): Capacitor bank switching, by closing cb3. techniques, respectively, and simple computations over their
Event 4 (E4): Voltage sag by opening cb4. outputs extract a number of features, out of which the most
Event 5 (E5): Voltage swell, by closing cb4. significant features are accumulated to form the target feature set.
The target features extracted are explained as below [31]:
Event 6 (E6): Voltage sag + harmonics, by opening cb5.

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 85

© The Institution of Engineering and Technology 2017
 Fig. 4 Performance comparison of the proposed STMHT and MSMG with
WT over ‘a’ phase voltage sample subjected to power quality events
(a) Capacitor switching (E3), (b) Voltage Sag with harmonics (E6), (c) Unbalanced
load switching (E8) with 128 samples/cycle

is defined as the mean of the data values of z(M).

Fig. 3 Performance comparison of the proposed STMHT and MSMG with
WT over ‘a’ phase voltage sample subjected to The ‘a’ phase voltage samples near the target DG's have been taken
(a) Islanding (E1), (b) Frequency plot of islanding, (c) LLL-fault (E2) with 128
as the input to the proposed MSMGF and STMHT techniques in
samples/cycle
this paper. Out of a large number of simulations, the target feature
sets for a selected number of operating conditions are shown in
A. Energy of the signal (x1): For a signal z(t), with samples zm with Tables 1 and 2.
In case of the proposed STMHT technique, it can be illustrated
m = 1, 2,3…, M, its energy is defined as
from the table that there is no overlapping in the target feature sets.
M On the contrary, the proposed MSMGF technique exhibits some
Energy = x1 = ∑ zm 2
(27) overlapping in the target feature sets, only for the some power
quality events. This is significantly illustrated in Figs. 5a–d,
m=1
B. Entropy (x2): Entropy is simply the average amount of the respectively. Based on this, it can be noted that all the classes can
be directly classified for the proposed STMHT. On the contrary, in
information from the event
proposed MSMGF, some of the events, such as islanding (E1),
M faults (E2) and power quality events such as unbalanced load
Entropy = x2 = ∑ zmlogb zm (28) switching (E8), balanced load switching (E9) and flicker (E10),
respectively, can be easily classified. However, for distinctively
m=1
classifying the voltage sag, swell, sag with harmonics and swell
where M = total sample number. with harmonics, we certainly need to fuzzify the boundaries of the
C. Maximum amplitude of the signal (x3): It is defined as the target features (E3, E4, E5, E6 and E7).
maximum value of the output of the STMHT and MSMGF over the
entire input signal. 4 Proposed multiclass classifier
D. Standard deviation (x4): It is defined as an attribute that 4.1 Fuzzy judgment tree
enumerates the amount of deviation in a set of data values.
Mathematically As can be revealed from Table 1, the maximum and minimum
values of the target parameters for all the events from E1 to E10,
M under different wind speeds and solar power, respectively, are
1
M m∑
x4 = (zm − z^ μ)2 (29) distinctive for STMHT. In case of the proposed MSMGF
=1 technique, a simple rigid judgment tree produces some incorrect
multiclass classifications, mainly due to the overlapping in the
where attribute values. In this regard, the application of the FJT has been
considered as a resourceful method to cover up the skilful
M information's in order to engross promising outcomes. To increase
1
M m∑
z^ μ = zm (30) the accuracy, a certainty factor (CF)-based fuzzy judgment tree is
=1 used instead of simple fuzzy logic-based classifier.

86 IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94

© The Institution of Engineering and Technology 2017
Table 1 Values of target parameter set for some random cases of STMHT
DG OC Class X1 X2 X3 X4 DG OC Class X1 X2 X3 X4
DG1 Z1 E1 0.6002 1.5901 0.2712 0.0696 DG1 Z1 E6 0.0302 1.2801 0.0452 0.0143
DG1 Z2 E2 0.1976 1.4301 0.2215 0.0389 DG1 Z2 E7 0.0727 1.3761 0.1403 0.0284
DG1 Z3 E3 0.2745 1.3021 0.2536 0.0458 DG1 Z3 E8 0.0009 0.9309 0.0129 0.0026
DG1 Z4 E4 0.0195 1.1050 0.0494 0.0129 DG1 Z4 E9 0.0094 1.6298 0.0271 0.0087
DG1 Z2 E5 0.0451 1.2124 0.1282 0.0199 DG1 Z2 E10 0.0079 0.7513 0.0082 0.0023
DG2 Z1 E1 0.5959 1.5826 0.2697 0.0691 DG2 Z1 E6 0.0298 1.2712 0.0441 0.0135
DG2 Z2 E2 0.2003 1.4411 0.2316 0.0397 DG2 Z2 E7 0.0733 1.3829 0.1409 0.0286
DG2 Z3 E3 0.2787 1.3054 0.2532 0.0461 DG2 Z3 E8 0.0010 0.9329 0.0131 0.0027
DG2 Z4 E4 0.0204 1.1007 0.0537 0.0131 DG2 Z4 E9 0.0096 1.6314 0.0274 0.0087
DG2 Z1 E5 0.0505 1.2010 0.1310 0.0204 DG2 Z1 E10 0.0083 0.8106 0.0079 0.0026
DG3 Z1 E1 0.5832 1.5762 0.2662 0.0684 DG3 Z1 E6 0.0307 1.3043 0.0459 0.0145
DG3 Z2 E2 0.1989 1.4596 0.2399 0.0390 DG3 Z2 E7 0.0701 1.3509 0.1402 0.0279
DG3 Z3 E3 0.2798 1.2991 0.2413 0.0485 DG3 Z3 E8 0.0009 0.9308 0.0129 0.0026
DG3 Z4 E4 0.0186 1.0988 0.0457 0.0128 DG3 Z4 E9 0.0098 1.6337 0.0278 0.0089
DG3 Z3 E5 0.0421 1.2291 0.1245 0.0194 DG3 Z3 E10 0.0082 0.8023 0.0076 0.0024
DG4 Z1 E1 0.5811 1.5635 0.2601 0.0672 DG4 Z1 E6 0.0314 1.3161 0.0461 0.0151
DG4 Z2 E2 0.2076 1.4484 0.2344 0.0402 DG4 Z2 E7 0.0745 1.4009 0.1418 0.0297
DG4 Z3 E3 0.2731 1.2914 0.2401 0.0431 DG4 Z3 E8 0.0008 0.9304 0.0128 0.0025
DG4 Z4 E4 0.0203 1.1108 0.0530 0.0132 DG4 Z4 E9 0.0099 1.6566 0.0281 0.0091
DG4 Z2 E5 0.0501 1.2341 0.1309 0.0202 DG4 Z2 E10 0.0079 0.7827 0.0075 0.0024

Table 2 Values of target parameter set for some random cases of MSMGF
DG OC Class X1 X2 X3 X4 DG OC Class X1 X2 X3 X4
DG1 Z1 E1 4.1818 1.1843 0.0300 0.3101 DG1 Z1 E6 4.4831 0.2691 0.0056 0.1314
DG1 Z2 E2 5.8212 1.0405 0.0714 0.4842 DG1 Z2 E7 4.4017 0.3421 0.0015 0.1238
DG1 Z3 E3 3.5470 0.8547 0.0096 0.1565 DG1 Z3 E8 0.0276 0.9811 0.0163 0.0180
DG1 Z4 E4 3.4982 0.3554 0.0030 0.1311 DG1 Z4 E9 0.0901 1.1037 0.0318 0.0127
DG1 Z2 E5 3.3799 0.2247 0.0012 0.1190 DG1 Z2 E10 0.0103 0.0757 0.0205 0.0025
DG2 Z1 E1 4.2652 1.2156 0.0319 0.3257 DG2 Z1 E6 4.4956 0.3061 0.0059 0.1418
DG2 Z2 E2 5.9158 1.0470 0.0741 0.4919 DG2 Z2 E7 4.4328 0.4501 0.0019 0.1292
DG2 Z3 E3 3.5381 0.8510 0.0093 0.1510 DG2 Z3 E8 0.0292 0.9849 0.0169 0.0190
DG2 Z4 E4 3.4677 0.3441 0.0035 0.1378 DG2 Z4 E9 0.0870 1.0154 0.0301 0.0122
DG2 Z1 E5 3.3804 0.2323 0.0013 0.1205 DG2 Z1 E10 0.0114 0.0799 0.0215 0.0028
DG3 Z1 E1 4.4414 1.2934 0.0392 0.3526 DG3 Z1 E6 4.5001 0.3086 0.0060 0.1441
DG3 Z2 E2 6.2521 1.0586 0.0785 0.5211 DG3 Z2 E7 4.4019 0.3400 0.0014 0.1301
DG3 Z3 E3 3.5675 0.9002 0.0099 0.1608 DG3 Z3 E8 0.0301 0.9904 0.0158 0.0200
DG3 Z4 E4 3.4346 0.3486 0.0029 0.1386 DG3 Z4 E9 0.0935 1.0283 0.0367 0.0134
DG3 Z3 E5 3.3511 0.2464 0.0009 0.1204 DG3 Z3 E10 0.0118 0.0821 0.0218 0.0037
DG4 Z1 E1 4.3611 1.2774 0.0378 0.3383 DG4 Z1 E6 4.4910 0.3002 0.0058 0.1409
DG4 Z2 E2 6.1756 1.0538 0.0764 0.5017 DG4 Z2 E7 4.3765 0.2816 0.0011 0.1225
DG4 Z3 E3 3.5895 0.9912 0.0107 0.1690 DG4 Z3 E8 0.0302 0.9914 0.0169 0.0196
DG4 Z4 E4 3.4262 0.3290 0.0028 0.1361 DG4 Z4 E9 0.0967 1.0295 0.0389 0.0139
DG4 Z2 E5 3.3881 0.2431 0.0014 0.1221 DG4 Z2 E10 0.0120 0.0881 0.0229 0.0039

The ability to split down a composite judgment into a criterion is achieved which indicates the procedure to exit. The
combination of some simpler decisions with interpretable outcomes selection of the stopping criterion is formulated in such a manner
and the potential to deal with uncertain information by fuzzy that each infected node can split a high of ten annotations. The
techniques are the most significant advantages of the FJT. In most appropriate condition formulated at each individual node in
contrast to the conventional single-staged classifiers, as the this paper utilises the Gini's diversity index [20]. Division between
computations of the FJT are used only for an unambiguous split of events rather than node miscellany is made possible for multiclass
prototypes. Hence it exhibits a reduced complexity. The hard classification, using the towing rule base. Splits preserving the
margins of the rigid judgment tree (RJT) are fuzzified based on correlated prototypes are strictly chosen.
some suitable membership functions. In order to establish an Adding to it, for a number of power quality problems with δ
efficient classification result based on a widespread investigation of number of patterns, with the percentage of annotations of νth
data sets, the judgment tree glances for all possible splits, pattern at ξth node defined as pδξz, the Gini's diversity index is
beginning from the initial node. On the whole, the data sets in defined as
individual iteration are split into two child nodes, until a stopping

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 87

© The Institution of Engineering and Technology 2017
 Fig. 6 Plot showing the membership functions for the features
(a) x2, (b) x3, (c) x4, for E3, E4 and E5 and (d) x4 for E6 and E7

Fig. 5 Plot showing the separation of classes E1–E10 by features x3 and 0 if u < u1
x2
(u − u1)
(a), (b) For MSMGF, (c), (d) For STMHT if u1 ≤ u < u2
(u2 − u1)
μ(u1, u2, u3) = (32)
δ (u3 − u)
if u2 ≤ u ≤ u3
~
gDI = ∑ pδξ[1 − pδξ] (31) (u3 − u2)
ν=1
0 if u3 < u
~
gDI is further vulnerable to alterations in the likelihood at each
The trapezoidal membership function with legs u1, u2, u3, u4 and u
node than to the misclassification imprecision. The size of the tree
as input, respectively, is defined as
(petite or hefty) considerably effects the swapping between the
highest precision in classification (HPC) and the computational 0 if u < u1
complexity associated with the technique. Further, the stiff
judgement regulations (if then else regulations) for creating the (u − u)
μ(u1, u2, u3, u4) = 1 if u1 ≤ u < u2 (33)
judgment tree is finely formulated by choosing the most important (u2 − u1)
features so as to obtain HPC. In the next step, the stiff judgment 1 if u2 ≤ u
rules of the RJT are fuzzified using triangular and trapezoidal
membership functions to formulate the FJT. The triangular The membership functions for various fuzzy sets of the features
membership function with legs u1, u2, u3 and u as input, x1–x4 are illustrated in Figs. 6a–d and the detailed value of the
respectively, is defined as membership points of the triangular and trapezoidal membership
functions are mentioned in Table 3.
As depicted in the literature [11], the application of CF based on
support and confidence indices, for deriving the association rules

88 IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94

© The Institution of Engineering and Technology 2017
Table 3 Detailed value of the membership points of the triangular and trapezoidal membership functions as shown in Fig. 7
Member shipname Class value a b c d
μ(x2)A1 E5 0.0000 0.1750 0.3500 —
μ(x2)A2 E4 0.1210 0.2668 0.4127 —
μ(x2)A3 E3 0.2475 0.6237 1.0000 1
μ(x3)B1 E5 0.0000 0.1230 0.2460 —
μ(x3)B2 E4 0.2250 0.2530 0.2810 —
μ(x3)B3 E3 0.2420 0.3360 0.4300 0.43
μ(x4)C1 E5 0.0000 0.0375 0.075 —
μ(x4)C2 E4 0.0120 0.056 0.1000 —
μ(x4)C3 E3 0.0400 0.1800 0.3200 0.32
μ(x4)D1 E6 0.1800 0.2100 0.2400 —
μ(x4)D2 E7 0.2300 0.2500 0.2700 —

Table 4 Detailed value of the CFs for fuzzy rules for some randomly selected feature sets for each class
Class E3 E4 E5 E6 E7
KFR1 0.979 0.011 0.010 0.966 0.034
KFR2 0.015 0.961 0.024 0.022 0.978
KFR3 0.014 0.021 0.965 — —

of the form Aα → Cα with ‘A’ as the antecedents and ‘C’ as then ‘E5’ with CF as KFR1.
consequents of the ‘α’ assessment rules, is very much significant. Rule2: If x2 is A2 && x3 is B2 && x4 is C2,
The support (Sα) of a particular rule ‘αth ‘ is an indication of the then ‘E4’ with CF as KFR2.
number of instances the consequents (Cα) and the antecedents (Aα) Rule3: If x2 is A3 && x3 is B3 && x4 is C4,
occur simultaneously, and mathematically defined as
then ‘E3’ with CF as KFR3.
Sα(Aα → Cα) = p(Aα ∩ Cα) Rule4: If x4 is D2,
∑n ∈ ζ μAi(Xn) (34) then ‘E6’ with CF as KFR4.
= Rule5: If x4 is D1,
ζ
then ‘E7’ with CF as KFR5.
where ‘i’ is the number of membership functions defined for a ‘n’
number of attributes (X) for classification, respectively. Then the firing strength of the individual fuzzy rules is obtained
In a similar manner, the confidence (Cα) of a particular rule ‘α’ in the following manner:
is an indication of the probability that the consequents (Cα) strictly
follows the respective antecedents (Aα) of the assessment rules, and Rule1: β1 = min (µ(x2)A1, µ(x3)B1, µ(x4)C1)
mathematically defined as Rule2: β2 = min (µ(x2)A2, µ(x3)B2, µ(x4)C2)
Rule3: β3 = min (µ(x2)A3, µ(x3)B3, µ(x4)C3)
p(Aα ∩ Cα) Rule4: β4 = (µ(x4)D2)
CFα(Aα → Cα) =
p(Aα) Rule5:
∑n ∈ ζ μAi(Xn) (35)
= ζ β5 = (μ(x4)D1) (38)
∑n = 1 μAi(Xn)
Finally, the decision making variables for the proposed FJT is
Finally, the CF (KFα) for particular rule ‘α’ is defined as derived as

KFα = CFα(Aα → Cα) − ρα (36) α1 = β1 ∗ KFR1, α2 = β2 ∗ KFR2, α3 = β3 ∗ KFR3, α4

= β4 ∗ KFR4, α5 = β5 ∗ KFR5 (39)
where ρα is the average confidence value for each rule defined as
Thus the final decision rule for classification using FAT can be
1
ζ written as
ζ − 1 n∑
ρα = CFα(Aα → Cα) (37)
=1 Decision 1 (D1): if α1 > α2 && α1 > α3,
n≠ζ
then disturbance class is ‘E5’
4.2 CF-based fuzzy rule base Decision 2 (D2): if α2 > α1 && α2 > α3,
then disturbance class is ‘E4’
The fuzzy judgment rules for the FJT comprise the rule base for
Decision 3 (D3): if α3 > α1 && α3 > α2,
multiclass event classification.
In the next step, we associate each of the fuzzy rules with a CF. then disturbance class is ‘E3’
Table 4 shows the CF of the fuzzy rules for some randomly Decision 4 (D4): if α4 > α5,
selected feature sets for each class. then disturbance class is ‘E6’
Based on these CFs, the fuzzy rule base for classification of the Decision 5 (D5): if α5 > α4,
events by FJT is formulated as then disturbance class is ‘E7’
Rule1: If x2 is A1 && x3 is B1 && x4 is C1,

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 89

© The Institution of Engineering and Technology 2017
Fig. 7 FJR-based threshold selection and event classification for
(a) STMHT and (b) MSMG

It is significantly observed from Table 4, that the CF of a particular The GAC (%) for each of the classes for ‘a’ phase voltage samples
rule is highest for the class which corresponds to the consequent of at different DG locations as input to the proposed classifiers, as
that particular rule. Figs. 7a and b describe the simple FJT-based illustrated in Table 5. As can be depicted from this table, maximum
event classification for STMHT and MSMG, respectively. accuracy is obtained with the proposed STMHT technique for most
of the classes. However, it is important to note that, the GAC (%)
5 Performance evaluation of proposed classifier slightly dwindles with the proposed MSMGF technique. The
superiority of the proposed classifier in terms of the GAC (%) for a
The pertinent attributes were extorted from the proposed STMHT large number of power system disturbances over some
and MSMGF techniques, respectively, for each training class. In conventional methods is illustrated in Table 6. It is very much
the next step, these attributes were engaged to form the FJT. The important to note from this table, that the GAC (%) for the
fuzzy laws, as developed from the FJT, are used to endorse some proposed STMHT and MSMGF techniques, respectively, is very
random test classes from the target feature set. much superior to the existing methods [11, 15, 32–37]. The
The general accuracy in classification (GAC in %) has been performance of the GAC (%) for the real-time implementation is
considered as the index to evaluate the performance of the also very much significant. Further, a confusion matrix (which
classification algorithms, which is defined as below: takes into account the overall data sets for all the classes) has been
presented for the proposed STMHT and MSMGF techniques,
Number of classes classified correctly respectively, in Tables 7a and b, where the diagonal matrix
GAC (%) = × 100 (40)
Number of misclassifications indicates the correctly classified class, whereas the off-diagonal
matrix indicates the misclassifications. The studies on dynamic
Some random feature sets (from the overall target feature set) are accuracy have also been done for all the power quality events
passed through the proposed certainty factor-based FJT classifier. according to IEEE std. 1159-2009. However, for space limitation

90 IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94

© The Institution of Engineering and Technology 2017
Table 5a Continued
Events (E) 128 samples/cycle GAC (%) by STMHT
DG1 DG2 DG3 DG4 DG1 DG2 DG3 DG4
(a) GAC (%) for different classes at different DG locations with 128 samples/cycle sampling rate
islanding (E1) 100 100 100 100 100 100 100 100
LLL-fault (E2) 100 100 100 100 100 100 100 100
capacitor switching (E3) 100 100 100 100 98.51 99.26 98.41 99.18
voltage sag (E4) 99.34 100 100 100 98.89 98.77 99.6 99.10
voltage swell (E5) 100 99.20 100 100 98.71 99.31 98.76 99.43
voltage sag with harmonics (E6) 100 100 100 100 99.21 98.72 99.39 99.27
voltage swell with harmonics (E7) 99.30 100 99.24 100 99.2 98.6 99.0 98.6
unbalanced load switching (E8) 100 100 100 100 99.2 100 100 100
balanced load (E9) 100 100 100 100 100 100 100 100
flicker (E10) 100 99.30 100 99.27 100 99.3 100 100

Table 5b Continued
Percentage of sag (E1), % Average classification accuracy (%) (duration 1–30 cycles)
DG1 DG2 DG3 DG4
(b) Classification accuracy of STMHT with different percentages of Sag according to IEEE std. 1159-2009
10 99.1 99 98.5 98.9
20 99.1 99.1 98.7 99
30 100 100 100 100
40 100 100 100 100
50 100 100 100 100
60 100 100 100 100
70 100 100 100 100
80 100 100 100 100
90 100 100 100 100
Avg 99.80 99.78 99.68 99.76

Table 5c Continued
Class GAC (%) by STMHT static threshold
DG1 DG2 DG3 DG4 DG1 DG2 DG3 DG4
(c) GAC (%) by STMHT for three different PQ disturbances at different DG locations with dynamic threshold according to IEEE std. 1159-2009
E3 100 100 100 100 98.86 99.31 99.28 98.81
E4 100 99.2 100 100 99.80 99.78 99.68 99.76
E7 100 99.2 99.2 100 98.71 98.89 99.17 98.10

Table 5d GAC (%) for each of the classes for ‘a’ phase voltage samples at different DG locations as input to the proposed
classifiers
Class GAC (%) by MSMGF static threshold
DG1 DG2 DG3 DG4 DG1 DG2 DG3 DG4
(d) GAC (%) by MSMGF for three different PQ disturbances at different DG locations with dynamic threshold according to IEEE std. 1159-2009
E3 98.2 99 98.3 99.1 97.91 98.29 98.12 97.81
E4 98.4 99.1 98.6 99.3 98.79 98.37 99.22 98.81
E7 99.2 99.2 98.4 98.4 98.32 98.79 97.75 97.91

all the results are not given, but results for only three events are the total computation requires [2(N − Y)Y] multiplications and [(N
given. The detailed classification results have been shown for − Y)Y] number of divisions and subtractions, where N is the total
voltage sag with STMHT and the results are given in Table 5c. The number of sample points of the discrete signal and each discrete
results of GAC (%) for three different PQ disturbances at different window consists of Y points. STMHT and MSMGF techniques do
DG locations with Dynamic threshold according to IEEE std. not involve decomposing the signal into multiple frequency bands
1159-2009 are given in Table 5d. which wavelet does, and hence the number of extracted features is
However, when it comes to power system analysis, it becomes higher in the case of WT than in MSMGF or STMHT. A detailed
more critical to evaluate the execution speed of an algorithm for comparison of the detection time (DT) for the conventional
timely detection of power quality disturbance. (wavelet) as well as the proposed (STMHT and MSMGF,
Thus, to establish the superiority of a mathematical analysis respectively) techniques has been presented in Table 8 which is
technique over existing methodologies, the simplest approach is to less than the threshold value of 2.0 s [24]. It is important to note
reduce the execution time [1, 32, 38, 39] which has been focused in that the computations have been performed in MATLAB
the author's paper. When calculating MSMGF, which include only environment in a computer with 2Duo processor (2.94 GHz), and
additions and subtractions, the total computation requires [8n(N − 2 GB Random Access Memory. As can be depicted from this table,
n + 1) + 8] additions and [8n(N − n + 1) + 8] subtractions for an N both the proposed STMHT and MSMGF techniques, respectively,
sample signal. Where n is the length of the SE. In case of STMHT,

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 91

© The Institution of Engineering and Technology 2017
Table 6 Performance of proposed method compared to some existing methods
Type of classifier with different detection algorithms GAC, % Year of publication
ST with probabilistic neural network 97.4 2008
2000
WT with neural network 94.37 1996
neural network 95.93 2002
WT with neuro fuzzy 96.50 2002
wavelet and fuzzy logic 98.70 2009
HHT with probabilistic neural network 97.22 2006
wavelet packet with SVM 97.25 2010
DWT with WT 98.18 2013
ST with fuzzy decision tree 98.66 2015
TT transform with SVM 98.08 2015
morphology with SVM 99.23 2015
cross-spectrum analysis with SVM 99.60 —
CF-based FAT, with MSMGF (proposed) 99.05 —
decision tree-based STMHT (proposed) 99.70 —
CF-based FAT, with MSMGF (proposed) with real-time implementation 93.1 —
decision tree-based STMHT (proposed) with real-time implementation. 93.7 —

Table 7 Confusion matrix for different classes for STMHT (proposed) and MSMGF (proposed)
Class E1 E2 E3 E4 E5 E6 E7 E8 E9 E10
(a) Confusion matrix for different classes for STMHT (proposed)
E1 130 0 0 0 0 0 0 0 0 0
E2 0 130 0 0 0 0 0 0 0 0
E3 0 0 130 0 0 0 0 0 0 0
E4 0 0 0 130 0 0 0 0 0 0
E5 0 0 0 0 130 0 0 0 0 0
E6 0 0 0 0 1 129 0 0 0 0
E7 0 0 0 0 0 0 130 0 0 0
E8 0 0 0 0 0 0 0 130 0 0
E9 0 0 0 0 0 0 1 0 128 1
E10 0 0 0 0 0 0 0 0 0 130
(b) Confusion matrix for different classes for MSMGF (proposed)
E1 130 0 0 0 0 0 0 0 0 0
E2 0 130 0 0 0 0 0 0 0 0
E3 0 0 129 0 1 0 0 0 0 0
E4 0 0 1 129 0 0 0 0 0 0
E5 0 0 0 0 130 0 0 0 0 0
E6 0 0 0 0 0 129 1 0 0 0
E7 0 0 0 0 0 1 129 0 0 0
E8 0 0 0 0 0 0 0 130 0 0
E9 0 0 0 0 0 0 0 0 130 0
E10 0 0 0 0 0 0 0 0 0 130

illustrate a faster response as compared to the conventional one disturbances with a much reduced computational complexity and
with an advantage of involving less integral computations. hence are potential candidates for real-time applications.
To further verify the real time capability of the proposed
classification algorithms, they are implemented on a 6 Conclusion
TMS320C6713 Starter Kit (DSK) [40] and embedded with
MATLAB coder. TMS 320C6713 is from Texas instruments and a This paper proposes two advanced SPTS, i.e. MSMGF and
32 bit floating point digital signal processor. The real-time data STMHT techniques, respectively, to detect and classify the
exchange (RTDX) had been used to exchange the data between the occurrence of multiclass power signal disturbances in a multiple
host pc and the target DSK for analysing the real-time DG-based microgrid network subjected to various operating
implementation. To provide the interface between PC & DSK, a conditions. Simpler in formulation, reduced computational
software tool Embedded Target for TI C6000 DSP and MATLAB complexity, faster in response and so on, are some of the
link for CCS is used by the RTDX. Some real-time implementation advantages of the proposed STMHT and MSMGF techniques,
results are given in Figs. 8a–c for islanding. From Figs. 8a–c, it is respectively, which are very much significant as compared to
clear that both the algorithms detect the occurrence of the events conventional SPT's. Further depending upon the overlapping in the
accurately in real-time environment. The accuracy obtained with values of the target feature sets, a fuzzy judgment tree has been
the real-time implementation is given in Table 6. Therefore the incorporated for classifying the multiclass disturbance patterns. As
proposed STMHT, MSMGF and FJT-based multiclass classifiers can be illustrated in the performance evaluation section, as
are very much reliable and accurate, as they exhibit the superior compared to all the techniques, STMHT demonstrates the
capability of GAC (%) for classifying a number of power system significant (highest) GAC (%) for all the multiclass disturbance
patterns. On the contrary, the proposed MSMGF, though has

92 IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94

© The Institution of Engineering and Technology 2017
 Table 8 Detection time for different events with the
proposed algorithm MSMGF and STMHT
Detection Events Time for detection,
technique ms
Wavelet islanding 26.9
LLL-fault 27.1
Sag with harmonic 28
capacitor switching 29.2
flicker 29.8
unbalanced load switching 29
MSMG islanding 20.2
LLL-fault 20.6
sag with harmonic 24
capacitor switching 24.4
flicker 22.7
unbalanced load switching 23
STMHT islanding 22.9
LLL-fault 23.2
sag with harmonic 24.9
capacitor switching 25.3
flicker 23.7
unbalanced load switching 24

7 References
[1] Mohanty, S.R., Kishor, N., Ray, P.K., et al.: ‘Comparative study of advanced
signal processing techniques for islanding detection in a hybrid distributed
generation system’, IEEE Trans. Sustain. Energy, 2015, 6, (1), pp. 122–131
[2] Ray, P.K., Kishor, N., Mohanty, S.R.: ‘Islanding and power quality
disturbance detection in grid-connected hybrid power system using wavelet
and-transform’, IEEE Trans. Smart Grid, 2012, 3, (3), pp. 1082–1094
[3] Hosseini, S.A., Amjady, N., Velayati, M.H.: ‘A Fourier based wavelet
approach using Heisenberg's uncertainty principle and Shannon's entropy
criterion to monitor power system small signal oscillations’, IEEE Trans.
Power Syst., 2015, 30, (6), pp. 3314–3326
[4] Logenthiran, T., Srinivasan, D., Khambadkone, A.M., et al.: ‘Multiagent
system for real-time operation of a microgrid in real-time digital simulator’,
IEEE Trans. Smart Grid, 2012, 3, (2), pp. 925–933
[5] Papaioannou, D.I., Papadimitriou, C.N., Dimeas, A.L., et al.: ‘Optimization &
Sensitivity Analysis of Microgrids using HOMER software – A Case Study’.
2014, pp. 35–37
[6] Lu, B., Li, Y., Wu, X., et al.: ‘A review of recent advances in wind turbine
condition monitoring and fault diagnosis’, IEEE Power Electron. Mach. Wind
Appl., 2009, pp. 1–7
[7] Milczarek, A., Malinowski, M., Guerrero, J.M.: ‘Reactive power management
in islanded microgrid – proportional power sharing in hierarchical droop
control’, IEEE Trans. Smart Grid, 2015, 6, (4), pp. 1631–1638
Fig. 8 TMS 320C6713 Starter Kit (DSK)-based output for islanding [8] Bayrak, G., Kabalci, E.: ‘Implementation of a new remote islanding detection
detection with both the proposed algorithms method for wind–solar hybrid power plants’, Renew. Sustain. Energy Rev.,
2016, 58, pp. 1–15
(a) a phase voltage signal, (b) STMHT and (c) MSMG [9] Senroy, N., Suryanarayanan, S., Ribeiro, P.F.: ‘An improved Hilbert–Huang
method for analysis of time-varying waveforms in power quality’, IEEE
smaller misclassification (as compared to STMHT and Trans. Power Syst., 2007, 22, (4), pp. 1843–1850
[10] Krishna, B.V., Baskaran, K.: ‘Parallel computing for efficient time–frequency
conventional wavelet), is computationally efficient, and has feature extraction of power quality disturbances’, IET Signal Process., 2013,
illustrated a faster response, with lowest detection time, for almost 7, (4), pp. 312–326
all the multiclass power system disturbances. Further, the [11] Kumar, R., Singh, B., Shahani, D.T., et al.: ‘Recognition of power-quality
robustness of the proposed hybrid detection and classification disturbances using S-transform-based ANN classifier and rule-based decision
tree’, IEEE Trans. Ind. Appl., 2015, 51, (2), pp. 1249–1258
techniques have been implemented and validated in real-time [12] Benmalek, M., Charef, A.: ‘Digital fractional order operators for R-wave
hardware-in-loop environment, where the outputs shown are detection in electrocardiogram signal’, IET Signal Process., 2009, 3, (5), pp.
satisfactory, and consequently vindicates the proposed algorithms 381–391
as potential candidates for real-time applications. [13] Granados-Lieberman, D., Garcia-Perez, A., Cabal-Yepez, E.: ‘Techniques and
methodologies for power quality analysis and disturbances classification in
power systems: a review’, IET. Gener. Transm. Distrib., 2011, 5, (4), pp. 519–
529
[14] Tse, N.C., Chan, J.Y., Lau, W.H., et al.: ‘Hybrid wavelet and Hilbert
transform with frequency-shifting decomposition for power quality analysis’,
IEEE Trans. Instrum. Meas., 2012, 61, (12), pp. 3225–3233
[15] Chilukuri, M.V., Dash, P.K.: ‘Multiresolution S-transform-based fuzzy
recognition system for power quality events’, IEEE Trans. Power Deliv.,
2004, 19, (1), pp. 323–330
[16] Thuraisingham, R.A.: ‘Estimation of Teager energy using the Hilbert Huang
transform’, IET Signal Process., 2015, 9, (1), pp. 82–87
[17] Chen, Y., Ji, T.Y., Li, M.S., et al. ‘Power quality disturbance detection based
on morphology singular entropy’. Int. Conf. IEEE Innovative Smart Grid
Technologies, Asia (ISGT ASIA), 3 November 215, pp. 1–5
[18] Sun, S., Jiang, Z., Wang, H., et al.: ‘Automatic moment segmentation and
peak detection analysis of heart sound pattern via short-time modified Hilbert
transform’, Comput. Methods Programs Biomed., 2014, 114, (3), pp. 219–230

IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94 93

© The Institution of Engineering and Technology 2017
[19] Gautam, S., Brahma, S.M.: ‘Guidelines for selection of an optimal structuring [36] Jaehak, C., Powers, E.J., Grady, W.M., et al. ‘Power disturbance classifier
element for mathematical morphology based tools to detect power system using a rule-based method and wavelet packet-based hidden Markov model’,
disturbances’, IEEE in Power and Energy Society General Meeting, 2012, pp. IEEE Trans. Power Deliv., 2002, 17, (1), pp. 233–241
1–6 [37] Weiming, T., Song, X., Lin, J., et al.: ‘Detection and classification of power
[20] Ribeiro, B., Pereira, H., Almeida, R., et al.: ‘Optimization of sitting posture quality disturbances based on wavelet packet decomposition and support
classification based on user identification’. IEEE 4th Portuguese Meeting on vector machines’. IEEE Proc. 8th Int. Conf. Signal Processing, 2006, vol. 4
Bioengineering (ENBENG), February 2015, pp. 1–6 [38] Jianmin, L., Teng, Z., Tang, Q., et al. ‘Detection and classification of power
[21] Farhan, M.A., Swarup, K.S.: ‘Mathematical morphology-based islanding quality disturbances using double resolution S-transform and DAG-SVMs’,
detection for distributed generation’, IET. Gener. Transm. Distrib., 2016, 10, IEEE Trans. Instrum. Meas., 2016, 65, (10), pp. 2302–2312
(2), pp. 518–525 [39] Shunfan, H., Li, K., Zhang, M.: ‘A real-time power quality disturbances
[22] Zhu, J., Ji, T.Y., Li, M.S., et al.: ‘Detection and classification of power classification using hybrid method based on S-transform and dynamics’, IEEE
disturbances using mathematical morphology with trapezoid structuring Trans. Instrum. Meas., 2013, 62, (9), pp. 2465–2247
elements and signal envelopes’. Int. Conf. IEEE Innovative Smart Grid [40] Chassaing, R.: ‘Digital signal processing and applications with the c6713 and
Technologies, Europe, 2014, pp. 1–6 c6416 DSK’, vol. 16, (John Wiley & Sons, 2004)
[23] Xueqin, Liu, Kennedy, J.M., Laverty, D.M.D., et al.: ‘Wide-area phase-angle
measurements for islanding detection – an adaptive nonlinear approach’,
IEEE Trans. Power Deliv., 2016, 31, (4), pp. 1901–1911 8 Appendix
[24] Safdar, R., Mokhlis, H., Arof, H., et al.: ‘A sensitivity analysis of different
power system parameters on islanding detection’, IEEE Trans. Sustain.
Energy, 2016, 7, (2), pp. 461–470
[25] Gautam, S., Brahma, S.M.: ‘Detection of high impedance fault in power The detailed parameters of the proposed test system are: utility:
distribution systems using mathematical morphology’, IEEE Trans. Power 120 kV, 60 Hz, DG's: DG1 and DG 3: 575 V, 9 MW wind farm (six
Syst., 2013, 28, (2), pp. 1226–1234 1.5 MW doubly-fed induction generator-based wind turbines),
[26] Gautam, S., Brahma, S.M.: ‘Guidelines for selection of an optimal structuring
element for mathematical morphology based tools to detect power system
DG2 and DG 4: 440 V, 1 MW
disturbances’. Proc. IEEE Power & Energy Society General Meeting, 2012, PV solar farms, transformers:
July 2012, pp. 1–6
[27] Ghosh, R., Chatterjee, B., Chakravorti, S.: ‘A method for unambiguous tr1: 50 MVA, 60 Hz, 120 kV/25 kV, Vbase = 25 kV,
identification of on-field recorded insulator leakage current waveforms
portraying electrical activity on the surface’, IEEE Trans. Dielectr. Electr. tr2 and tr4: 10 MVA, 60 Hz, 25 kV/575 V, Vbase = 25 kV,
Insul., 2016, 23, (4), pp. 2156–2164 tr3 and tr5: 2 MVA, 60 Hz, 25 kV/440 V, Vbase = 25 kV.
[28] Kar, S., Samantaray, S.R.: ‘Time-frequency transform-based differential
scheme for microgrid protection’, IET. Gener. Transm. Distrib., 2014, 8, (2),
pp. 310–320 Transmission lines: TL1, TL2, TL3, TL4, TL5, TL6: each 25 kV, 20
[29] Mitra, A., Chatterjee, D.: ‘Active power control of DFIG-based wind farm for MVA, 20 km PI sections, Vbase = 25 kV, r0 = 0.1153 Ω/km, r1 =
improvement of transient stability of power systems’, IEEE Trans. Power
Syst., 2016, 31, (1), pp. 82–93 0.413 Ω/km, l0 = 0.00103 H/km, l1 = 0.00332 H/km, C0 = 0.1133 ×
[30] Li, X., Hui, D., Lai, X.: ‘Battery energy storage station (BESS)-based 10−11 F/km, C1 = 0.0511 × 10−11 F/km.
smoothing control of photovoltaic (PV) and wind power generation
fluctuations’, IEEE Trans. Sustain. Energy, 2013, 4, (2), pp. 464–473 Loads: L1: 10, 3.5 MVAR, L2 and L4: 6 MW, 2.5 MVAR, L3 and
[31] Bercher, J.F., Vignat, C.: ‘Estimating the entropy of a signal with L5: 0.5 MW, 0.2 MVAR.
applications’, IEEE Trans. Signal Process., 2000, 48, (6), p. 1687–1694
[32] Milan, B., Dash, P.K.: ‘Measurement and classification of simultaneous Disturbance elements: Rf = 0.0001 Ω, Cf = 1 MVAR, Pf = 1
power signal patterns with an S-transform variant and fuzzy decision tree’, MW, 0.5 MVAR.
IEEE Trans. Ind. Inf., 2013, 9, (4), pp. 1819–1827 Battery: 1-MWh.
[33] Stuti, S., Mishra, S., Singh, B.: ‘Empirical-mode decomposition with Hilbert
transform for power-quality assessment’, IEEE Trans. Power Deliv., 2009, 24, Converters:
(4), pp. 2159–2165
[34] Masoum, M.A.S., Jamali, S., Ghaffarzadeh, N.: ‘Detection and classification C2 and C3: 440 V VSC-based inverters.
of power quality disturbances using discrete wavelet transform and wavelet
networks’, IET Sci. Meas. Technol., 2010, 4, (4), pp. 193–205 BC1 and BC2: 440 V, dc–dc boost converters.
[35] Mladen, K., Rikalo, I.: ‘Detect and classify faults using neural nets’, IEEE
Comput. Appl. Power, 1996, 9, (4), pp. 42–47

94 IET Signal Process., 2018, Vol. 12 Iss. 1, pp. 82-94

© The Institution of Engineering and Technology 2017