IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy (CISPSSE-2020), July 29-31,

2020, Odisha, India

DSP Based Online Power Quality Events Detection

and Classification Using Hilbert Huang Transform
and Random Forest Method
1st Mrutyunjaya Sahani 2nd Sasmita Choudhury
Student Member, IEEE Student Member, IEEE
Department of Electrical and Electronics Engineering Department of Electrical Engineering
Siksha ‘O’ Anusandhan Deemed to be University Goverment college of Engineering
Bhubaneswar, India Keonjhar, India
[email protected] [email protected]

3rd Susanta Kumar Rout 4th Debadatta Amaresh Gadanayak

Student Member, IEEE Student Member, IEEE
Department of Electrical and Electronics Engineering Department of Electrical and Electronics Engineering
Siksha ‘O’ Anusandhan Deemed to be University Siksha ‘O’ Anusandhan Deemed to be University
Bhubaneswar, India Bhubaneswar, India
[email protected] [email protected]

Abstract—In this paper, empirical mode decomposition (EMD), spike, and DC offset which makes sustenance of power quality
Hilbert transform (HT), and random forest technique are com- a precarious issue. Disturbances in the power system are
bined to detect and classify the power quality events (PQEs) classified basically into stationary and non-stationary where
in real-time. Empirical mode decomposition (EMD) is used to
decompose the non-stationary PQEs into the mono-component the non-stationary disturbances are enormously present in
mode of oscillations, known as intrinsic mode functions (IMF). the power system network. These disturbances lead to the
The most efficacious features are extracted by applying Hilbert disintegration of the execution of transmission and distribution
transform on individual IMF. Four features such as standard system components, so indeed diminutive unsettling influences
deviation of magnitude, Hilbert energy, Tsallis entropy, and in the original signal needs to be checked so as to take
standard deviation of phase are computed from the AM-FM
signal of Hilbert transform to train the random forest classifier. appropriate preventive activity to avoid power quality issues.
Random forest (RF) classifier, a congregated form of numerous Primarily preprocessing, optimal feature extraction and clas-
self-determining decision trees that has a unique ability to deal sification are followed by recognition and detection of power
well with uneven data sets with missing variables and faster system problems. The time and frequency domain information
learning speed compared with other machine learning methods. from PQ disturbances are extracted by utilizing the meth-
The main objective of this paper is to assess the worth of
the proposed HHT-RF method implemented in the hardware ods of Fourier transform (FT) and wavelet transform (WT)
prototype using a digital signal processor (DSP) in the power through the process of preprocessing and feature extraction
quality events detection and classification in real-time. [1-2]. Fourier analysis is deficient of time information during
Index Terms—Empirical mode decomposition (EMD), random transforming to frequency domain so it does not have sufficient
forest (RF), Hilbert transform (HT), intrinsic mode function time information about the event on the other hand also
(IMF), non-stationary power quality events, real-time analysis,
Digital signal processor (DSP). bounded by stationary signals only. Most PQEs issues are
non-stationary in nature so for that there is a requirement
I. I NTRODUCTION of method like a short-time Fourier transform (STFT) and
wavelet transforms that would give frequency components
Online Power quality (PQ) signal disturbance’s recognition included with the timing of event of such disturbances [3-
and classification is the foundation of PQ problem manage- 7]. As non-stationary signals are the integration of stationary
ment in this rapidly growing power system with the multiple signals hence STFT [8] decomposes it into time-frequency
uses of nonlinear loads, fast-acting power electronic devices, domains within the sliding window which provide the timing
computers, lighting controls, and the unavoidable unbalanced of the event of such disturbances. STFT cannot adequately
power system. Quality of power is remarkably weakened analyze transient phenomena whereas, the WT with multi-
due to the event of unsettling influences such as transient, resolution capability proves to be an efficient analyzer of
harmonics, flicker, sag, momentary interruption, swell, notch, nonlinear and transient phenomena by disintegrating the signal
978-1-7281-7274-3/20/$31.00 
c 2020 IEEE into multiple time-frequency levels and, therefore, maintaining

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY HAMIRPUR. Downloaded on August 28,2023 at 12:08:00 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy (CISPSSE-2020), July 29-31, 2020, Odisha, India

the transient characteristics of the analyzed signal and proved

appropriate for investigation of non-stationary signals. But it
is limited by the corrupted performance beneath the noisy
circumstances.
Ultimately this leads to the evolution of another method
called Stockwell’s transform (ST) which is an invertible time-
frequency spectral localization technique by combining the
methods of WT and STFT. Extracted features from Stock-
well’s transform provides a frequency-dependent resolution for
automatic recognition and classification of PQ disturbance by
utilizing classifiers like support vector machine (SVM), fuzzy
logic (FL) and artificial neural network (ANN), etc. [9-13].
Researchers became fascinated to use accurate and efficient
advanced signal processing methods to extract suitable power
quality indices for the recognition and classification of PQ
events precisely from the enormous data to avoid any unde-
sirable situation. Empirical mode decomposition (EMD) has
the unique quality to detect some optimal features of PQ
disturbances [14]. Any complicated data set can be segmen- Fig. 1. Different IMFs of sag with impulsive transient and multiple spike
tized into a finite and often small number of intrinsic mode signal
functions (IMFs) while processing through EMD. It represents
the signals in terms of slow and fast oscillation and the
categorization of signals is local and totally data based. EMD The obtained IMFs using EMD have the same count of zero-
computation does not depend on any base past known value crossing, local minima, and local maxima. The decomposed
of the signals due to which EMD is extremely effective for IMFs are also symmetric with respect to the value of the local
any nonstationary and nonlinear signals like PQ disturbances. mean. The EMD iterative sifting process have a number of
Hilbert Huang transform (HHT) is a time-frequency energy steps as follows.
method to examine the stationary, non-stationary and nonlin- Step-1: Compute the value of local minima and maxima of the
ear signals [15-16]. HHT generates quadrature signals and input PQE signal r(t) and also figure out the value of lower en-
thereby analytical signals for which instantaneous amplitude, velope amin (t) and upper envelope amax (t) using cubic spline
frequency, and phase can be evaluated. In this paper, empirical interpolation technique to determine the value of envelope
mode decomposition (EMD) with Hilbert transform along mean v(t) using the formula v(t) = [amin (t) + amax (t)]/2.
with random forest technique (RF) is proposed to identify, Step-2: Calculate g(t) = r(t) − v(t). if the g(t) have mean
analyze, and classify the signal and concurrent PQEs. After value is zero and the numbers of zero-crossing and extrema are
the decay of unique signals by EMD, components that have same or at most differ by one, then the corresponding signal
similar oscillatory characteristics are categorized to the same of g(t) is called IM F1 , else apply step-1 until g(t) satisfy the
decomposition level known as IMFs. The Hilbert transform above mentioned condition.
(HT) is utilized to extricate the appropriate highlights from the Step-3: The g(t) is treated as new original signal and the
individual IMFs. The highlight vectors are utilized to prepare new residue k(t) = r(t) − g(t). If k(t) is less than the
and test utilizing a recently presented random forest (RF) predefined stopping value of 0.05 and have standard deviation
classifier [17]. (SD) between two consecutive sifting is given by,

 T T

II. H ILBERT H UANG TRANSFORM  2

2
SD(n) = | gn−1 (t) − gn (t) | / | gn−1 (t) | < 0.05
The advanced signal processing algorithm namely empirical t=0 t=0
mode decomposition is the primary algorithm of Hilbert
Huang transform (HHT) to extract the intrinsic mode functions (1)
(IMFs) from the original non-linear and non-stationary power
quality events (PQEs) signals. The Hilbert transform (HT) where n illustrates the nth difference between the original
is applied to each and every IMFs individually to extract signal r(t) and the envelope mean v(t). Otherwise apply the
competent features of the PQEs signals. step-2 to extract the IM F2 , IM F3 , IM F4 unless until an
IM F = 0. The PQE decomposed signal EMD is shown in
A. Empirical Mode Decomposition Fig. 1. The original PQE signal r(t) is regenerated using the
equation given below.
The empirical mode decomposition (EMD) is applied on m
the PQE signal to decomposed into a number of mono- 
r(t) = gi (t) + km (t) (2)
component modes named as intrinsic mode functions (IMFs).
i=0

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY HAMIRPUR. Downloaded on August 28,2023 at 12:08:00 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy (CISPSSE-2020), July 29-31, 2020, Odisha, India

where m represents the IMF number and they are orthogonal The performance of the proposed method is tested using
to each other. fourteen PQEs of synthetic, real, and synthetic with noise
signals presented in Table I. Fig. 2 shows the magnitude
response, phase response, frequency response, and energy
B. Hilbert transform
spectrum of PQE signal using HT for the detection of multiple
The Hilbert transform is applied on the imaginary part of synthetic PQEs.
the extracted real-valued IMF signal into analytic form for
the purpose of feature extraction [14]. The complex-valued
analytic signal representation of IMFs is defined as,

fA (t) = gi (t) + jH{gi (t)} (3)

where H{gi (t)} represents the Hilbert transform of ith ex-
tracted IMFs gi (t) from the original PQE signal r(t).

III. F EATURE E XTRACTION

Moreover, the feature extraction step is used to detect and
classify the power quality disturbances. The HT is applied on
extracted IMFs to obtained the Hilbert transformed array. In
this paper, a number of relevant features are extracted from the
obtained Hilbert transformed array and four useful features
are used to construct the feature vector for the purpose of
detection and classification of PQE signals. The four features
Yk are calculated using ith number of Hilbert transformed Fig. 2. Detection of multiple power quality events.
IMFs Hi (j) with L window length. The k illustrated as the
range of features and j represents the index of the element TABLE I. T HE DIFFERENT PQE S PATTERNS
present in Hi (j). The details of the four instantaneous features Types PQE Labels Types PQE Labels
are interpreted as follows. Transient PL 1 Flicker PL 8
The standard deviation of the instantaneous IMF is considered Sag with Transient PL 2 Swell with Transient PL 9
as the first feature and is defined as, Voltage Swell PL 3 Voltage Sag PL 10

 L
Sag with Harmonics PL 4 Harmonics with Flicker PL 11
 1 
Y1 = 
2 Harmonics with Notch PL 5 Spike with Transient PL 12
| Hi (j) − mean(Hi ) | (4)
L − 1 j=1 Harmonics PL 6 DC Offset PL 13
Voltage Notch PL 7 Momentary Interruption PL 14
The energy of the instantaneous IMF is treated as the second
feature and is enumerated as,
L
 2 IV. R ANDOM F OREST TECHNIQUE
Y2 = | Hi (j) | (5)
j=1 A solitary decision tree is useful for understanding, yet
its forecast control is regularly not sufficiently solid. Along
The above mentioned two statistical features in time-domain these lines outfit technique like random forest are actualized
are used to analyze the non-linear and non-stationary PQE to support the execution of decision trees containing sub
signals efficiently. The third feature is named as Tsallis entropy decision trees with certain diversity. Random forest is a most
to measure the disorder, uncertainty, and irregularities of the developed, supervised, integrated learning classifier utilizing
PQE signals and is defined as, numerous decision trees. It creates solid prescient models,
L
1 − j=1 pj prepares quickly, and gets rid of the issue of over-fitting. Each
Y3 = K (6) tree is developed at an irregular with arbitrariness infused
q−1
in two ways. To begin with, by developing each tree on an
where K is Boltzmann constant, pj is probability status of irregular subset of the preparation information test. Secondly,
L. The last felicitous feature Y4 is extracted from the standard amid the choice procedure of properties with the goal that the
deviation of phase. It gives the unique information of the PQEs splitter at any hub is resolved at random. The RF system is
signal and computed as, effectively proposed for grouping of highlights of various signs

 L which has been disintegrated utilizing EMD and highlights are
 1 
Y4 = 
2 extricated. Expanding a gathering tree has indicated significant
| φi (j) − mean(φi ) | (7)
L − 1 j=1 improvement in characterization exactness.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY HAMIRPUR. Downloaded on August 28,2023 at 12:08:00 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy (CISPSSE-2020), July 29-31, 2020, Odisha, India

V. E XPERIMENTAL V ERIFICATION OF THE P ROPOSED

M ETHOD

In this section, the simulation and experimental verification

of the proposed HHT-RF method as shown in Fig. 3 sre
clearly explained by importing fourteen types of single as
well as combined PQEs patterns. The different PWEs patterns
hardware prototypes are designed and assembled to generate
and record the practical PQE patterns through the high-speed
DSP kit.

Fig. 5. Hardware laboratory setup.

B. Result and Discussion

The advanced time-frequency based HHT is combined with
the unified learning capabilities RF classifier to empower the
classification accuracy for the single as well as combined PQE
patterns. The proposed HHT-RF method performs outstanding
classification accuracy in both the noisy and noise-free envi-
ronment of the highly non-stationary complex PQE patterns.
Fig. 3. Online PQE monitoring system based on the proposed HHT-RF The MATLAB/Simulink software environment is developed
method. to execute the program in a i3 − 3000U @3.3 GHz CPU.
The most efficient features are extracted as shown in Fig.
4 from the robust time-frequency HHT for the quick and
A. Hardware Design and Assembly
accurate localization of the combined PQE patterns. Two
The proposed PQE detection and classification algorithm hundred PQE patterns are recorded from each PQE class,
has also been verified using laboratory scale real-time events. from which one hundred patterns of the individual class are
For this purpose, 9 hardware modules are designed to generate used to train and test the proposed HHT-RF method for
9 different basic PQEs, such as sag, swell, transient, harmon- testing, examining, and validating the overall performances.
ics, notch, flicker, impulsive transient, DC offset, and momen- The obtained superior classification accuracy of the fourteen
tary interruption. High-speed MOSFET switches are used to multi-class PQE patterns is 99.7% for synthetic signal in
control the timing of the events. Sags and swells are created ideal case and the classification accuracy of all the individual
by momentary increase and decrease in loading. Transients PQE class is enumerated in Fig. 3. The overall performance
and momentary transients are generated by capacitor switching of the proposed HHT-RF method is compared with other
at various inception angles. Steady harmonics is created by prevalent methods as shown in Table II to prove its simplicity,
using a three-phase diode rectifier with R-L load. DC offset is feasibility, and practicability for real-time recognition of PQE
created by varying the load resistance with an extra inductor patterns. Most of the existing prevalent methods recognize
added to the source impedance. Similarly, notches, flickers, the single PQE patters effectively but the proposed HHT-RF
and momentary interruptions are created by controlling the method can recognize both the single as well as combined
ON-OFF period of a MOSFET switch connected in series PQE patterns effectively. Higher learning speed, superior clas-
with a load at different load resistance to inductance ratios. sification accuracy, short event detection time, robust anti-
After creating the basic hardware modules, hybrid PQE events noise performances, real-time ability, and practicability are
are generated using more than one basic module at a time. the most distinguished advantages of the proposed HHT-RF
For example, to generate harmonic with flicker both the
harmonic and the flicker modules are employed. An AD202JY TABLE II. T HE PROPOSED HHT-RF METHOD COMPARISON WITH RE -
voltage transducer is used to obtain the PQE data. The TMS CENTLY DEVELOPED METHODS
320C6713, a 32-bit digital signal processing evaluation board
Signal
from the Texas instruments is used for the verification of the Different PQEs Types Classification
decomposition
proposed algorithm. The powerful TMS 320C6713 processor classifier considered of PQEs Accuracy
algorithm
is capable of carrying out the calculations of the detection
WT SVM 9 Practical 94.22 [15]
algorithm upon the inbuild analog to digital converter (ADC)
ST DT 13 Practical 92.69 [16]
outputs continuously as shown in Fig. 5. The recognition
GMOCUW Thresholding 13 Practical 95.79 [17]
capability of the algorithm is checked by interfacing LEDs
HHT RF 7 Practical 97.4 (proposed)
to the general purpose input and output (GPIO) pins such that
HHT RF 14 Synthetic 99.7 (proposed)
appropriate LEDs glow on the detection of the events.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY HAMIRPUR. Downloaded on August 28,2023 at 12:08:00 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy (CISPSSE-2020), July 29-31, 2020, Odisha, India

Confusion Matrix
98 0 0 0 0 0 0 0 0 0 0 0 0 0 100%
1
7.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 100 0 0 0 0 0 0 0 0 0 0 0 0 100%
2
0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 0 100 0 0 0 0 0 0 0 0 0 0 0 100%
3
0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 0 0 100 0 0 0 0 0 0 0 0 0 0 100%
4
0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 0 0 0 100 0 0 0 0 0 0 0 0 0 100%
5
0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 0 0 0 0 100 0 0 0 0 0 0 0 0 100%
6
0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
PQE Output Class

0 0 0 0 0 0 100 0 0 0 0 0 0 0 100%
7
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 0 0 0 0 0 0 98 0 0 0 0 0 0 100%
8
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
0 0 0 0 0 0 0 0 100 0 0 0 0 0 100%
9
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
2 0 0 0 0 0 0 2 0 100 0 0 0 0 96.2%
10
0.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.1% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0% 3.8%
0 0 0 0 0 0 0 0 0 0 100 0 0 0 100%
11
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0% 0.0%
0 0 0 0 0 0 0 0 0 0 0 100 0 0 100%
12
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0% 0.0%
0 0 0 0 0 0 0 0 0 0 0 0 100 0 100%
13
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0% 0.0%
0 0 0 0 0 0 0 0 0 0 0 0 0 100 100%
14
0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 7.1% 0.0%
98.0% 100% 100% 100% 100% 100% 100% 98.0% 100% 100% 100% 100% 100% 100% 99.7%
2.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 2.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.3%
10

11

12

13

14
1

Target Class

Fig. 4. Classification accuracy of the proposed HHT-RF method.

method for the recognition of the single as well as combined ing the most suitable twelve features in both the noise-free
PQE patterns in both the noisy and noise-free environments. and noisy environments. Both the simulation and laboratory
The proposed HHT-RF method can be implemented for the experiment prove the remarkable recognition performance of
superior recognition of real-time PQE patterns to avoid power the proposed HHT-RF method for both the synthetic and
system pollution. practical PQEs. The outstanding classification capability, short
event detection time, simplicity and practicability of HHT-RF
VI. C ONCLUSION method as compared to the existing prevalent method prove
The less computational complex empirical mode decom- that it is the utmost choice to monitoring the single as well as
position is used to extract most suitable IMFs for the com- combined PQE patterns in real-time.
putation of PQIs. Four efficacious PQIs are extracted from
the Hilbert transformed array for better classification accuracy. R EFERENCES
The magnitude response, frequency response, phase response
and Hilbert energy spectrum show the quick, clear and accu- [1] Bracewell, Ronald Newbold, and Ronald N. Bracewell, “The Fourier
rate detection capability for the combined PQE patterns. The transform and its applications,” vol. 31, no. 99, New York: McGraw-
Hill, 1986.
high-speed T M S320C6713 digital signal processor starter
[2] Wang, Zhongde. “Fast algorithms for the discrete Wavelet transform and
kit is more suitable to embed the proposed HHT-RF method for the discrete Fourier transform,” IEEE Transactions on Acoustics,
in MATLAB/Simulink with code composer studio software Speech, and Signal Processing, vol. 32, no. 4, pp. 803–816, 1984.
environments for the quick recognition of the complex PQE [3] Chen, J., Rostami, J., Peter, W.T. and Wan, X.,“The design of a novel
mother Wavelet that is tailor-made for continuous Wavelet transform in
patterns in real-time. The ensemble learning based RF clas- extracting defect-related features from reflected guided wave signals,”
sifier produces outstanding classification accuracy by import- vol. 110, pp. 176–191, 2017.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY HAMIRPUR. Downloaded on August 28,2023 at 12:08:00 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Computational Intelligence for Smart Power System and Sustainable Energy (CISPSSE-2020), July 29-31, 2020, Odisha, India

[4] Alickovic, E., Kevric, J. and Subasi,“Performance evaluation of empiri-

cal mode decomposition, discrete wavelet transform, and wavelet packed
decomposition for automated epileptic seizure detection and prediction,”
Biomedical Signal Processing and Control, vol. 39, pp.94–102, 2018.
[5] Bhattacharyya, A., Sharma, M., Pachori, R.B., Sircar, P. and Acharya,
U.R., 2018.“A novel approach for automated detection of focal EEG
signals using empirical Wavelet transform,” Neural Computing and
Applications, vol. 29, no. 8, pp. 47–57, 2018.
[6] Mohan, Chinnem Rama, and Siddavaram Kiran. “Image enrichment
using single discrete Wavelet transform multi-resolution and frequency
partition,” In Artificial Intelligence and Evolutionary Computations in
Engineering Systems, pp. 87–98, Springer, Singapore, 2018.
[7] García Iglesias, Daniel, Nieves Roqueñi Gutiérrez, Francisco Javier De
Cos, and David Calvo. “Analysis of the high-frequency content in human
QRS complexes by the continuous Wavelet transform: An automatized
analysis for the prediction of sudden cardiac death,” Sensors 18, vol. 1,
no. 2, pp 560–568, 2018.
[8] Wang, Y. and Orchard, J., “On the use of the Stockwell transform for
image compression, In Image Processing: Algorithms and Systems VII,”
International Society for Optics and Photonics, vol. 7245, p. 724504,
February 2009.
[9] Wang, Yanwei, and Jeff Orchard. “The discrete orthonormal Stockwell
transform for image restoration,” 16th IEEE International Conference,
pp. 2761–2764, IEEE, 2009.
[10] Li, T.Y., Zhao, Y., Nan, L., Fen, G. and Gao, H.H.,“A new method for
power quality detection based on HHT. In Zhongguo Dianji Gongcheng
Xuebao,” (Proceedings of the Chinese Society of Electrical Engineer-
ing,) vol. 25, No. 17, pp. 52–56, September, 2005.
[11] Huang, N.E., Wu, M.L., Qu, W., Long, S.R. and Shen, S.S.,“Applications
of Hilbert–Huang transform to non-stationary financial time series
analysis,” Applied Stochastic Models in Business and Industry, vol. 19,
No. 3, pp. 245–268, 2003.
[12] Peng, Z.K., Peter, W.T. and Chu, F.L.,“An improved Hilbert–Huang
transform and its application in vibration signal analysis,” Journal of
Sound and Vibration, vol. 286, No.(1-2), pp. 187–205, 2005.
[13] Svetnik, V., Liaw, A., Tong, C., Culberson, J.C., Sheridan, R.P. and
Feuston, B.P.,“Random forest: a classification and regression tool for
compound classification and QSAR modelling,” Journal of Chemical
Information and Computer Sciences, vol. 4, No. 36, pp. 1947–1958,
2003.
[14] Bergholz, R., Rossel, M., Dutescu, R.M., Vöge, K.P. and Salchow,
D.J.,“Facilitating the analysis of the multifocal electroretinogram using
the free software environment R,” European Journal of Ophthalmology,
pp.0–0, 2017.
[15] Maria Dolores Borrás, Juan Carlos, and Juan Carlos ño, “Disturbance
ratio for optimal multi-event classification in power distribution net-
works,” IEEE Transactions on Industrial Electronics, vol. 63, no. 5, pp.
3117–3124, May 2016.
[16] M. Biswal, and P.K. Dash, “Measurement and classification of simul-
taneous power signal patterns with an S-transform variant and fuzzy
decision tree,” IEEE Transactions on Industrial Informatics, vol. 9, no.
4, pp. 1819–1827, November 2013.
[17] Y. Zhang, T. Y. Ji, M. S. Li, and Q. H. Wu, “Identification of
power disturbances using generalized morphological open-closing and
close-opening undecimated Wavelet,” IEEE Transactions on Industrial
Electronics, vol. 63, no. 4, pp. 2330–2339, April 2016.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY HAMIRPUR. Downloaded on August 28,2023 at 12:08:00 UTC from IEEE Xplore. Restrictions apply.