Chapter-1

Introduction
1.1Background

In an A.C power system voltage and current waveform are expected to be sinusoidal with
constant amplitude and frequency, however due to the increasing uses of non linear loads in
power system mostly power electronics equipment, Uninterruptible power supply,arc furnaces
and controlled motor drives, pericodically distortions in current and voltage waveform
become more. As a matter of fact voltage and current posses a set of sinusodial waveform of
variable amplitude and phase having frequinces which are integer multiple of fundamental
frequency, these frequncy multiplies of this fundamental frequency are called harmonics
frequencies. If suitable filtering is not undertaken then these power electronics equipment will
introduce inter harmonics (Having frequency non integer multipule of fundamental but greater
than fundamental frequency) and Sub harmonics (Having frequency non integer multipule of
fundamental but lower than fundamental frquency) components to the power system. Both
harmonics and Inter harmonics have adverse effects such as increased I 2R losses, over voltage
unbalancing, mal operation of releays, saturation of transformer core etc. It is pertinent that
accurate estimation of harmonics in distroted power system current/voltage signal is essential
to effectively design filter for elimination of harmonics.

Application of Power electronics based devices in the field of industry (Arc and industrial
furnace, Industrial lasers, rolling mills, cement mills, welding, excavators), residence sector
(air conditioning, lighting, space heating, refrigerators, dryers, fan etc), Telecommunication
(Battery chargers, power supplies), Transportation sector (traction control of electric
vehicle’s), Utility systems (High voltage dc transmission) etc is the main causes of poor
quality power now a days. Due to the applications of such devices the V-I Characteristics is
not linear which indicates the production of harmonics in the system, the increased severity of
harmonics pollution in power network day to day attracted the attention of many researchers
to develop different types of efficent device to improve the power quality by reducing the
harmonics level. To quantify the magnitude of harmonics content present in power signal,

1
Total Harmonics Distoration (THD) is used as an index. The aforesaid adverse affect of
harmonics necessitate guidelines to maintation acceptable harmonics level in the power
system in terms of THD. To avoid these undesirable effect initially Line reactors, isolation
transformers, Low pass harmonic filters were used but they are ineffective due to their
inability to adapt to network characteristic variation and passive filter suffer from large size,
fixed and effect of resonance etc. Recently some active Power Filters (APFs) have been
widely investigated for compensation of harmonics in electric power system. Shunt active
power filter, Series active power filter and hybrid filters (Combination of active and passive)
but the performance of such active filters depend on the contorl mechanisim and methods
applied for generation of switching signal for the active power filter.

Again in Power System, Frequency is one of the most important and sensitive parameter in
power system. Any variation in power system is eventually reflected the change in frequency.
A change in frequency leads to change in system reactance and the operation of several relays
such as reactance relay is affected. Frequency is a measure of mismatch between power
generation and load demand. If load demand is greater than power generation under frequency
situation arises and if generation is greater than load demand over frequency situation arises,
in either case change in frequency poses a threat to efficiency, safety of entire system and
increase in chances of system collapse, thus frequency is an integral part of power system
protection, power quality monitoring, and operation and control of devices using digital
technology. Hence the accurate estimation and tracking of system frequency is of utmost
important. Due to development of several electronics and other non linear devices, the present
power system is subjected to several undesirable conditions such as presence of noise and
harmonics etc. Keeping in mind several undesirable conditions and also taking into account
the dynamic phenomenon of frequency variation conventional frequency estimation based on
constant frequency assumption are not suitable for achieving accurate frequency estimation.

This chapter presents a progressive review on power system Frequency estimation, harmonics
estimation and harmonics elimination. Power System Frequency and Harmonics estimation
methods based on signal processing approach such as Recursive Least square (RLS),
Extended Least Square (ELS), Kalman filtering(KF), Least Mean square (LMS) etc to soft
computing approach such as Fuzzy logic, neural network ,Genetic algorithm and Evolutionary

2
Computation etc are discussed. Harmonic elimination methods starting from passive filtering,
active filtering to hybrid filtering along with different control techniques such as synchronous
detection technique, indirect current control technique, Hysteresis current control, Fuzzy logic
controller are also reviewed.

1.2 A Review on Power System Frequency Estimation using Signal Processing

Techniques

The rapid growth of Signal processing Techniques and soft computing techniques make
modern frequency measurement flexible and a variety of techniques have been developed in
the recent years for the same purpose. Some of the best techniques for frequency estimation
are described in this chapter.

Frequency or the period of a signal can be measured by Zero crossing detection [1], when
measuring the frequency of a signal, cycles of a reference signal is measured over one or more
time periods, however Multiple periods of measurement helps to reduce errors caused by
phase noise. Accurate measurement can be achieved by this method at the expense of slow
measurement rates. Zero crossing detection method or its modification using curve fitting of
voltage samples is the simplest approach for frequency estimation [2] but the estimation
accuracy gradually decreases when high contents of harmonics are associated with the signal,

Duric, M.B et al; [3] proposed a new approach to the design of a digital algorithm for network
frequency estimation. Derivation of the proposed algorithm is based on Fourier and zero
crossing technique. Fourier method is used as digital filter and zero crossing technique is
applied to the cosine and sine components of the original signal which can be corrupted by
higher harmonics. Discrete Fourier Transform (DFT) is used by many researchers for
frequency estimation of a signal, The conventional DFT shows excellent performance when
the signals contains fundamental and integer harmonics component [4]. presence of decaying
dc component in a signal and the implicit data window in DFT, introduce fairly large errors in
the estimation when frequency deviates from the nominal value.[5] To improve the
performance of DFT, some approaches has been mentioned by some authors [6-8] like
feedback loop by turning the sampling interval, adjusting the data window length, changing
the nominal frequency used in DFT iteratively, correcting the gains of orthogonal filter and

3
tuning the weighted factor recursively respectively. Theoretically the decaying component
can be completely removed from the original waveform once its parameter can be obtained.
Some of the authors added some samples to calculate the parameter of the decaying
component based on this idea. The effect of DC components by DFT is eliminated by using
the outputs of even-sample set and odd-sample set. Some authors estimate the parameter of
the DC of decaying component by using the phase-angle difference between voltage and
current.

Fast Fourier Transform (FFT) is one of the conventional methods for frequency estimation
and it is based on a Fourier series model of the data and the data are composition of harmonic
signals. This analysis is computationally efficient and produces reasonable results for a large
class of signal process [9]. Though this method possess this type of advantage it has
associated with some disadvantage like frequency resolution,i.e the ability to distinguish the
spectral responses of two or more signals and irregular windowing of data that occurs at the
time of processing with FFT. Windowing manifests itself as leakage in the spectral domain-
energy in the main lobe of spectral responses that are present [10-11].

Sidhu et al; [12] proposed a revised digital algorithm called Smart Discrete Fourier
Transforms to estimate the frequency of a sinusoidal signal with harmonics in real time. This
algorithm smartly avoids the errors which are coming due to the deviation of frequency from
nominal frequency and always associated with all the advantages of DFT [4], although this
approach is suitable for measurement of frequency over a wide-range, the on-line application
requires a trade-off between the accuracy and computational complexity.

T.Lobes et al; [13] Proposed Prony estimation technique along with Discrete Fourier
Transform (DFT) which is a static state algorithm for power system frequency estimation with
a variable data window to eliminate the noise and harmonics associated with a signal. Least
square algorithm along with orthogonal FIR digital filter presented by the same author [13]
for measurement of frequency in the operating condition of a power system.This algorithm is
capable of producing a correct and noise free estimate for near nominal, nominal and off
nominal in very short duration.

4
Jin Kwon Hwang et al; [14] presented a Novel-DFT based frequency estimation technique by
introducing three digital filter to reduce the frequency error developed by noise and leakage
effect of the negative fundamental frequency in a single phase signal analyzed by DFT based
algorithm[4]. Tomas Radil et al; [15] proposed one algorithm based on leakage compensation
by best fitting a theoretical spectrum of a rectangular windowed single-tone signal on the
spectrum of the analyzed signal for the accurate estimation of the signal’s frequency. The
proposed algorithm is several times faster than the multi harmonic sine fitting algorithm. The
proposed algorithm is suitable for monitoring frequency in power systems. It can also be
applied in other areas, particularly when the accurate estimates of the signal’s amplitude and
phase are required.

Arghya Sarkar, [16] proposed a novel digital signal processing algorithm for online estimation
of the fundamental frequency of the distorted power system signals. The basic algorithm relies
on the development of an efficient variance reduction algorithm and design of a new stable
band pass infinite impulse response (IIR), second-degree digital integrator (SDDI) with
reduced approximation error. Compared with the well-established technique such as the
enhanced-phase-locked-loop (EPLL) system, the proposed algorithm provides higher degree
of immunity and insensitivity to harmonics and noise and faster response during step
frequency change.

Karimi.H et al; [17] presented A method for estimation of power frequency and its rate of
change, the proposed scheme accommodates the inherent nonlinearity of the frequency
estimation problem by providing a fast and accurate estimation of the frequency when its
deviation from the nominal value is incremental or large. The estimator is based on a newly
developed quadrature phase-locked loop concept.

The dominant frequency component of a input signal and its frequency estimation is proposed
by P.J moore [18] on introducing a phase-locked loop (PLL) system. The mechanism of the
proposed PLL [18] is based on estimating in-phase and quadrature-phase amplitudes of the
desired signal. Mohsen Mojiri et al; [19] proposed an Adaptive Notch Filter (ANF) which can
be employed and furnished with pre- and post-filters to devise a method for estimation of
power system frequency and its rate of change They discussed the adjustment of the ANF
parameters and the design of pre- and post-filtering stages. The dynamic responses of the

5
proposed method with regard to step, ramp, and oscillatory changes of frequency are faster
than those of the PLL-based method. The structural simplicity of the proposed estimator
renders it suitable for digital implementation both in hardware and software environments.

Wu Jiekanget al; [20] proposed a high-accuracy, wide-range frequency estimation method

based on the principle of numerical differentiation. The fundamental frequency of non-
sinusoidal signal voltage and current of a power system can be estimated with a structure
similar to the basic parameter estimator, the estimator is simple in design and implementation
and it is very effective for tracking the real-time frequency of the power system.

YiliXial et al; [21] proposed a novel technique for online estimation of the fundamental
frequency of unbalanced three-phase power systems based on Clarke’s transformation and
widely linear complex domain modeling, the proposed method makes use of the full second-
order information within three-phase signals, thus promising enhanced and robust frequency
estimation. The proposed method is also less sensitive to the variations of the three-phase
voltage amplitudes over time and in the presence of higher order harmonics.

Least Mean Square (LMS) [22] algorithm is adopted where the formulated structure looks
very simple and it has been observed that this algorithm is found to be accurate under various
systems changing condition to estimate correct measure of frequency. Pradhan et al; [23]
proposed a Least Mean Square algorithm in complex form to estimate the frequency of a
power system. This estimation of frequency is verified in the presence of noise, with
frequency jump and data collected from real time system. The presence of 3 rd harmonic in the
signal does not affect the performance of the algorithm as the 3rd harmonic component is
eliminated during Clarks transform. But the presence of 5th harmonic component affects the
performance of the algorithm, so a Butterworth Filter used for pre filtering shows the
correctness of the estimation with less error.

A Variable Step Size LMS (VSSLMS) has been proposed [24] to get more accurate and better
convergence in estimation over conventional LMS algorithm. Disturbances exist in a signal do
not affect the estimation performance using VSSLMS algorithm. Step size of this algorithm is
adjusted by autocorrelation of square of time averaging estimate error and previous error. The
auto-correlation error is a good measure of the proximity to the optimum and it rejects the

6
effect of uncorrelated noise sequence in the step size update. However, this VSSLMS provides
faster convergence at early stages of adaptation while there is little deviation in the later stage.

The conventional LMS technique based on adaptive linear filtering possesses the advantage of
simplicity in its underlying structure. However, it suffers from the problem of poor
convergence rate if the step size for adaptation is fixed [25] This step size, in general depends
inversely on input power, i.e, it takes more time to learn about its input when step size is
small and vice versa. Time-varying step size is usually employed to overcome this poor
convergence problem. If the LMS of the error is only considered as the cost function to be
minimized, with respect to the dynamic variation the linear weights of the filter may go
unbounded or take longer time to respond because of the stalling effect [26]. In order to avoid
the drifting of weight involved in the estimation mechanism, B.Subudhi et.al proposed a
variable leak adjustment technique in which a variable adaptation step size is incorporated to
attain faster convergence. To enhance the convergence characteristics and to reduce the error
of the LMS algorithm in power system frequency estimation Ray proposed Extended Least
square [27]

Soliman Abdel-Hady [28] proposed a new application for linear Kalman Filter algorithm for
power system frequency estimation. The filter uses the digitized samples of the three-phase
voltages or current waveform signals. These three phases are transformed into two phases,
using the well-known αβ –transformation matrix. Having obtained the signal of the two new
phases, a complex phasor is constructed using the new two-phase voltages. Kalman filter is
then applied to extract the frequency and phase angle of the fundamental component of the
complex phasor.

An approach based on Recursive Least Square (RLS) Algorithm applied to frequency

estimation of the instantaneous power system [29] the Three-phase voltage signal is
transformed to a complex form which is easy to be handled by the proposed approach. The
RLS Algorithm is more suitable for online frequency estimation due to its rapid convergence
rate. This algorithm recursively finds the coefficients that minimize a weighted linear least
squares cost function related to the input signals. When compared with other algorithms, the
RLS algorithm exhibits the feature of rapid convergence rate. However, this benefit comes at
the cost of high computational complexity.

7
Adaptive LMS algorithm-based adaptive filters are used to estimate the discrete Fourier
coefficients of sine and cosine terms of noisy sinusoidal signals, whose frequencies are known
a priori. The standard RLS technique is used by many researchers but it is computationally
complex. Using approximations, a new recursive Gauss–Newton adaptive filter is proposed by
Das [30] to estimate fundamental and harmonic phasors of power system voltages or currents
and their variations because of sudden disturbances take place in a power system. Further, the
filter possesses computational simplicity of the normal gradient technique and has the speed
of convergence of the Newton method.

A.Pradhan et al, [23] Presents an arc cosine function –free technique for frequency estimation
to reduce the burden of computation with little decline in frequency estimation accuracy. A
leak factor updatation algorithm has been proposed for variable leakage factor in VLLMS
[24]. This leak adaptation in the proposed VLLMS has the advantage of using measurable
signals in the system to perform the adjustment of the leak factor.

Recently Soft Computing (SC) techniques are receiving more attention as optimization
techniques for many industrial applications. It is an evolving collection of methodologies,
which aims to exploit tolerance for imprecision, uncertainty and partial truth to achieve
robustness, tractability, and low cost. SC provides an attractive opportunity to represent the
ambiguity in human thinking with real life uncertainty. Fuzzy logic (FL), Neural Networks
(NN) and Evolutionary Computation (EC) are the core methodologies of soft computing; it
can solve problems that have not been able to be solved by traditional analytic methods. In
addition, SC yields rich knowledge representation (symbol and pattern), flexible knowledge
acquisition (by machine learning from data and by interviewing experts) and flexible
knowledge processing (inference by interfacing between symbolic and pattern knowledge),
which enable intelligent systems to be constructed at low cost and high machine intelligence
quotient (HMIQ).

This section describes some of the recent developed soft computing methods applied for
frequency estimation in power system signal.

Neural Network and Genetic Algorithm (GA) have been used in [31], for estimation of power
system frequency. In this proposed algorithm, the learning of weights of NN was carried out

8
by GA. Authors have compared the performance of this proposed technique with the
conventional error back propagation and LMS algorithm. But they found that the proposed
algorithm outperforms over the other two. They have judged the performance using
simulation only and also observed that though the algorithm gives better performance still it
suffers from problem in training of the network.

M. Gupta et al; [32], proposed a faster training algorithm for estimation purposes. The author
first applied only Gradient Descent (GD) algorithm separately for the estimation of frequency
and observed that it has the disadvantages of getting stuck in local minima. Then applied PSO
separately and observed that the square of the error fluctuates randomly and it may take much
iteration to converge. To avail the advantages of both the technique, a hybrid algorithm has
been proposed by the same author in the same work to estimate the power quality parameter
estimation. In those new hybrid algorithm chances of getting stuck in the valley of local
minima becomes almost nil. Simulated results prove the superiority of the proposed hybrid
algorithm (combination of GD& PSO) in terms of lesser number of iterations to converge.

A. Sundarrajan [33], proposed Evolutionary Algorithms (EA) like, Enhanced Particle Swarm
Optimization (EPSO), Multi Objective Particle Swarm Optimization (MOPSO) and Stochastic
Particle Swarm Optimization (SPSO) to overcome the premature convergence problem in a
standard PSO. These algorithms reduce transient oscillations and also increase the
computational efficiency for frequency estimation,

P. K. Ray et al; [34], proposed RLS-Adaline and KF-Adaline algorithms for power system
frequency estimation approaches, the weights of the Adaline are updated using RLS/KF
algorithms. Frequency of power system signal is estimated from final updated weights of the
Adaline. Neural estimator was found to be an effective estimator [35]. It consists of an
adaptive perception of neuron called Adaline. Since KF and RLS both are recursive in nature,
online estimation is possible and KF can be used for both filtering and estimation, KF and
RLS algorithms have been employed in the proposed hybrid algorithm for updating the weight
in Adaline. Both RLS-Adaline and KF-Adaline estimators’ track the power system signal in
different cases such as signal corrupted with noise, in presence of harmonics and in presence
of sub harmonics and inter harmonics. Evolutionary Computation technique [36] is a
population based search algorithm; it works with a population of strings that represent

9
different potential solutions. It enhances its search capability and the optima can be located
more quickly when applied to complex optimization problems. An EC technique called
Bacterial Foraging Optimization (BFO) is developed and it is combined with RLS and
Adaline [38] to power system frequency estimation. BFO is one of the recent bio-inspired
computing used by many researchers in different areas of optimization.

Four new hybrid algorithms such as RLS-Adaline, KF-Adaline, RLS-BFO and Adaline-BFO
[39] are presented. The performances of the first two proposed hybrid algorithms i.e RLS-
Adaline and KF-Adaline are dependent on the initial choice of weight vector W and
Covariance matrix P. By using an optimal choice of weight vector, faster convergence to the
true value of signal parameter can be achieved. After the optimization of the weight vector,
online tracking of frequency of signal can be carried out. Both the algorithms track the
frequency of signal at different level of noises and different signal changing conditions but the
performance of tracking using KF-Adaline is better than RLS-Adaline.

Fuzzy linear regression is proposed in [40] for frequency and harmonics evaluation in a power
network, which used digitized voltage signals as fuzzy numbers for estimation of frequency
and harmonics components of voltage signal. Sampling frequency effects, data window size
and degree of fuzziness on the estimated parameters has been investigated and presented.

The frequency and the rate of frequency change are estimated by the non recursive Newton-
type algorithm [40, 43] using generator swing equation, the recursive algorithm form is
improved with a strategy of sequential tuning of the forgetting factor. By this, the algorithm
convergence and accuracy are significantly improved.

1.3 A Review on Power System Harmonics Estimation

Harmonics are a mathematical way of describing distortion in voltage or current waveform.

The term harmonic refers to a component of a waveform occurs at an integer multiple of the
fundamental frequency. Estimation of the harmonic components in a power system is a
standard procedure of the assessment of quality delivered power. Several methods such as
Discrete Fourier Transforms, Least Square Error technique, Kalman filtering, Adaptive Notch
Filters etc; has been used for harmonic estimation of distorted signals in power system. Some
of the methods used for harmonics estimation purposes are described here.

10
Fast Fourier transform (FFT) [52-54] is an efficient method for harmonics estimation and it
produces reasonable results for a large class of signal processes. it has some advantages but
still there are several performance limitations of this approach. The most distinctive limitation
is that of frequency resolution, which incorporates spectral responses of two or more signals.
It is very crucial to develop better tools to avoid possible technical hazards which may come
up due to their existence. Windowing of the data which is the leakage in spectral domain
manifest itself as second limitation. As many measured process are being brief in duration and
having short time –varying spectra, It is troublesome to analyze short data records. To
alleviate the limitations of the FFT approach, many new spectral estimation methods have
been proposed during the last few decades [55-57]. Advantages of the new methods depend
strongly upon the Signal-to-Noise Ratio (SNR).

In Kalman Filter (KF) [59], individual harmonics injection source is treated as a random state
variable. Error covariance analysis of harmonic injection source is treated as a random state
variable, was used to determine the optimal metering locations in power systems. Based on
this optimal arrangement, the KF was able to estimate and track each harmonic injection in
power system [60] KF estimates amplitude and phase when frequency is fixed. However,
when frequencies vary, it is unable to retune itself to the frequency changes. Similarly, Fourier
Linear Combiner using single layer neutral network able to estimate harmonics at static
Frequency but during frequency change tracking time becomes much larger and there is more
error in estimation. Das P.K. et al; [61] presents an approach combining both Fourier Linear
Combiner and Extended Complex Kaman Filter (ECKF) which will overcome the above
mentioned problem.

Rosendomacias J.A. et al; [62] Presented the basic theory of the STDFTand kalman filter
approach as well as the kalman filter models available for harmonics estimation, STDFT
shows quite a uniform convergence within a period to the proper value of the harmonic,
despite the presence of higher harmonics. However it can not deal properly with the presence
of decaying DC Component, which produces ripple around the correct harmonic magnitude.

MaamarBettayed et al; [63] proposed Several variants of Recursive Least Square (LS)
algorithm such as Weighted Least Square (WLS), Recursive Least Square information form
(RLS-I), Recursive Least Square Covariance form (RLS-C), Extended Least Square form 1

11
(ELS-1) and Extended Least Square form 2 (ELS-2) for estimation of power system
Harmonics using a noisy harmonic signal from an AC bus of a six-pulse Rectifier as a test
signal and applied various RLS algorithms to signals having different SNR values.

An online estimation technique of harmonic signal based on LS and Total Least Square (TLS)
optimization criteria has been presented [71]. The SVD technique [64] is very interesting for
signal reconstruction and estimation. The application of this methodology in an offline
estimation of higher harmonics in Power transmission system, Linear Least Square method for
harmonic detection in a Power system using Singular Value Decomposition (SVD) is
proposed by [65] The size and location of large harmonic sources which are smaller in size in
comparison to suspious buses in practical power system are unknown before state estimation.

Lio et al;[66], proposed a systematic approach to identify and estimate harmonic sources in
power networks, when the number of harmonic meters is less than the number of unknown
state variables. Dominguez et al. [124] presented the introduction of digital filter for
estimation of Harmonics components of a signal. The digital filter CDF (Characteristics
Harmonics Digital Filter) incorporated in treatment of samples before applying Discrete
Fourier Transform (DFT). The convergence of the algorithm during signal change is also
improved.

Yilmatz et al; [67] suggested parametric spectral estimation methods for the estimation of
harmonics, inter-harmonics and sub-harmonics, Co-Variance and Modified Co-Variance
methods are applied for estimation of harmonics. A Mother Wavelet Transform based
approach for the study of time-varying power system harmonics has been developed where
time frequency localization chacteristics are embedded in Wavelets. Mandel [68] suggested
Ensemble Kaman Filter (EnKF), is a new version of the Kalman Filter and is an important
data assimilation component of ensemble forecasting. This ref. described the derivation and
practical implementation of the basic Version of EnKF [68] this paper suggested that EnKF
can be implemented without access to the observation matrix but only an observation function
is required.

Soft computing is a concept that has come into prominence in recent times and its application
to power system analysis is still more recent. This section explores the application of soft

12
computing techniques in the area of power system harmonics estimation, soft computing as
opposed to conventional “hard” computing, is a technique that is tolerant of imprecision,
uncertainty, partial truth and approximation. Some of the important branches of soft
computing (SC) are Artificial Neural Networks (ANNs), Fuzzy Logic (FL), Genetic
Algorithm (GA), and BFO etc.
Das P.K. et [125] proposed a technique on Fuzzy LMS for estimation of harmonics voltage
and current signals in power network using fuzzy gain scheduling method for the adjustment
of step size to provide faster convergence and noise rejection for tracking fundamental as well
as harmonics components from signals.

Joorabian et.al, [69] described decomposition of Total Harmonics estimation problem into a
linear and non-linear problem, Linear Estimator (Least Squre (LS)) has been used for
amplitude estimation and an adaptive linear combiner “Adaline” which is very fast and
simple is used for harmonics phase estimation. Improvement in convergence and processing
time is archived in this algorithm. This algorithm estimates correctly for static, dynamic and
fault signal but estimation is difficult for inter and sub harmonic components.

If there is any frequency drift in signal, then conventional FFT based on fixed measurement
window is unable for effective power system monitoring. However, Lie et al. [70] applied the
Least Square technique with ANN to harmonics extraction in time varying situation. This
proposed method is capable of dealing simultaneously the measurement of varying frequency,
amplitude and any harmonic components present in the power system.

Mori et al. [81] presented a method based on back propagation learning for feed-forward
neural network for harmonics prediction. S. Ghodratollaht et al. [72] presented an adaptive
neural network based on Genetic Method called GAP (Genetic AdalinePerceptrons) for
tracking the harmonics components of current and voltage waveforms in faulted power
system, at each iteration of Adaline, GAP uses GA for selection of optimized value for
learning parameter.

Bettayeb and Qidwai [73] presented a new algorithm for estimation of harmonics using GA’s.
The proposed algorithm estimates phase of power system signal using GA. After the
estimation of phase, amplitude has been estimated using Least Square (LS) algorithm. Signals

13
taken across load from a two-bus three-phase system with a full-wave six-pulse bridge
rectifier were applied to this algorithm.

Mishra [74] presented to estimate the harmonic components present in power system
voltage/current waveforms. The foraging strategy becomes adaptive by using Takagi-Sugeno
scheme. Linear Least Square has been combined with B.F.O and Takagi-Sugeno scheme used
for estimation of amplitude.

De Arruda, et al; [75] Presented a new methodology to estimate harmonics distortion in power
system, based on measurements of a limited number of given sites. The algorithm utilizes
Evolutionary Strategy (ES), a developed branch of evolutionary algorithm, the main
advantage in using such a technique relies upon its modeling facilities as well as its potential
to solve fairly complex problem, Ray Pravat et al; [76] presented the Artificial Neural
Networks and Evolutionary Computation techniques, which, focuses on exploiting estimation
of harmonics in power system. This in turn will be useful for designing active filters to ensure
power quality (pure sinusoidal voltage or current signal in a power system). It is also
demonstrated through simulation that how Neural Network and Evolutionary Computing
(Bacterial Foraging Optimization (BFO)) are combined to achieve accurate estimation of
different harmonics components of a distorted power system signal. Adaline-BFO algorithm
and the BFO algorithm have been applied for estimation of harmonics of the voltage obtained
across the inverter terminals of a prototype Photo-Voltaic (PV) system. From the obtained
results, it is confirmed that the proposed Adaline-BFO algorithm provides superior estimation
performance when compared with the Discrete Fourier Transform (DFT), Kalman Filter (KF)
and BFO algorithms.

1.4 Reviews on Harmonics Elimination

The effect of Harmonics due to linear and non-linear load in power system was not a major
issue in the last 40 years. However, in the recent year the number of harmonic producing
devices is increasing rapidly from low-voltage to high voltage application. These nonlinear
loads increases the generating capacity of harmonics, The increased severity of harmonics
pollution in power network has attracted the attention of many researchers to develop different
types of device to mitigate the harmonics and enhance the power quality. A progressive

14
development of various devices and control algorithm used for Harmonics elimination in
power system to enhance the power quality is discussed in this section.

Current harmonics are produced by [97, 120] nonlinear load such fluorescent lighting with
electronic ballasts, switch mode power supply (SMPS), battery chargers, rectifiers, inverters,
three phase power converter fed drives, arc furnaces, arc welding, discharge lighting and
saturable reactors etc. in the power system leading to current and voltage waveform distortion.
All these loads draw the non-sinusoidal currents resulting in current harmonics and are
injected back into the supply system through the Point of Common Coupling (PCC), which
causes severe deterioration of power factor and other adverse effect such as overheating of
transformer, increase in R.M.S value of supply current, losses, over loading in the system,
unnecessary tripping and interference with telecommunication lines and poor power system
efficiency etc. [98, 122]

To avoid these undesirable effect initially Line reactors, isolation transformers, K-Factor
transformers, tuned harmonic filters (fixed capacity or automatic switched multiple banks),
IGBT based fast switched harmonic filters, Low pass harmonic filters were used but they are
ineffective due to their inability to adapt to network characteristic variation. Gonzalez, D. A.,
et.al [126] proposed shunt passive filters for harmonic mitigation in power system. Shunt
passive filters are configured with inductance, capacitance and resistance elements and tuned
to control harmonics. Shunt passive filters are advantageous over series compensators as they
have compensate harmonics as well as reactive power, in addition they have not carried large
currents hence associated losses are less. Ludbrook, A. et al; [127], proposed high pass filter
for notch reduction in a power distribution system having large electronic loads. Therefore,
recent progress in switching devices has resulted in the formulation of several active filter
topologies such as shunt, series and hybrid active filters [99-100]. based on their objectives,
system configuration, power circuits and control strategy, these filters not only used for
current or voltage filtering but also used for voltage dips, damping, isolation, reactive power
control for power factor correction and voltage regulation etc.

A comprehensive review of active filter (AF) configurations, control strategies, selection of

components, other related economic and technical considerations and their selection for
specific applications are presented. The most important configuration widely used in active

15
filtering applications for current harmonic reduction and power factor improvement is the
Shunt Active Power Filter. A SAPF consists of a controllable voltage or current source
inverter. The voltage source inverter (VSI) based shunt APF is the most commonly used type,
due to its well known topology and straight forward installation procedure. SAPF acts as
harmonic current source which injects an anti-phase but equal magnitude of the harmonic and
reactive current as that of nonlinear load. As a result components of harmonic currents
contained in the load current are cancelled and the source current remains sinusoidal and in
phase with the respective phase to neutral voltage, Series APFs are less common than the
shunt APF. This is because of the fact that they have to handle high load currents which will
increase their current rating considerably compared SAPF especially on the secondary side of
the interfacing transformer. This increases the I2R losses. However, the main advantage of
series APF over shunt APF is that they are ideal for voltage harmonics elimination [101]. It
provides the load with a pure sinusoidal waveform, which is important for voltage sensitive
devices such as power system protection devices. With this feature, series APF is suitable for
improving the quality of the distribution source voltage

The utilization of fast switching devices in APF application causes switching frequency noise
to appear in the compensated source current and interfere with neighboring sensitive
equipment. To overcome the limitations, hybrid active power filter are designed. Active power
filter keeps the power system balance under the condition of unbalanced and nonlinear loads
[98, 102] The Performance of the SAPF mainly depends upon the design of power inverter,
types of current controller and methods to obtain the reference current.

The Controller is the most important part of the active power filter and recently a lot of
research is being conducted in this area [99].Mostly a Proportional- Integral (PI) controller is
used to control the DC link capacitor voltage as well as to estimate peak value of reference
source current for both single phase and three phases APF [103-105]. The output of PI
controller is multiplied with unit vector of source voltage to generate reference source current.
Unit vector implies a sine wave having unit peak value with phase same as that of source
voltage. As PI controller has large impact on source current harmonics, a low pass filter is
connected at the output of PI controller to reduce the source current THD. Some advance
technologies, such as Fuzzy Logic, Artificial Neural Network (ANN) and Genetic Algorithm

16
have been used in literature [106-108] to generate reference source current. P. Kumar and A.
Mahajan [106] compares different soft computing techniques for generating reference source
current. They found that with application of soft computing technique, APF give very good
response under frequent load variation. In [107] comparison of PI controller and Fuzzy Logic
Controller is done for controlling the DC voltage of capacitor. The generation of reference
current using the combination of ANN and Fuzzy Logic is explained in [128]. In this literature
all the analysis have been carried out in discrete time domain, the main benefit of this
controller is that it can handle nonlinearity. Harmonics are estimated using neural network
[108] and real power loss by circuit elements of APF is estimated using PI controller. Both PI
controller and NN are used to generate reference source current. A three phase shunt active
power filter was proposed by H. Akagi using instantaneous active and reactive power theory
[109]. In this control strategy, reference source currents are calculated instantaneously using
instantaneous source voltages and load currents. Further development in this strategy was
done by S. Bhattacharya, who calculated d-q (direct-quadratic) components of instantaneous
three phase currents [129]. This literature gives concept about synchronous reference frame
and calculated reference source current instantaneously without sensing the source voltage. A
modified reference current extraction method [130] is proposed using both p-q(active –
reactive) and d-q theory . In [130], reference source current is calculated using real power
balance of the system. The peak value source current required to balance the real power loss
of the circuit elements is calculated. The peak value of source current required to provide real
power to nonlinear load is calculated. Finally both the peak values are added to give the peak
value of reference source current.

Modulation scheme plays an important role in reducing the source current THD. Mostly two
types of switching modulation schemes are applied to active power filter, Hysteresis
modulation (HM) and Pulse Width Modulation (PWM). But for a single phase APF both HM
and PWM switching schemes are classified into two types, unipolar modulation and bipolar
modulation. Application of unipolar modulation scheme is discussed in [118-119] and in [118-
119], unipolar hysteresis modulation based switching scheme applications technique analyzed
and bipolar hysteresis modulation scheme [110] presented. A brief description of unipolar and
bipolar modulation schemes is given below. As shown in Fig. 1.1 for positive source current,
unipolar modulation employs +V and 0 , and for negative source current it employs -V and 0 .

17
But bipolar modulation scheme only employs +V and -V both for positive and negative
current.

Fig. 1.1 (a) source current, (b) Unipolar Modulation, (c) Bipolar
Modulation
In unipolar Modulation, even harmonics are found to be absent So THD of source current in
bipolar modulation is twice of unipolar modulation scheme. In PWM switching scheme,
switching frequency remains constant, where as in HM switching scheme, switching
frequency may varies as the load changes. To control the switching frequency, an adaptive
hysteresis band based current controller for APF is presented [111]. In this control strategy
hysteresis band width changes with change of load but it requires complicated mathematical
calculations. B. Mazari proposed a method of updating hysteresis band using fuzzy logic to
avoid mathematical calculation [116]. All the above discussed hysteresis band controllers are
based on two level hysteresis band. A three level hysteresis band based current controller is
presented. THD of source current reduced significantly in three level hysteresis band [131]
based controller of APF. Although most of the research on APF is based on reference current
extraction method and switching schemes, current control of APF also plays a significant part
in analyzing stability of the complete system, robustness under external disturbances and
reducing THD of the source current. Current of APF can be controlled directly or indirectly. In
direct current control method, the sensed coupling inductor current used directly in the

18
controller of APF, in indirect current control method sensed source current is used in the
controller to generate switching pulses. In literature [103-104], [112], indirect current
controller is implemented, whereas literature [155], [167] are based on direct current
controller. Mostly indirect current control techniques are applied as it is easier to implement. A
resonant current controller is presented in [112]. A Lyapunov stability based current control
strategy is presented in [105]. The concept of equilibrium points and linearization of single
phase shunt APF is explained in literature [132]. Both model reference adaptive controller
[132], Lyapunov stability controller [105] are very good method for analyzing stability of the
system. In [132], transient response of the system is significantly improved, which is better
than the PI controller based APF. The sliding mode (SM) control [178] is applied to three
phase shunt APF. Also in both literature [103] and [104] SM control strategy is applied to
single phase shunt APF. SM controller via feedback linearization [110] is applied to shunt
APF . Coming into the SM control strategy and feedback linearization control strategy, it is
found that there is good development of these control strategies applied to other field of power
electronics.

To avoid the drawbacks of variable switching frequency, a PWM based constant

switching frequency SM controller of DC-DC converter [133] is presented. Similarly fixed
switched frequency SM [134] controller for single phase VSI is developed. To make the boost
converter robust under variation of input voltage and load, an adaptive SM controller [135] is
presented. A variable sliding surface based position control of DC motor [136] is presented.
Maximum power point tracking method of photovoltaic system using SM controller [137] is
analyzed. In this literature instead of sensing voltage of the capacitor connected across the
photovoltaic system, current flowing through the capacitor is used in the SM controller. In
Dynamic stability of the photovoltaic system [138] connected to grid is analyzed using zero
dynamic of the system. This process is nothing but the feedback linearization of grid
connected photovoltaic system [139]. A robust Partial Feedback Linearization (PFL) scheme
of the photovoltaic system for maximum power point tracking is presented. This control
algorithm is mainly robust under parameter uncertainties. Also an Exact Feedback
Linearization (EFL) scheme is presented. The complete description of SM control strategy is
explained. Similarly a detail explanation of feedback linearization method is given. Voltage

19
regulation and robustness toward any type of load under distorted source conditions is
discussed.

Salem Rahmania et al;[99] made a comparison of the performance of the single-phase shunt
active power filter (SPSAPF) and the single-phase shunt hybrid power filter (SPSHPF) that
adopt both an indirect current control scheme with a unipolar pulse width modulation
(UPWM) strategy. The SPSHPF topology includes, in addition to the components of the
SPSAPF, a power factor correction capacitor connected in series with a transformer. Smitha
Krishnan et al; [140] proposed a new shunt active power filter composed of a three-phase
converter and a Scott transformer. The controller part includes fuzzy controller and FBD
detection method of compensating current. Since the voltage source converter is important
part in APF, more care is given to design of DC side capacitor. The controlling of DC
capacitor voltage along with harmonic current control is adopted here to improve the system
behavior. Efficiency of operation can be increased by modifying fuzzy controller with fuzzy-
hysteresis controller, the performance of proposed shunt active power filter has been verified
and the results obtained gives better accuracy in comparision to others technique.

The current control with a fixed hysteresis band has the disadvantage that the switching
frequency varies within a band because peak - to - peak current ripple is required to be
controlled at all points of the fundamental frequency wave. Kale et al [111] have proposed an
adaptive band controller for APF. The adaptive hysteresis band controller changes the
hysteresis bandwidth as a function of reference compensator current variation to optimize
switching frequency and THD of supply current. This paper proposes a fuzzy-adaptive
hysteresis band control, where the hysteresis bandwidth can be easily calculated with the help
of a fuzzy logic controller (FLC). Another important task in the active filter design is the
maintenance of constant DC voltage across the capacitor connected to the inverter. This is
necessary because there is energy loss due to conduction and switching power losses
associated with the diodes and IGBTs of the inverter in APF, which tend to reduce the value of
voltage across the DC capacitor.

1.5 Motivations of the Work

20
  Several System identification techniques such as Least Square(LS), Recursive least
square (RLS), Extended least square (ELS) , Least mean square (LMS) and Kalman
Filter (KF) have been applied to power system frequency estimation. To increase the
speed of convergence, to reduce computational and settling time, an immediate
motivation is to apply some improved recursive techniques to power system frequency
estimation.
 Different harmonics estimation technique based on Recursive least square (RLS),
Least Mean Square (LMS), Kalman filter (KF) and variants of Kalman Filter such as
Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF) and Ensemble
Kalman Filter (EnKF) have been studied but the motivation is to develop an efficient
harmonics estimation technique which will be capable of estimating the harmonics
components at different critical situations of power system signal such as variation in
frequency, amplitude in presence of inter harmonics and sub harmonics and also for
the case of dynamics signal.

 There is scope in developing hybrid identification technique employing both classical

and optimization techniques for harmonics estimation. In that hybrid technique,
estimation problem can be treated as an optimization problem, so that estimation error
becomes very much reduced on minimizing a cost function which is sum of the
squared estimation error. .
 For Improvement in power quality, developed estimation algorithm may be utilized in
addition to different control algorithm for designing shunt active power filter to reduce
power quality index.
 Many works in literature, validate the estimation and elimination methods using
power system synthetic signal .The performance of the developed estimation and
elimination technique should be verified using dataset collected from industrial set up
and development of prototype experimental set up using fast processor such as
Microcontroller, DSP and dSPACE etc.

1.6 Problem Statement

21
The problem addressed in the thesis comprises of the following three sub problems such as
frequency estimation, harmonics estimation and harmonics elimination described below.

PC
Non linear
C
Load
Source Load
current current
PD
PG

Active Harmonics Frequency estimation

power filter Estimation Algorithm Algorithm

Fig. 1.2 Schematic of estimation and elmination problem

1.6.1 Frequency Estimation Problem

In the present day modern power systems, fast and accurate frequency estimation has become
quite vital. Any mistake in accurate estimation of frequency may threaten the frequency
stability or at least lead to system operation problems.The fundamental frequency is one of the
most important and sensitive parameters of a power system. Any imbalance between the
system load and generation directly affects the frequency and leads to frequency deviations.
So it is necessary to maintain frequency at its nominal value or as close as possible to that
value within a narrow limit. Deviation of frequency from its rated value is an indication of
imbalance between real power generation and load demand. Any mistake in accurate
estimation of frequency could cause inadequate load shedding by frequency relays, which
would eventually be resulted in a major grid collapse in a power system, Thus the frequency
estimation problem is concerned with developing efficient algorithms with a view to obtain
accurate estimation of frequency of the resulting distorted power system voltage signal.

1.6.2 Harmonics Estimation Problem

22
Increased demand of nonlinear loads such as power electronics devices in commercial,
industrial, residential, Telecommunication, traspotation sector and utility system etc, instigate
a significant amount of harmonics into power system which result in deviation of signals from
their sinusoidal nature. The distortion of current signal being complex will introduce inter-
harmonics and sub-harmonic components to power system if not properly filtered. As
harmonics and inter harmonics has hazardous effect on power system, appropriate and
effective filters must be designed to eliminate harmonics. Hence, the harmonics estimation
problem signifies to buildup accurate estimation algorithms for computing amplitude and
phase of the harmonics of the distorted voltage/current signal.

1.6.3 Harmonics Elimination problem

Power quality gradually decreases as the demand of the nonlinear load increases day by day,
the main cause of degradation of power quality is harmonics distortion in voltage/current
signals which has several adverse effect in the power system such as overheating of motor,
transformer and capacitor, increase in conduction loses and eddy current losses, premature
damage of electrical equipment, important data loss from computers used in offices, meter
reading with higher percentage of error, etc. Therefore mitigation of harmonics from voltage
and current wave forms has become a major concern of power and control researchers.
Initially Passive filters have been used to improve the power quality by compensating voltage
and current harmonics, but these are no longer used due to high cost, large size and resonance
effect. Also there is no possibility of using same passive filter for different loads. These
problems can be overcome by use of active power filter (APF) and several types of active
filter has been used to suppress the harmonics in power system but still it is necessary to
improve the quality of active power filter. In this context harmonics elimination problem is
intended to develop efficient control algorithms for active power filter to eliminate the
harmonics and enhance the power quality.

1.7 Objectives

23
  To apply an improved iterative algorithm such as Improved Recursive Netwon Type
Algorithm (IRNTA) to power system frequency estimation problem and to compare its
estimation performance with the existing classical techniques such as Recursive Least
square (RLS) and Kalman Filter (KF).
 To propose an efficient estimation technique such as VLLMS, to avoid more
estimation error and less speed of convergence of LMS in power system harmonics
estimation problem. Estimation performance of the proposed VLLMS is validated
using arduino duo micro controller as a processing platform.
 To achieve improvement in percentage error in estimation, processing time in
computation and performance at different critical conditions of power system, attempts
have been made in combining classical technique with optimization technique such as
MGA and BFO.
 To develop control algorithms such as synchronous detection technique with HCC and
indirect current control technique with FLC for designing active power filter to power
system harmonics elimination problem.
 To validate the proposed estimation and elimination technique on data obtained from
laboratory experimental set up using fast processors such as arduino duo
microcontroller, dSPACE 1104 and also from industrial set up.

1.8 Thesis/Book Organization:

Chapter 1 provides a brief back ground on the area of research of frequency and harmonics
estimation and design of filter using different control algorithm for harmonics elimination, it also
includes motivation and objective of the thesis followed by problem ststement. Thesis organisation is
discussed at the end of the chapter.
Chapter 2 Presents an Improved Recursive Netwon Type Algorithm (IRNTA) based
frequency estimation technique , The performance of such a formulation is studied for several critical
cases that often arise in a power system, e.g., sudden change in frequency, amplitude and phase of

24
signal, in presence of harmonics in the signal and the results obtained are compared with another two
iterative techniques such as Recursive Least Square (RLS) and Kalman Filter (KF) Finally, frequency
estimation of the laboratory data collected from normal working day of a Laboratory and is also
presented.
The current signals obtained from the captive power plant of a Ferro alloys industry is also
investigated for frequency estimation.
Chapter 3 Suggests a New estimation techniques that is Variable Leaky Least Mean Square
(VLLMS) applied to power system Harmonics estimation problem, after the presentation of estimation
of frequency in the previous chapter, estimation of harmonics is carried out in this chapter. For the
purpose, power system signal modelled in a parametric form. Then the weight of the unknown
parameter is updated using VLLMS algorithm.From the updated weights of unknown parameter,
fundamental as well as harmonics components of amplitude and phase are estimated. The performance
of the proposed VLLMS algorithm is compared with other existing algorithms such as Recursive Least
Square (RLS) and Least Mean Square (LMS), for different critical cases that may arises in a power
system such as sudden change in amplitude of signals and also dynamics signal.
The experimental prototype hardware setup is designed in the Labortary. Finally Harmonics estimation
of Labortary generation signal i.e. across the output of a single phase PWM inverter, using
arduino due microcontroller is also presented.
Chapter 4 Describes two hybrid algorithms such as KF-MGA and KF-BFO for
estimation of harmonics components of a power system. In chapter 3 only application of
signal processing technique have been discussed for estimation of harmonics but in this
chapter we have switch over to combination of signal processing and optimization technique
for the improvement of percentage error in estimation,processing time in computation and
performance in presence of inter and sub harmonics component. A comparative harmonics
estimation performance of two proposed strategy is made to show the superiority between
them. A labaratory proto type is also developed for generation of power system data inorder to
validate the efficies of the proposed algorithm.
Chapter 5A novel controllers that uses an indirect current controller in conjunction
with Fuzzy Logic Controller (FLC) with different fuzzy rules has been implemented for
harmonic elimination in a single phase system, A SAPF is designed using combined
synchronous detection and two level hysteresis current controller techniques for elimination of
harmonics in a three phase distribution system. A prototype experimental setup is designed in

25
dSPACE platform for verification of proposed elimination technique.
Chapter 6 provides comprehensive summary and conclusions of all estimation
approaches for power system frequency and harmonics estimation as well as all the control
techniques implemented for active power filtering. It also focuses on contributions of thesis
and scope for futute work

26
 Chapter 2
Power System Frequency Estimation Using Signal Processing
Techniques
2.1 Introduction
Power Quality (PQ) has become an important research subject due to the increased
number of non-linear components in modern power systems. The power electronic equipment
introduces harmonic distortions that can deteriorate power system quality and affect the
system performance [120]. In order to maintain the specified PQ level, it is crucial to control
the quality of the supplying voltage. Therefore, harmonic filters must be used to reduce the
level of harmonic distortions. The operation of these filters relies on the quality of the
monitoring of the harmonic distortion. The monitoring relies on the quality of transducers
used in the measurement scheme, as well as on the methodology for the assessment of signal
distortions. In the past, a number of methods have been developed to estimate harmonic
distortions in power systems. One of widely used methods is the Fast Fourier Transform
(FFT). This method is simple and efficient. However it is sensitive to the changes of the
fundamental signal frequency [48]. Furthermore, to achieve a fast convergence, the FFT data
window must be short enough, which might lead to limitations in terms of extraction of
frequency components. Least Square (LS) method has also been widely used as a parameter
estimator [46, 74]. It minimizes the sum of the squares between measured and observed
values [86]. In different applications, LS has shown the estimation accuracy with the presence
of large disturbance and frequency deviations. Another method, Kalman Filter (KF), improves
the quality of estimation by considering more rigorously the understanding of the signal noise
[85, 43, 47]. In a number of cases LS and KF have shown their advantages as compared to
FFT based techniques in processing signals consisting of variation in frequency. However, the
dynamics involved in LS and KF raises concern since it exhibits poor performance [44] with
respect to sudden change in amplitude, phase or frequency of signal i.e during dynamic
changes in power system parameters, convergence of the algorithm is slow maens it takes
more time to track the parameters of the signal. So far as identifications of non-stationary
systems are concerned, [95] discusses on time frequency signal filtration applied to recursive

27
method of modal parameter identification. Model order of the system is reduced and signal
components are separated by using adaptive wavelet filtering. For identification of Linear
Time Varying (LTV) systems [41] having non stationary properties and small magnitude
vibration, a typical subspace based technique is used to extract the observability range space
using the Singular Value Decomposition (SVD) of a general Hankel matrix. The varying
transition matrix is estimated at each moment through the SVD of two successive Hankel
matrices. A vector vibration response measurement based [42] parametric identification for
time varying structure is reported, where the identification is based on three simultaneously
measured vibration response signal obtained during a single experiment. An ARMA (Auto
Regressive Moving Average) [82] with varying co-efficient is also a very familiar choice in
using the adaptive estimate methods. As these models were initially proposed for scalar
(single-input and single-output) systems, the use of them is quite difficult in multivariable
systems. So, to address the more complex problem, state space model is preferred. Popularity
of the state space model is due to the latest development in the subspace-based methods for
state space model realization. Accurate spate space model of multi variable systems can be
obtained directly from input-output data using subspace methods.

In this chapter, an Improved Recursive Newton Type Algorithm (IRNTA) [43] based
frequency estimation [24] technique is developed. The performance of such a formulation is
studied for several critical cases that often arise in a power system, e.g., sudden change in
frequency, amplitude and phase of signal, in presence of harmonics in the signal and the
results obtained are compared with another two iterative techniques such as Recursive Least
Square (RLS) and Kalman Filter (KF) Finally, frequency estimation of the laboratory data
collected from normal working day of a Laboratory and also industrial setup is discussed.

The remaining of the chapter is organized as follows. Section 2.2 presents the frequency
estimation using the existing signal processing technique. Section 2.3 shows the proposed
algorithm and formulation of equations. Section 2.4 describes the frequency estimation based
on the proposed algorithm. Section 2.5 discusses the simulation results of the proposed
algorithm. Section 2.6 presents some experimental setups and test results for the data obtained
fromthem. Section 2.7 concludes the chapter.

28
2.2 Frequency Estimation using existing signal processing Techniques
2.2.1 Recursive Least Square (RLS) Algorithm applied to Frequency
Estimation

Let a distorted power system signal buried with noise is represented by the following structure
A(t )  A1 sin  t + 1  +  (t ) (2.1)

To estimate the signal A(t ) , the amplitude ( A1 ) , phase (1 ) and frequency ( f ) , equ (2.1)
can be written in discretized form as

A(k )  A1 sin kT cos 1 + A1 cos kT sin  +  (k )

A(k )   sin kT cos kT    T +  (k )

(2.2)

Where
   11  A1 cos 1
   21  A1 sin 1
Further notational simplification of (2.2) can be made by expressing this in regressor form
given by
A( k )  h(k ) +  ( k ) (2.3)

Where  (k ) Signal noise.

Using the RLS estimation technique, the parameters can be estimated using the following
computing steps
 
 ( k )   (k - 1) + K ( k )e(k )
(2.4)

Where  (k )  current value of estimate

 (k - 1)  Past value of estimate
K (k )  Kalman Gain

The error in the measurement is given by


e k   A k  - h k   (k - 1) (2.5)
T

The gain K is updated using the following expression

29
K (k )  P ( k - 1) h(k )[I + h(k ) T P ( k - 1) H ( k )] -1
(2.6)

Where P (k )  Error Covariance matrixand  (0    1) = Forgetting factor, the covariance

matrix can be updated using the following updatation law as given by

P (k )  [ I - K (k ) h(k ) T ]P ( k - 1) / 
(2.7)

Equations (2.4) to (2.7) are initialized at k  0 . Initial covariance matrix P (0) is usually
chosen to be very large. i.e. P  I , where  is a large number and I is a square identity
matrix.

After getting the final estimate of      T , the fundamental amplitude ( A1 ) and

phase (1 ) can be estimated as given below

A1  ( A1 cos 1 ) 2 + ( A1 sin 1 ) 2
A1   2 +  2 (2.8)

tan 1 


1  tan -1 (2.9)


Once the estimates of amplitude and phase are obtained, then fundamental frequency ( f )


can be estimated as follows. f is given by f  can be evaluated from the noisy
2
measurement A(k ) using equation (2.1) as given below
A(k )
sin(kT + 1 ) 
A1

 A( k ) 
kT + 1  sin -1  
 A1 

1  A(t ) 
f  [sin -1   - 1 ] (2.10)
2kT  A1 

2.2.2 Kalman Filtering (KF) for frequency Estimation

30
 The discretized voltages signal as described in section 2.2 is also considered for applying
KF to frequency estimation. The regressor form of signal as described in (2.3) is also
taken. Then applying Kalman Filtering algorithm to (2.3), unknown parameters

     T are estimated as follows

 
K  k   P  k / k - 1 (h P k / k - 1 h T + Q) -1 (2.11)
Where K is the Kalman gain, H is the observation matrix, P  I is the covariance
matrix, where  is a large number and I is a square identity matrix. Q Is the noise
covariance of the Signal,so the covariance matrix is related with Kalman gain with the
Following equation
  
P (k / k )  P ( k / k - 1) - K (k ) h P (k / k - 1) (2.12)

Hence the updated estimated state is related with previous state with the following
equation.
  
 ( k / k )   (k / k - 1) + K (k )( A(k ) - h  (k / k - 1))

(2.13)

After updatation of the unknown parameter matrix,  by Kalman Filtering, using

equations (2.8-2.10), frequency is estimated.

2.3 Signal Model Presentation and IRNTA Algorithm Development

Let us assume the following observation model of the input signal (arbitrary voltage or
current);
A(t )  h( x, t ) +  (t ) (2.14)

Where A(t ) is an instantaneous signal at time t ,  (t ) is a random noise, X is a suitable

parameter vector, and h . is expressed as follows:
M
h( x.t )  A0 e -t +  Ak sin( kt +  k )
k 1
(2.15)
For the generic model (2.15), a suitable vector of unknown parameters is given by

x  [ A0 ,  ,  , A1 ,....., AM , 1 ,.....,  M ]T
(2.16)

31
Where A0 is the magnitude of the decaying dc Component at t  0 ,   1 / T , T being the
time constant , M is the highest order of harmonics present in the signal,  is the
fundamental angular velocity i.e equal to 2 f , f being frequency , AK is the magnitude of
the k th harmonics. The number of unknowns, i.e., the model order, is

n  2M + 3 (2.17)

The Model (2.15) can be simplified, e.g., containing only the fundamental harmonic. This is
due to the fact to reduce the order of the system and our requirement i.e. determination of
frequency can be met with that simplification.The Model selection depends on the application,
i.e., on the features of the input signal processed.The vector of unknown model parameters
(2.16) can be estimated by applying non recursive NTA numerical algorithm (2.16) given by

� � ^
x k +1  x k + ( J kT J k ) -1 J kT ( A- h( x k )) (2.18)

Where k is an iteration index, J is an  m, n  jacobian matrix, A is an  m.1 measurement

vector, h is an  m.1 vector of nonlinear functions determined by the model assumed ,and
m is the number of signal samples belonging to the data window.The use of the NTA

necessitates inversion of normal equation matrix N   J KT J K  at every iteration. This

-1

requires much of CPU time, particularly when the model order is high. The real time
computation can be significantly reduced by applying the following recursive form of NTA
algorithm. So performance of computing using IRNTA becomes improved.

^  ^
x k +1  x k + Pk +1 J k +1 ( Ak +1 - h( x k )) (2.19)

Pk +1  1 /  k +1 (( Pk - ( Pk j k +1 j kT+1 Pk ) /( k +1 + j kT+1 Pj k +1 ))

(2.20)
T
jm  �
�j1, j2, j3 �
�is
Where the m-th row of the Jacobian matrix.For example, the third
element (i.e., the first derivative of angular velocity) is given by the following equation
M
j3  h( x) /    Ak kt cos(kt +  k ) (2.21)
k 1

32
Other elements can also be derived, in the same way.
The new recursive algorithm presented requires the appropriate selection of the sampling
frequency, the initial guess for X 0 and the forgetting factor  . Normally, one selects  near

1.0  e.g .,0.95  . Here,  is tuned to the dynamics of the input signal processed.

The quality of estimation depends upon on the pre-selected forgetting factor  . The faster
convergence and the bigger sensitivity to random noise can be achieved by setting  less than
1 0.5 .on the other hand the slower convergence and less sensitivity to random noise follows

for  near 1  e.g.   0.99  .

2.4 IRNTA based frequency estimation

The noisy single phase power system signal can be represented as

y (t )  A1 sin( 0 t + 1 ) +  (t )
(2.22)

Where  (t ) is a random noise, The vector of unknown parameter is given by

x ( k )  [ A1e  0 e 1e ]
(2.23)

A1e : Estimated amplitude, 0e : Estimated Angular Frequency

1e : Estimated Phase, The signal model in regressor form can be written as

hx( k )  A1e sin( 0 e kT + 1e ) (2.24)

Jacobin matrix is given by

 hx( k ) hx( k ) hx( k ) 

J  
 A1e  0 e 1e 
(2.25)

On applying the recursive form of NTA Algorithm

33
 Pk +1  1/ k +1 (( PK - ( Pk jk +1 jkT+1 Pk ) /(k +1 + jkT+1 Pjk +1 )) (2.26)
� � �
x k +1  x k + PK +1 J K +1 ( AK +1 - h( x k )) (2.27)

After estimating the unknown parameter x , Frequency can be calculated as


f 0 e (k )  x 12 (k ) / 2 * pi
(2.28)

2.5 Simulation Studies

This section presents simulation results performed with the IRNTA algorithm. We present the
performance of the algorithm in estimating frequency in different cases of power system
signal. MATLAB/SIMULINK is used for the purpose. The initial values of Covariance matrix
for all the three algorithms presented (RLS, KFand IRNTA) and forgetting factors for RLS
and IRNTA are given in Table -2.1. The model order selection is algorithm specific. In case of
RLS and KF, model order depends on system structure matrix and in case of IRNTA, it
depends on order of Jacobian matrix. For selection of starting parameters of RLS, KF and
model order, [18] is referred. Different cases of simulated signals that may represent several
well-known properties of real-time power system voltage signal are considered for the
simulation study.

Table-2.1 Parameters used for simulation studies (RLS, KF and IRNTA)

Algorithms  ( P  I ) 

RLS 100 0.96

KF 100 -
IRNTA 100 0.96

2.5.1 Sinusoidal signal in presence of noise

A 50 Hz signal with constant frequency but with random noises is generated with a 1
millisecond sampling interval. The comparative estimation performance of proposed
algorithm for SNR of 40 dB is shown in Fig. 2.1. Fig.2. 2 shows MSE in the estimation of

34
frequency of signal at SNR 40 dB. It is found from the Fig. that maximum MSE in case of KF
followed by RLS with noise is of the order of 10 -3. But MSE, in case of IRNTA algorithm,
converges to zero after 10th sample. Hence IRNTA outperforms over other two algorithms
(RLS and KF). A quantitative assessment of the results for SNR of 40 dB case is presented in
Table 2.2 Besides frequency values, the Table also lists frequency deviation and
computational time for the respective algorithms. The frequency deviation is computed as
frequency deviation = mean ((true value–estimated value) /true value). It may be noted that
the accuracy in estimation of frequency using IRNTA is better (error in estimation is only
0.006%) compared to other two algorithms. The computational time for estimation in case of
IRNTA is also least among these three methods.

53 RLS
KF
IRNTA
52
F req u ency in H z.

51

50

49

48

47

0 20 40 60 80 100
Sample No.

Fig. 2.1 Estimation of frequency from noisy signals with SNR 40 dB

35
 -3
x 10
5
RLS
KF
4 IRNTA

3
MSE

0
0 20 40 60 80 100
Sample No.

Fig. 2.2 Estimation performance in MSE of frequency of signal

Table 2.2 Comparative assessment of RLS, KF and IRNTA algorithms

Parameter RLS KF IRNTA

Estimated Frequency 49.84 49.78 49.997

Frequency deviation, % 0.32 0.44 0.006

Computational time (seconds) 0.1024 0.1375 0.0994

2.5.2. Jump in frequency in the signal

The next case is considered to be the signal having jump in frequency. The frequency change
from 50 Hz to 49 Hz using the proposed algorithm is presented in Fig 2.3. It is observed from
this Fig. that the proposed approach follows very closely the actual system frequency in case
of IRNTA as compared to other two algorithms.

36
 51
RLS
KF
50.5 IRNTA
F req u en cy [H z]

50

49.5

49

48.5
0 20 40 60 80 100
Sample No.

Fig. 2.3 Estimation during sudden frequency change of 49 Hz from 50 Hz

2.5. 3. Jump in amplitude of signal

Here the sudden change in amplitude of signal is taken into account. Fig. 2.4 shows the signal
whose amplitude changes from 1 p.u. to 1.5 p.u. at 40 th sample. Fig. 2.5 presents the
comparative estimation performance of frequency using the proposed algorithm, from this
Fig. it is found that IRNTA estimates accurately the frequency even if there is some change in
amplitude of signal.

37
 2

A m p litu d e in P .U .
1.5

0.5

-0.5

-1

-1.5
0 20 40 60 80 100
Sample No.

Fig. 2.4 Single phase signal during change of amplitude at 40th Sample

51
RLS
KF
50.5 IRNTA
F req u en cy in H z.

50

49.5

49

48.5

48
0 20 40 60 80 100
Sample No.

Fig. 2.5 Estimation of Frequency during sudden change in amplitude

2.5. 4 Change in phase of signal

Fig. 2.6 shows a single phase signal when there is change in phase from 60 0 to 1200 at 40th
sample. Fig. 2.7 shows the estimation of frequency during sudden change of phase. Here also
estimation accuracy of IRNTA is more as compared to RLS and KF algorithms .

38
 1

A m p litu d e in P .U . 0.5

-0.5

-1
0 20 40 60 80 100
Sample No.

Fig. 2.6 Single phase signal during change of phase at 40th Sample

80
RLS
75 KF
IRNTA
F req u en cy in H z .

70

65

60

55

50

45

40

35
0 20 40 60 80 100
Sample No.

Fig. 2.7 Estimation of Frequency during sudden change of phase

2.5. 5 In the presence of harmonics

Next we consider the problem of estimating fundamental frequency (50 Hz) from signals
having harmonics content in them. The common case of 3 rd harmonic is considered. Fig.2.8

39
shows the single phase signal containing this harmonic. Fig. 2.9 shows the comparative
estimation of frequency using proposed algorithm from the signal with harmonics.

1
A m p litu d e in P .U .

0.5

-0.5

-1
10 20 30 40 50
Sample No.

Fig. 2.8 Single phase signal containing harmonics

75 RLS
KF
70 IRNTA
F re q u e n c y [H z ]

65

60

55

50

45

40

35
0 20 40 60 80 100
Sample No.

Fig. 2.9 Estimation of Frequency in presence of harmonics

2.5. 6 In presence of Sub and Inter harmonics

The performance of the proposed algorithm is then evaluated when sub and inter harmonics
are present in a signal. The frequency, amplitude and initial phase of the sub and inter

40
harmonics considered here are as 20 Hz, 0.4 p.u. and 650 and 130 Hz, 0.3 p.u. and 550
respectively. Estimation of Fundamental frequency of the signal using the three algorithms is
shown in Fig. 2.10. In this case also proposed algorithm estimates better compared to other
two.

80
RLS
75 KF
IRNTA
70
F req u ency [H z]

65

60

55

50

45

40

0 20 40 60 80 100
Sample No.

Fig. 2.10 Estimation of Frequency in presence of sub- and inter- harmonics

2.6. Experimental Results

So far performance of the IRNTA algorithm has been studied on simulated signals. It would
be interesting to have the same on some experimental data that captures many more features
arises due to the introduction of measurement and instrumentation errors. This section
presents such studies.

2.6.1 Validation on Real time (Data Collected from Industrial setup)

41
Three phase current signals generated at the terminals of the generators installed at the captive
power plant, FACOR Power Ltd, Bhadrak, Odisha, India are collected at a sampling rate of 1
kHz.

G 11KV
Generator Rating

Output MW/MVA-52/65
T/F
Terminal Voltage-11KV
132KV
P.F-0.8 lagging

Excitation Voltage-55V
CT Secondary
Excitation Current-12.5Amp

Connection-Star
Input: 0-1
Phase-3 Phase A.C Transducer AOutput: 4-20mA

Frequency-50Hz Bus Bar

Digital
DDCMIS
Distributed
Control
Monitory
Information
system

3 Phase Current

Fig. 2.11 Schematic Diagram of collecting industrial data

Fig. 2.11 shows the schematic diagram for the data collection set up that involves current
transformer to facilitate the measurement. The generated voltage is fed to 11KV/132KV
Generator transformer, the output of GT fed to the busbar. For acquiring digital current data,
the generator is connected to 400Amp/1Amp CT. The output of CT is fed to the Transducer of
output range 4-20 mA. Current signal of mA range is fed to DDCMIS (Digital Distributed

42
Control monitory information system) which consists of several components such as panel
terminal block, Analog input card (AT-810), Communication interference, Profibus
Communication, Controller (ABB 800m), Server, etc. DDCMIS provides digital data of
generator MW, MVA, pf, voltage and current. In the present work, we obtained three-phase
digital current data from the output of DDCMIS. The data almost corresponds to 50Hz signal
with noises. Since the captured signal does not have any typical visually dectable feature, we
omit its presentation. From these three-phase digital current data, frequency of the system is
estimated using the three algorithms (RLS, KF and IRNTA). Fig. 2.12 shows the performance
of the algorithms and it is observed that the KF generates largest error of more than 0.2 Hz.
Whereas IRNTA closely follow the actual frequency with errors less than 0.02 Hz. So error in
estimation in case of IRNTA is minimum and hence more effective.

50

49.8

49.6
Frequency[Hz]

49.4

49.2 RLS
KF
IRNTA
49
0 50 100 150 200 250 300
Sample No.

Fig. 2.12 Estimation of Frequency from industrial data

2.6.2 Validation on real-time (laboratory setup) data

43
The Data is obtained in laboratory from the supply on normal working day as per the
experimental setup shown in the Fig.2.13 and Schematic Diagram of Experimental Data
collection is shown in Fig.2.14.

Fig. 2.13 Experimental set up for Laboratory data generation

Specifications of the Instruments used are:

1. Autotransformer: Maximum load 15 Amps, 4.05 kVA, input 240V and output 260V @
50/60 Hz frequency.
2. Isolation Transformer: 230/115V (2 in Nos.) (Supply to measuring devices isolation
and between Supply and DSO isolation)
3. Nonlinear Load: a single phase diode bridge rectifier with a 100 Ω resistor in series
with a 250mH inductor at the dc side.
4. Digital Storage Oscilloscope (Tektronix with kit no TPS2014): Band Width-100 MHz,
Sample rate-1 GS/s, 4 channel, Record length-2500 data points, PC Connectivity- 9
pin female port and PC Communication software.
5. Voltage Probe: A 10X voltage probe is used to capture the distorted signal from the
6. PC: 1.46 GHz CPU and 1GB RAM, desktop computer. It stores the data in the form of
discrete form with the help of pc communication software, which is captured by the
DSO.
The waveform is stored in the Digital Storage Oscilloscope and then through the Open Choice
PC Communication software it is acquired in the personal computer. From the digital voltage

44
data, Estimation of Frequency is carried out using proposed algorithm. As per International
Electro-technical Commission 61000-4-30 [9], for computing the power quality parameters,
10 cycles in a 50-Hz system which is 200-ms windowing at a sampling time of 0.4 ms has
been used. The variation of estimated frequency with the proposed algorithm is shown in
Fig.2.15, which shows that the estimated value closely approaches the actual value of 50 Hz.

SINGLE
Auto ISOLATION PHASE R-L
Transformer TRANSFORMER(1:1)
FULLWAVE LOAD
DIODE
BRIDGE
230V,50 HZ RECTIFIER

SUPPLY

DSO PC

Fig.2.14 Schematic Diagram of Experimental Data collection

50.4 RLS
KF
50.2 IRNTA
F r e q u e n c y in H z .

50

49.8

49.6

49.4

49.2

49

48.8

48.6

0 50 100 150 200 250

Sample No.

Fig. 2.15 Estimation of Frequency of Experimental data

2.7 Chapter Summary

45
In this chapter, a Newton Type algorithm is presented for power system frequency estimation.
In this proposed algorithm, unknown parameter is updated in recursive form to estimate the
frequency. The estimation performance of the proposed algorithm is compared with two
existing recursive techniques such as Recursive Least Square (RLS) and Kalman filter (KF).
However, initialization of covariance matrix is very much important for RLS and KF
algorithms because improprer choice of these parameters may lead to more computational
burden, more convergence time and more estimation error. To overcome these problems in the
existing methods, proposed Improve Recucrsive Newton Type Algorithm (IRNTA) has been
developed. Both simulation and experimental studies have been performed to study the
efficacy of the proposed algorithm and a comparision made beween the three algorithms
(IRNTA, RLS and KF). Moreover, it has been observed that the proposed algorithm (IRNTA)
not only superior in minimizing the error but also superior in terms of convergence rate and
computational aspect as compared to RLS and KF.

46
 Chapter-3
Real time Harmonics Estimation of Distorted Power
System Signal
3.1 Introduction
Estimation of power system harmonics is now a day’s one of the important issues of power
system engineers. Due to significant growth of nonlinear loads in power systems, mostly
power electronic equipments, uninterruptible power supplies, arc furnaces and controlled
motor drives, current and voltage signals of power system become highly distorted. Main
sources of inter harmonics and sub harmonics in power system are power electronics devices
(e.g., cycloconverters), arcing loads (e.g., welding machines and arc furnaces), which are
widely used in rolling mills and linear motor drives [80]. Recently, grid integration of
Renewable energy sources based on power electronic converters also contribute to the
increasing concern for the betterment of power quality [124-125]. The aim of this work is to
develop efficient harmonics estimation technique and to design filter for removal of unwanted
harmonics distortion in power system.
In literature [83, 69, 72], so many algorithms have been applied for power system
harmonics estimation. The fundamental one is based on the Fast Fourier Transform (FFT). But
it fails in presence of inter-harmonics and variations in system fundamental frequency,
because of leakage and picket-fence effects [53-55],some methods [55, 34] are provided to
improve these drawbacks. Kalman Filter (KF) [58, 63] is the right choice for estimation of
harmonics contained in a power system signal [59-61]. However, to optimize estimations, the
higher-order terms in the Taylor’s expansion for this method were neglected because of a
nonlinear function including the formulation of measurements. As a matter of fact, estimation
of distorted signals may occur incorrectly or take longer to converge and even diverge.A
popular method for parameter estimation is Least Square (LS) algorithm [65]. The algorithm
is very powerful in estimating system parameters and is widely used for estimating harmonics
and their deviations in a signal. But online estimation is not possible in this case because of
it’s batch processing methodology.

47
 A power system digital harmonic analyzer[80] having built in FFT, where 68000
microprocessor- based instrument samples the waveform and calculates the harmonics of
signal for both current and voltage waveforms. Method based on Taylor-Kalman-Fourier [59]
was also used for instantaneous phasor and harmonics estimation of power system signals. A
method based on filter bank (FB) with adaptive filtering [82] was used for estimation of
parameters describing harmonics and inter harmonics in a power signal. Each harmonic and
inter harmonic component is estimated recursively. Advantages and disadvantages of adaptive
filters such as cascade FIR comb filters and resonator based filters for estimation problem
were discussed and an improved &suitable combined algorithm [141] based on above two
filters using decoupled modules and multirate sampling was proposed for frequency and
harmonics estimation of power signal. Using wavelet transform [142], analysis on harmonics
distortion was carried out by many researchers. A fast and accurate approach for real time
estimation for time varying harmonics of voltage and current signal based on rotational
invariance technique (ESPIRIT)[143] assisted adaptive wavelet neural network (AWNN) is
presented. AWNN provides quick estimate whereas ESPIRIT handles for time varying signal
with higher accuracy.

At the same time, soft computing techniques, e.g. Fuzzy Logic, Neural Networks, Genetic
Algorithm and simulated annealing have also been used for harmonics estimation [74, 91,
121]. However, such soft computing approaches suffer from poor convergence rate and larger
computational burden. Artificial Neural networks have also been extensively studied as a
means of harmonic extraction [121].But, the initial choice of weight vector and learning
parameters affect the convergence characteristics of the Neural type Estimator.

The conventional LMS technique suffers from the problem of poor convergence rate if the
step-size for adaptation is fixed. This step size in LMS, depends inversely on input power, i.e.
it takes more time to learn about its input when step size is small and vice versa. Poor
convergence problem of LMS can be overcome by using time varying step size [23].

If the Lest Mean Square of the error is considered as the cost function to be minimized, then
due to the dynamic variation of signal, the linear weights of the filter may go unbounded or it
may take longer time to respond due to the stalling effect [23]. To overcome this, one may
employ the leaky LMS algorithm that incorporates the magnitude of the weights in the cost

48
function to avoid the parameter drifting or the stalling problem [24]. One may, however,
make the leak adjustment as a variable one so that it converges slowly along the worst-case
eigen direction (the direction of the eigenvector corresponding to the smallest eigenvalue of
the error auto-correlation matrix) as opposed to the best-case eigen direction (the direction of
the eigenvector corresponding to the largest eigen value) [92-93]. Although such leaky LMS
algorithm has been employed in stereophonic acoustic echo cancellation problem [53],
effectiveness of the same in estimating power system harmonics appears to be not investigated
so far. Moreover, the frequent dynamical changes in a power system motivate one to
implement such an algorithm for efficiently estimating power system harmonics.

In this Chapter, a Variable Leaky LMS (VLLMS) [91-93] based harmonics estimation
technique is developed. For the purpose of modelling power system signal in a parametric
form, the VLLMS algorithm is applied on this signal. The performance of such a formulation
is compared with other existing algorithms applied to harmonics estimation problem, e.g.,
RLS [63] and LMS [23] for several critical cases that often arise in a power system, e.g.,
sudden change in amplitude of signal, in presence of sub and inter-harmonics in the signal and
also for dynamic signal. Finally, harmonics estimation of laboratory generated signal i.e.
across the output of a single phase PWM inverter, using arduino due microcontroller is also
presented. In all the cases, the performance of the proposed algorithm is found to be superior
compared with that of RLS and LMS, algorithms. Moreover, a significant feature of the
VLLMS algorithm is to restrict the estimated parameters to drift beyond limit gives an
advantage in implementing other correcting circuitry based on the estimator.

The remaining of the chapter is organized as follows. Section 3.2 presents the proposed
algorithm and formulation of equations. Section 3.3 describes the harmonics estimation based
on the proposed algorithm. Section 3.4 discusses the comparative simulation results of the
proposed algorithm. Section 3.5 presents some experimental setups and test results for the
data obtained from them. Section 3.6 concludes the Chapter.

49
 3.2 Algorithms applied to Power System Harmonics Estimation
3.2.1 VLLMS based Harmonics Estimation
Let us assume the voltage or current waveforms of the known fundamental angular frequency
 as the sum of harmonics of unknown magnitudes and phases. The general form of the
waveform is

N
y (t )   An sin( n t +  n ) + Adc exp(- dc t ) +  (t ) (3.1)
n 1

Where N is the number of harmonics,  n  n 2f 0 , f 0 is the fundamental frequency,  (t ) is

the additive noise, Adc exp(- dc t ) is the dc offset decaying term.

After discretization of Eq. (3.1) with a sampling period, T one obtains the following
expressions

N
y (k )   An sin(n kT + n ) + Adc exp(- dc kT ) +  (k ) (3.2)
n 1

Invoking Taylor series expansion of the dc decaying term, Adc exp(- dc t ) and retaining

only first two terms of the series

y (k ) can be obtained as

N
y (k )   An sin( n kT +  n ) + Adc - Adc dc kT +  (k ) (3.3)
n 1

For estimation amplitudes and phases Eq.(3.3) can be rewritten as

N
y (k )   [ An sin( n kT ) cosn + An cos( n kT ) sin n ] + Adc - Adc dc kT +  (k ) (3.4)
n 1

Eq. (3.4) can be rewritten in parametric form as follows

y (k )  H (k ) X (3.5)

H ( k )  [sin(1 kT ) cos(1 kT ) ... sin( N kT ) cos( N kT ) 1 - kT ]T (3.6)

The vector of unknown parameter

50
 X  [ A1 cos(1 ) A1 sin(1 ) ... An cos( n ) An sin( n ) Adc Adc dc ]T (3.7)

The VLLMS algorithm [141] is applied to estimate the state. The algorithm minimizes the
square of the error recursively by altering the unknown parameter X k at each sampling
instant using equation (3.8) given below

X k +1  (1 - 2  k  k ) X k + 2  k ek yk (3.8)

Where the error signal is


ek  y k - y k (3.9)
Step size  k is varied for better convergence of the VLLMS algorithm in the presence of
noise.

k +1   k +  k Rk2 (3.10)

Where Rk represents the autocorrelation of e k and ek -1 . It is computed as

Rk  Rk -1 + (1 -  )e k e k -1
(3.11)
Where  is an exponential weighting parameter and 0    1 , and  (0    1) and   0
control the convergence time.The variable leakage factor  k [132] can be adjusted as

 k +1   k - 2 k ek yk X k -1 (3.12)
After the updating of the vector of unknown parameter using LMS algorithm, amplitudes,
phases of the fundamental and nth harmonic parameters are derived as

An  X 2
2N + X 22N -1  (3.13)

 X 
 n  tan -1  2 N  (3.14)
 X 2 N -1 

3.2.2 Harmonics Estimation using Least Mean Square (LMS) algorithm

The LMS algorithm is applied to estimate the state. The algorithm minimizes the square of
the error recursively by altering the unknown parameter X k at each sampling instant
using equation (3.15) given below
X k  X k -1 +  k ek yk (3.15)
Where the error signal is

51
 
ek  y k - y k

The step size  k is varied for better convergence of the LMS algorithm in the
presence of noise.
 k +1   k + Rk2 (3.16)
Where Rk represents the autocorrelation of e k and ek -1 . It is computed as
Rk  Rk -1 + (1 -  )ek ek -1

(3.17)
Where  is an exponential weighting parameter and 0    1 , and  (0    1) and
  0 control the convergence time.

3.2.3 Harmonics Estimation using RLS algorithm

The signal as described in section 3.1 is taken; the vector of unknown parameter X ,
as in (3.7) is updated using RLS as
X is updated using Recursive Least Square Algorithm as
 
X ( k + 1)  X ( k ) + K ( k + 1)e( k + 1) (3.18)

Error in measurement is

e( k + 1)  y ( k + 1) - H ( k + 1) T X ( k )

(3.19)

The gain K is related with covariance of parameter vector

K (k + 1)  P (k ) H ( k + 1)[1 + H (k + 1)T P( k ) H ( k + 1)]-1 (3.20)

The updated covariance of parameter vector using matrix inversion lemma

P (k + 1)  [ I - K (k + 1) H (k + 1) T ]P (k ) (3.21)
These equations are initialized by taking some initial values for the estimate at
instants t ,  (t ) and P . As the choice of initial covariance matrix is large it is taken
as P  I , where  is a large number and I is a square identity matrix.
After the updating of the vector of unknown parameters using Recursive Least Square
(RLS) algorithm, amplitudes, phases of the fundamental and nth harmonic parameters
and dc decaying parameters can be derived using (3.13)-(3.14)

52
 Start

Initialize,,,,, , and

Model the signal in parametric form using

(3.5)

Evaluate Error=Estimated voltage –Actual voltage using (3.9)

Update Step Size, Leakage Factor and Weight Vector on minimizing

error from (3.10), (3.12) and (3.8)

No
Is final iteration
reached

Yes

Get final updated Weight Vector

Estimate Amplitudes and Phases for Fundamental and nth

Harmonics using (3.13) and (3.14)

Stop

Fig.3. 1 shows the flow chart of the estimation scheme of VLLMS algorithm

53
3.3Simulation Results
3.3.1 Static signal corrupted with random noise and decaying DC component
To evaluate the performance of the proposed VLLMS algorithm in estimating harmonics
amplitude and phase, discretized signal having fundamental frequency 50 Hz is generated in
MATLAB. The power system signal used for the estimation, besides the fundamental
frequency, contains higher harmonics of the 3rd, 5th, 7th, 11th and a slowly decaying DC
component [90]. This kind of signal is typical in industrial load comprising power electronic
converters and arc furnaces.

y (t )  1.5 sin(t + 800 ) + 0.5 sin(3t + 600 ) + 0.2 sin(5t + 450 ) + 0.15 sin(7t + 360 )
(3.22)
+ 0.1sin(11t + 300 ) + 0.5 exp(-5t ) +  (t )

The above signal is corrupted by random noise (t)  0.05randn having normal distribution with
zero mean and unity variance. All the amplitudes, given are in p.u. values.Different
parameters [17,207,208] used in the simulation studies are selected on the basis of series of
experiments performed during the implementation of this method and are given in Table. 3.1

Table 3.1 Simulation Parameter

Initial
Parameter  Initial W    max  min 
P0 Initial 
Value 0.001 0.018 0.99 0.97 0 0.008 0.0001 0.01 1.1

54
 actual
RLS
2 LMS
Amplitude in p.u.

VLLMS
1

-1

-2
0.04 0.05 0.06 0.07
Time in sec.

Fig.3.2. Actual and Estimated signal using RLS, LMS and VLLMS

1.6

1.4

1.2
Amplitude in p.u.

0.8

0.6

0.4
RLS
0.2 LMS
VLLMS
0
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.3a. Estimation of amplitude of fundamental component of signal

55
 80

75
Phase in deg.

70

65

60
RLS
55 LMS
VLLMS
50
0 0.02 0.04 0.06 0.08
Time in sec.
Fig.3.3b. Estimation of phase of fundamental component of signal

0.4
RLS
0.35 LMS
VLLMS
0.3
Amplitude in p.u.

0.25

0.2

0.15

0.1

0.05

0
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.4a. Estimation of amplitude of 5th harmonic component of signal

56
 80
RLS
60 LMS
VLLMS
Amplitude in p.u.

40

20

-20

-40
0 0.02 0.04 0.06 0.08 0.1
Time in sec.

57
 Fig. 3.4b Estimation of Phase of 5th harmonic component of signal

0.4
RLS
0.35 LMS
Amplitude in p.u.

VLLMS
0.3

0.25

0.2

0.15

0.1

0.05

0
0 0.02 0.04 0.06 0.08 0.1
Time in sec.

Fig.3.5 Estimation of amplitude of 11th harmonics component of signal

58
 100
RLS
LMS
VLLMS
50
Phase in deg.

-50

-100
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.6. Comparison of MSE in Estimation of signal

Fig.3.2 shows actual vs. estimated value of signal using three different algorithms. Actual vs
estimated signal almost matches with each other with very less deviation in case of LMS
algorithm. Figs.3.3a and3.3bshow the tracking of fundamental amplitude and phase of signal
in presence of random noise and decaying dc components using RLS, LMS and VLLMS
algorithms Fig.3.4a and3.4b presents the comparative estimation of 5th harmonic amplitude
and phase of signal. Figs.3.5a and3.5b shows the estimation of 11 th harmonic amplitude and
phase of signal respectively. In the above estimation process, proper choice of co-variance
matrices optimally tune RLS algorithm. The RLS algorithm exhibits oscillations in the
estimated amplitude of fundamental and harmonics components in the presence of a distorted
signal and noise.These results are quite significant in tracking steady-state both fundamental
as well as harmonics components of a power system over a period of 24 hour for the
assessment of power quality and harmonics distortions. Fig.3.6 compares Mean Squared
Errors in the estimation of signal using three different algorithms such as RLS, LMS and
VLLMS. It is found from Fig.6 that MSE is in the order of 10 -3.Fig. 3.7 presents a sensitivity
analysis of estimation of power system harmonics using VLLMS algorithm. It gives an idea

59
regarding variation of estimation error in signal with respect to variation of sampling
frequency in estimation. It is found that estimation error in signal is very much reduced with
increase in sampling frequency and the minimum estimation error is 0.001 at a sampling
frequency of 3 kHz. From this Figs.3.1-3.7, it is seen that estimation accuracy using VLLMS
is more as compared to other two such as RLS and LMS.

Fig.3.7. Variation of Estimation error with sampling frequency

Table 3.2 Harmonics Parameters Estimated values using VLLMS

Parameters 3rd 5th 7th 11th

Amplitude 0.498 0.1999 0.1498 0.1013

Phase 59.97 44.96 35.87 29.995

Table 3.2 shows the amplitude and phase estimates of all harmonic components using VLLMS
in presence of noise and dc decaying components. On comparing the actual values of
parameters, it is found that VLLMS is a good estimator so far as estimation accuracy of
harmonic components is concerned.

60
3.3.2. Estimation of harmonics in presence of amplitude drift
Fig.3.8 shows the estimation of 5 th harmonic amplitude using the above three
algorithms. It is observed that all algorithms track the 5th harmonics change in amplitude from
0.2 to 0.5 p.u.with oscillations in case of RLS and LMS estimation.

0.8
RLS
0.7 LMS
VLLMS
Amplitude in p.u.

0.6

0.5

0.4

0.3

0.2

0.1
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.8. Estimation of 5th harmonic amplitude during amplitude drift

3.3.3. Harmonics estimation of signal in presence of inter and sub-harmonics

To evaluate the performance of the proposed algorithm in the estimation of a signal in
the presence of sub-harmonics and inter-harmonics, a sub-harmonic and two inter-harmonics
components are added to the original signal. The frequency of sub-harmonic is 20 Hz, the
amplitude is set to be 0.2 p.u. and the phase is equal to 75 degrees. The frequency, amplitude
and phase of one of the inter-harmonic is 130 Hz, 0.1p.u. and 65 degrees respectively. The
frequency, amplitude and phase of the other inter-harmonic is 180 Hz, 0.15p.u. and 10 degrees
respectively. Fig.3. 9 and 3.10 shows the estimation of phase and amplitude of a sub-harmonic
and an inter-harmonic respectively.

61
 85

80
75
Phase in deg.

70
65
60

55
RLS
50
LMS
45 VLLMS
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.9. Estimation of sub-harmonics having phase 75 deg.

0.35
RLS
0.3 LMS
VLLMS
Amplitude in p.u.

0.25

0.2

0.15

0.1

0.05

0
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.10. Estimation of inter-harmonics having amplitude 0.1 p.u.

62
In the simulation studies the performance index (a measure of accuracy in estimation)  is
estimated by

N 
 ( y ( k ) - y (k ) ) 2

 k 1
N
 100 (3.23)
 y(k )
k 1
2


where y (k ) and y (k ) are actual and estimated signal respectively. In this case the
significance of the performance index  is that it provides the accuracy of the estimation
algorithm. Small value of  corresponds to more accurate estimation and vice versa.

The performance indices of all the three algorithms are given in Table 3.3 From which it can
be seen that VLLMS achieves significant improvements in terms of reducing error for
harmonicsestimation in comparison to other two algorithms. VLLMS exhibits superior
estimation performance and is more robust as compared to RLS and LMS owing to the fact
that unlike in RLS, the performance of the VLLMS is not influenced by the initial choice of
covariance matrix used in RLS.

Table 3.3 Comparision of Performance Index

SNR 40 dB 30 dB 20 dB

RLS 0.064 0.72 0.853

LMS 0.042 0.51 0.638

VLLMS 0.0034 0.24 0.418

3.3.4. Harmonic Estimation of a Dynamic Signal

To examine the performance of VLLMS algorithm in tracking harmonics and its
robustness in rejecting noise, a time-varying signal of the form

y (t )  {1.5 + a1 (t )} sin( 0 t + 80 0 ) + {0.5 + a3 (t )} sin(3 0 t + 60 0 )

(3.23)
+ {0.2 + a5 (t )} sin(5 0 t + 45 0 ) +  (t )

is used where the amplitude modulating parameters a1 (t ), a 2 (t ) and a3 (t ) are given by

63
a1  0.15 sin 2f 1t + 0.05 sin 2f 5 t
(3.24)

a 3  0.05 sin 2f 3 t + 0.02 sin 2f 5 t (3.25)

a 5  0.025 sin 2f 1t + 0.005 sin 2f 5 t (3.26)

f 1  1.0 Hz, f 3  3.0 Hz, f 5  6.0 Hz.

In case of static signal, which is discussed in subsection A, signal parameters such as

amplitude, phase and frequency do not change with respect to time but in case of a dynamic
signal, in the above example a1 (t ), a 2 (t ) and a3 (t ) change with respect to time. Here, the
random noise  (t ) has a normal distribution of zero mean, unity variance and amplitude of
0.05 randn.

The estimation of time varying 5th harmonic signal in the presence of random noise is
shown in Fig.3. 11 It is observed that there is more oscillation in estimation using RLS but
using other two algorithms such as LMS and VLLMS, the estimated value of amplitudes and
phases of signal closely matches with the actual with more accurate estimation using VLLMS
algorithm.

64
 0.22

0.21
Amplitude in p.u.

0.2

0.19

Actual
0.18
RLS
LMS
0.17 VLLMS
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.11. Estimation of 5th harmonics amplitude of dynamic signal

3.4 Experimental Studies and Results

In view of real-time application of the algorithm for estimating harmonics in a power system,
voltage data generation was accomplished in the laboratory by giving the 230 V supply
voltage to a regulated dc power supply. Then output of dc voltage regulator is fed to single
phase PWM inverter to generate the required ac test signal with harmonics. This 230 V ac test
signal is fed to measurement signal pre-processing circuit.

3.4.1. Measurement Signal Pre-Processing Circuit: It consists of the following

components as discussed below.

1. The 9 V step down Transformer: Since we will be measuring 230 V ac it has to be

stepped down to a suitable voltage range to be operated on and also the Transformer
provides sufficient isolation necessary to keep ourselves as well as the sophisticated
electronics from harm’s way.

65
 2. Voltage Divider (1/10): The output of the Transformer, when unloaded can give up to
14-17 volt Peak. Which can be further reduced by using a Voltage divider which
divides it 10 times which is the ratio of the resistances 10K and 100K
3. Voltage Offset Adding Circuit: The voltage after the voltage divider has both a positive
and negative peak, which is not suitable for our ADC. Therefore, we add an OFFSET
using the Rail to Rail swing OPAMP LM358.
4. The OPAMP functions as a voltage follower, and buffers the mid-point voltage that
appears at the junction of R3 & R4. This significantly reduces the impedance of the
voltage source, resulting in enhanced performance.
5. The Final Voltage Divider: Since the original circuit gives out a signal in the range of 5
V, but since we have used a microcontroller, which has an ADC capable of only
measuring a signal in the range of 3.3 Volt, a final Voltage divider has been used to
keep the voltage in range suitable for the ADC.
3.4.2 Voltage Supply Circuit (230 V ac to 5 V dc) of the microcontroller:
1. 230 / 6 V Step-Down Transformer: It steps down the voltage to 6 V to be used by our
power Supply Circuit for the microcontroller as well as the voltage offset circuit.
2. DiodeBridge Rectifier: A bridge rectifier made of four diodes rectifies the 6 V ac of the
transformer secondary, i.e. only positive parts of the signals i.e.dc.
3. Smoothing Capacitor: A 1000uF 25 V smoothing capacitor smoothes out the voltage
coming from the bridge rectifier.
4. CD7805 Voltage Regulator: It eliminates any ripple present in the voltage supply after
the capacitor so that a maximum allowable ripple of .005 Volts is present, i.e. the
Output is close to pure dc suitable for the sensitive microcontroller.
3.4.3. Arduino Due (Atmel SAM3X8E ARM Cortex-M3 CPU):

This is the microcontroller, which takes the analog voltage samples, at 20 samples for a cycle
for 5 cycles and then estimates the harmonics using the VLLMS algorithm implemented in C-
language. The variables are printed to a serial communication port which can be accessed /
recorded from a computer serial port monitoring software. These variables are then imported
and plotted in MATLAB.The experimental setup for implementing the proposed algorithm is
shown in Fig.3.12. The detailed circuit diagram of the experimental setup is given in Fig.3.13

66
Fig.3.12. Experimental setup of Arduino Due Microcontroller based hardware for
implementing the proposed algorithm

Fig.3.13. Circuit Diagram of the Experimental setup

67
 400
Actual
Amplitude in volt. 300 Estimated
200

100

-100

-200

-300

-400
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.14. Estimation of signal from real data

300

200
Amplitude in volt.

100

-100
Funda
-200 3rd
5th
-300 7th
11th
-400
0 0.02 0.04 0.06 0.08 0.1
Time in sec.

Fig.3.15. Estimation of Fundamental and other harmonic signals of real data

68
 400
Funda
350 3rd
5th
300
Amplitude in volt.

7th
250 11th

200

150

100

50

0
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.16. Estimation of Fundamental and other harmonics amplitudes of real data

100

50
Phase in deg.

Funda
-50 3rd
5th
7th
11th
-100
0 0.02 0.04 0.06 0.08 0.1
Time in sec.
Fig.3.17 Estimation of Fundamental and other harmonics phases of real data

69
 Fig.3.14 presents the estimation accuracy of the proposed algorithm on comparing the
actual and estimated signal. After initial half cycle, actual real generated signal and estimated
signal closely match with each other. Fig.3.15 shows estimation of fundamental and other
harmonics components of real signal. Fig.3.16 and Fig.3.17 show the fundamental and other
harmonics amplitudes and phases of real generated signal using VLLMS algorithm. Hence the
obtained results are satisfactory for the application with real data.

3.5 Chapter Summary

This Chapter presents a VLLMS algorithm for accurate estimation of amplitude and phase of
the harmonic components of distorted power system signal. The proposed method is based on
applying a leak adjustment technique so that drifting of parameters of the weight vectors is
avoided. Since step size is also updated in the proposed VLLMS algorithm, convergence of
the algorithm becomes faster. Improved harmonics estimation performance has been achieved
by using the VLLMS algorithm as compared to other two existing methods such as RLS and
LMS. Several computer simulation tests have been conducted to estimate harmonics in a
power system signal corrupted with random noise and decaying dc offsets. The performance
index of estimation of VLLMS in case of distorted signal at 20 dB noise has been found as
0.41 %. The above value is 0.1% for the case of experimental data. These are the minimum
value as compared to other estimation methods. The simulation results in various cases of
power system signal at SNR 20 dB and experimental results show the robustness of the
proposed VLLMS for effective harmonic estimation.

70
 Chapter 4
Development of Hybrid Algorithms for Power System
Harmonics Estimation
4.1Introduction

Estimation of harmonics of a distrubuted power system signal using different signal

processing technique such as RLS, LMS and VLLMS are discussed in chapter 3. In order to
achieve further improvement in computational time, percentage error and tracking time in
estimation, hybrid estimation algorithm using both signal processing and optimization
technique are presented in this chapter.

In the estimation process of these hybrid technique, unknown parameters are first estimated
using MGA/BFO Algorithm, optimized output of MGA/BFO are taken as initial value of
unknown parameter for starting the estimation using Kalman Filter Algorithm. The proposed
hybrid estimation approaches are verified in both simulation and Experimentation.

The difficulty in estimation of harmonics is due to the fact that harmonic generating loads are
dynamic in nature, which produces harmonics with time varying amplitudes. Thus, fast
methods of measuring and estimating harmonic signals are required. BFO [38] ,[70]is applied
by Mishra [74] to estimate the harmonic components present in power system voltage/current
waveforms. BFO rests on a simple principle of the foraging (food searching) behavior of
E.Coli bacteria in human intestine. Dash et al. [121] applied a neural network approach that is
adaptive by Widow-hoff delta rule for the estimation of harmonic components in a power
system. The authors [121] [38] adjusted the learning parameter of Adaline so that error
between the actual and desired output is minimized. In this context to tackle these problems,
KF based Bacterial Foraging Optimization (BFO) and Modified GA (MGA) are presented in
this chapter. Here in these techniques the power system signals with harmonics are estimated
using either BFO or MGA. Further KF is used to update the weights adaptively so that the
estimated output converges to the desired output of the signal.

71
In this research work, two hybrid approaches of KF-MGA and KF-BFO are proposed for the
improvement in percentage error in estimation, processing time in computation and
performance in presence of inter and sub-harmonic components. A comparative study of two
proposed strategy for estimation of harmonics in distorted power signals is made to show the
superiority in performance. A laboratory prototype is developed for generation of power
system data in order to validate the efficacies of the proposed algorithm

The chapter is organized as follows. Section 4.2 provides idea about power system harmonic
estimation problem. Section 4.3 gives the structure of two hybrid estimation schemes applied
to power system harmonics. Section 4.4 and 4.5 describes methods of individual KF-MGA
and KF-BFO schemes for harmonics estimation. Section 4.6 shows the simulated results for
different cases of harmonics estimation. Section 4.7 describes the experimental set-up
developed to validate the efficacy of the proposed algorithm. Section4. 8 conclude the chapter.

4.2Power System Harmonics Estimation Problem

Negative effects of harmonics currents and voltages are signal interference, over voltages,
data loss and circuit breaker failure, as well as equipment heating, malfunction and damage.
Harmonics are responsible for introduction of noise on telephone and data transmission lines,
huge increases in computer data loss, excessive heating in transformers and capacitors
resulting in shortened life or failure. Rotor heating and pulsating output torque caused by
harmonics can result in excessive motor heating and inefficiency. Let us assume the voltage or
current waveforms of the known fundamental angular frequency  as the sum of harmonics
of unknown magnitudes and phases

The general form of the waveform is

N
y (t )   An sin(n t + n ) + Adc exp(- dct ) +  (t ) (4.1)
n 1

Where, N is the number of harmonics  n  n 2f 0 ; f 0 is the fundamental frequency; μ(t) is

the additive Noise; Adc exp(- dc t ) is the probable d.c. decaying Term.

The discrete time version of (4.1) can be represented as

72
 N
y (k )   An sin( n kTs +  n ) + Adc exp(- dc kTs ) +  (k ) (4.2)
n 1

Approximating decaying term using first two terms of Taylor series as

y dc  Adc - Adc dc kTs (4.3)
Now Eq.(4.2) becomes
N
y ( k )   An sin( n kTs + n ) + Adc - Adc dc kTs +  ( k ) (4.4)
n 1

The nonlinearity arises in the model is due to phase of the sinusoids. From the discrete signal

4.3 Structure of the Proposed KF-MGA or KF-BFO Hybrid Estimation Scheme

applied to Harmonic Estimation

Input Power system(plant)

Desired
Output
+
Optimize unknown Output of
parameters using MGA/BFO as _
MGA/BFO initial weight of Estimated
Output Error
KF

Weight updating
using KF

Fig 4.1 Structure of KF-MGA or KF-BFO estimation scheme

Fig. 4.1 shows the estimation scheme of proposed KF-MGA or KF-BFO combined algorithm.
First input signal is fed to either MGA or BFO algorithm. Unknown parameters (weight
vectors before initialization) are optimized using any one of the said algorithm. Optimized
output of MGA / BFO is taken as the initial values of weights for KF algorithm. Then weights
are updated using the steps of KF algorithm. Fundamental as well as harmonic components
are estimated from final updated weights of KF. For estimation amplitudes and phases,
equation can be written as [4.4]

73
 [ An sin(nkTs )cosn + An cos(nkTs) sin n]
(4.5)
N
y( k )  
+ A - A  kT +  (k )
n1 dc dc dc s

The discrete signal y (k ) can be modelled in state space form as

x (k + 1)   I  ( 2 N + 2 )( 2 N + 2 )
[ x1 (k ) x 2 (k ) ... x 2 N -1 (k ) x 2 N (k ) x 2 N +1 (k ) x 2 N + 2 (k )]T (4.6)

y (k )  H (k ) x(k ) +  (k ) (4.7)

Where state transition matrix    I  ( 2 N + 2 )( 2 N + 2 ) (4.8)

Measurement matrix
H (k )  [sin(1 kTS ) cos(1 kTS ) ... sin( N kTS ) cos( N kTS ) 1 - kTS ]T
(4.9)
And

x  [ A1 cos(1 ) A1 sin(1 ) ... An cos( n ) An sin( n ) Adc Adc dc ]T

(4.10)

The vector of unknown parameter can be updated as

x( k )  [ x1 (k ) x2 (k ) ... x2 N -1 (k ) x2 N ( k ) x2 N +1 (k ) x2 N + 2 (k )]T (4.11)

The optimized output of the unknown parameter using either MGA or BFO algorithm is taken
as the initial values of unknown parameter for estimation using KF. The vector of unknown
parameter can be updated as

 
x(k + 1)  x(k ) + K (k + 1) ( k + 1) (4.12)

Error in measurement is

 (k + 1)  y (k + 1) - H (k + 1)T x (k ) (4.13)

74
 The gain K is related with covariance of parameter vector

K (k + 1)  P ( k ) H ( k + 1)[ H (k + 1)T P ( k ) H ( k + 1) + R ]-1 (4.14)

R is measurement noise covariance Rk at k th instant, R is initially taken as 100  [ I ]100100

Here the model relating the measurements to the states is obtained using the first principle
modeling. Deterministic behavior of the model is found from this modeling. Based on this
consideration process noise covariance is neglected and so process noise covariance Q is
taken as zero.

The updated covariance of parameter vector using matrix inversion lemma

P( k + 1)  [ P( k ) - K ( k + 1) H ( k + 1)T P( k )] (4.15)

These equations are initialized by taking some initial values for the estimate at instants k
H(k)and P. As the choice of initial covariance matrix is large it is taken P=αI where α is dc
decaying parameters are derived as

An  ( x 22N + x 22N -1 )

(4.16)

 x 
 n  tan -1  2 N 
 x 2 N -1 

(4.17)

Adc  x 2 N +1

(4.18)

x 
 dc   2 N + 2  (4.19)
 x 2 N +1 

Because

75
x  [ A1 cos(1 ) A1 sin(1 ) ... An cos(n ) An sin(n ) Adc Adc dc ]T

(4.20)

BFO or MGA is employed to optimize the initial weights x given in equation (4.20) by
minimizing the cost function as given below.

Ns Ns 
J (i, n, m, l )    (t )   [ y (t ) - y (t )]2
2
(4.21)
t 1 t 1

4.4Modified GA (MGA)
The following two modifications have been proposed Modification in parent selection
Modification in crossover mechanism

4.4.1 Parent Selection

Depending upon the values of fitness function, pairs of strings are selected from the
population pool at random for forming a mating pool. In a simple GA approach this is termed
as reproduction. And the strings are selected into the mating pool by simple Roulette wheel
selection. In this proposed algorithm, the following modifications are applied for the selection
of parents so that the strings with large values of fitness are copied more into the mating pool.

The first parent in each reproduction is the string having the best fitness value. The second
parent is selected from the ordered population using normal selection technique.

4.4.2Proposed Crossover
Crossover is an algorithm for artificial mating of two individual chromosomes with an
expectation that a combination of genes of individuals of high fitness value may produce an
offspring with even higher fitness. It represents a way of moving in the solution space based
on the information derived from the existing solutions. This makes exploitation and
exploration of information encoded in genes.

In this proposed algorithm, the following modifications have been proposed with an intuition
to have better trade-off between exploration of unknown solution space and exploitation of
already known knowledge of solution to find the global optimum in less number of

76
generations. In this work, one point crossover also called Holland crossover is adopted with a
probability Pc  0.6, 0.95 with modifications in exchange of chromosomal material In a
binary coded chromosome if the value of right most bits is changed 1  0, 0  1 , the search
point in the search space shifts to a nearby point. This helps in refining the optimum point in
the already known search space. As one proceeds towards the left from the right most bit of
the chromosome, the shifting of search point in the search space increases and it depends on
the position of the bit in the chromosome whose value is changed. The shifting is highest with
the change in the left most bit. This facilitates to explore new region in the search space by
shifting the search point wide apart from the current optimum position in the search space.

Therefore, it is evident that the exploitation of already known region or exploration of

unknown region in the search space is relatively depending upon the position of the bit in the
chromosome whose value changes. In a chromosome change in the bits towards the right from
the middle position contribute more towards the exploitation of already known region.
Similarly, change in the bits towards the left from the middle position contributes more
towards Exploration of new region in the search space. This is shown in the Fig 4.2.

Middle Point

a7 a6 a5 a4 a3 a2 a1 a0

Exploration Exploitation

Fig.4.2 Example of Exploration and Exploitation in the Search Space

Thus the positional dependency of crossing site in respect of middle point of the chromosome
helps to maintain diversity of the search point as well as improve the value of already known
optimum value. Here the mechanism of crossover is not same as that of one point crossover.
In this proposed scheme, the exchange of chromosomal material between two parents is made
considering the position of crossover site with respect to the midpoint of the chromosome. If
the crossover site falls towards the right of the midpoint of the chromosome, the right side

77
 chromosomal material from the crossover site of the fitter parent is replaced with that of other
parent to form one offspring. If the crossover site falls towards the left of the middle position
of the chromosome, the left side chromosomal material from the crossover site of the fitter
parent is replaced with the other parent to form on offspring.

Fig.4.3 shows an example of crossover procedure. Thus by generating one random number,
only one offspring is produced by crossover. For each pair of parent, two random numbers are
generated to produce two offspring. The flow chart of a modified GA for obtaining optimized
unknown parameters (weight vectors before initialization) is shown in Fig.4.3

1 0 0 1 1 1 0 1
Filter 1 0 0 1 1 1 0 1
Parent 1
Crossover
Site
Parent 2 0 1 0 0 0 1 1 0

0 1 0 1 1 1 0 1

Offspring 1 Offspring 2

Crossover site falls towards right Crossover site falls towards right of
of the Middle point the Middle point

Fig.4.3Example of Crossover

78
 Start

Randomly generate initial population

Simulate system and evaluate performance index

Select 2 chromosomes from the population using

Proposed GA

Perform crossover on the 2 chromosomes selected using

Proposed GA

Perform mutation on the chromosomes obtained

Replace these for the least fit chromosomes of the existing population

Yes
Check
Done
Convergence
No
No
Done Is gen<

MAX gen?
Yes
Gen = Gen+1

Fig.4.4 Flowchart for Optimal weight vector using Modified GA(MGA)

4.5BFO Algorithm

The survival of species in any natural evolutionary process depends upon their fitness criteria,
which relies upon their food searching (foraging) and motile behavior. The E. coli bacteria
that are present in our intestines have a foraging strategy governed by four processes, namely,
chemo taxis, swarming, reproduction, and elimination and dispersal [178]

79
4.5.1 Chemotaxis
The characteristics of movement of bacteria in search of food can be defined in two ways, i.e.
swimming and tumbling together known as chemo taxis. A bacterium is said to be ‘swimming’
if it moves in a predefined direction, and ‘tumbling’ if moving in an altogether different
direction. Depending upon the rotation of the flagella in each bacterium, it decides whether it
should go for swimming or for tumbling, in the entire lifetime of the bacterium.

4.5.2 Swarming
For the bacteria to reach at the richest food location (i.e. for the algorithm to converge at the
solution point), it is desired that the optimum bacterium till a point of time in the search
period should try to attract other bacteria so that together they converge at the solution point
more rapidly. This can be achieved by using a penalty function based upon the relative
distances of each bacterium from the fittest bacterium.

4.5.3. Reproduction
The original set of bacteria, after getting evolved through several chemotactic stages reach the
reproduction stage. Here, the best set of bacteria gets divided into two groups. The healthier
half replaces the other half of bacteria, which gets eliminated, owing to their poorer foraging
abilities. This makes the population of bacteria constant in the evolution process.

4.5.4. Elimination and Dispersal

In the evolution process a sudden unforeseen event can occur, which may drastically alter the
smooth process of evolution and cause the elimination of the set of bacteria and/or disperse
them to a new environment. Most ironically, instead of disturbing the usual chemo tactic
growth of the set of bacteria, this unknown event may place a newer set of bacteria nearer to
the food location.

80
4.6 ProposedBFO Algorithm

1. Initialization of BFO Parameter

2. Elimination-dispersal loop: l=l+1
3. Reproduction loop: m=m+1
4. Chemo taxis loop: n=n+1
a) For i=1,2,….S
Compute value of cost function J(I,n,m,l)


J    2  t    Y  t  - Yˆ (t ) 
Ns Ns 2

t 1 t 1

Let J SW (i, n, m, l , )  J (i, n, m, l ) + J cc ( x (n, m, l ), P (n, m, l ))(i.e., add on the cell-to-cell

i

attractant effect for swimming behavior).Where

S i S
  P

J cc ( x, P (n, m, l ))   J cc ( x, x i (n, m, l ))   - d attract exp -  attract  ( x m - x i m ) 2 
i 1 i 1   m 1 
S
 P

+  hrepellant exp(- repellant  ( x m - x i m ) 2 
i 1  m 1 

J last  J SW (i, n, m, l )

End of for loop

b) For i  1,2...S take the tumbling /swimming decision

. Tumble
Generate a random vector  (i ), on - 1,1

c) Update parameter
 (i )
x i (n + 1, m, l )  x i (n, m, l ) + u  C (i )
 (i ) (i )
T

This results in an adaptable step size in the direction of tumble for set of solution of parameter i

d) If minimum (J) is less than tolerance limit then break all the loops.
5. If J  N C , go to 4, In this case continue Chemot axis since the life of bacteria is not over.
6. Reproduction
a) For the given m and l ,and for each i=1,2,3…..S,
N C +1
Let J health  J
j 1
SW (i, n, m, l )

Sort parameter in ascending J health

b) S r  S / 2 no.of set parameter with highest J health will be removed and other S r no. of set of
parameters with the best value split

81
7. If m  N rs go to 3, In this case ,Specified reproduction steps is not reached ,so start the next
generation in the chemo tactic loop.
8. Elimination dispersal

For i=1,2,…..S, with probability Ped ,eliminate and disperse each set of parameters

9. Obtained Optimized values for weights (parameter)

82
 Start Y
Z

Cost functionJ(i,j)
Initialize Initialize
unknown para of
Variables
KF as output
ofBFO

No
Elimination and Dispersal J(i,j)<J(i,j-
Loop Counter E=E+1 1) Update
unknown
i=i+1 parameters
Yes using KF
Yes

E>Ned Swim, Yes

SL=SL+1
Z N<Nf
No
No
Reproduction Loop SL<Ns No
Counter R=R+1

Yes Yes Tumble

R>Nre No
i>S
No

Yes Determine Ampl. and

Chemotactic loop phase from updated
X Counter, K=K+1 X unknowns

Yes
End
K>Nc

No

Fig.4.5. Flow chart of KF-BFO Algorithm

83
Fig 4.5 describes the proposed KF-BFO estimation scheme. This algorithm has two distinct
parts, in first part BFO is applied to optimize the unknown parameters and in second part
optimized output of BFO is again updated using KF algorithm. As a result the error between
the desired and the estimated output is minimized.

4.7Simulation Results and Discussion

4.7.1 Static signal corrupted with random noise and decaying DC component.
The power system signal corrupted with random noise and decaying DC component is taken.
The signal used for the estimation, besides the fundamental frequency, contains higher
harmonics of the 3rd, 5th, 7th, 11th and a slowly decaying DC component. This kind of signal
is typical in industrial load comprising power electronic converters and arc furnaces.

y (t )  1.5 sin(t + 80 0 ) + 0.5 sin(3t + 60 0 ) + 0.2 sin(5t + 450 )

+ 0.15 sin(7t + 36 0 ) + 0.1 sin(11t + 300 ) + 0.5 exp(-5t ) +  (t ) (4.21)

The signal is corrupted by random noise having normal distribution with zero mean and
unity variance. In the simulation work, in harmonics estimation, the values of different
parameters taken during both simulation and experimentation work are given in Table 1. The
no. of parameters to be optimized, =12 (without inter and sub- harmonics case), =18 (with
inter and sub-harmonics case) are taken. During simulation, an initial value of Measurement
noise covariance is taken as and process noise covariance is taken as zero. Where is an
identity matrix. The Genetic Algorithm parameters used in the present research work are
tabulated in

Table 4.1 Values of parameters used for simulation and experimental work employing BFO

S P NS NC Nre Ned Ped C(I) dattract Wattract hrenellent Wrenellent

100 12,18 3 5 10 10 0.25 0.001 0.05 0.3 0.05 10

Table – 4.2GA Parameters

84
Population size Maximum no. of Crossover probability Mutation

generations (Max. Gen.) Probability

100 40 0.6 0.01

2
Ampl., P.U.

KF-BFO
1.5 KF-MGA

1
0 10 20 30 40 50 60 70 80 90 100
Sample No.
0.6
Ampl., P.U.

KF-BFO
0.5 KF-MGA

0.4
0 10 20 30 40 50 60 70 80 90 100
Sample No.
0.25
Ampl., P.U.

KF-BFO
0.2 KF-MGA

0 10 20 30 40 50 60 70 80 90 100
Sample No.

Fig.4.6 Estimation of amplitude of Fundamental 3rd,5th harmonics

85
 0.16

A m pl., P .U .
0.155

0.15

0.145 KF-BFO
KF-MGA

0 10 20 30 40 50 60 70 80 90 100
Sample No.

0.12
A m pl., P .U .

KF-BFO
KF-MGA
0.11

0.1

0.09
0 10 20 30 40 50 60 70 80 90 100
Sample No.

Fig. 4.7 Estimation of amplitude of 7th and 11th harmonics

P hase,deg.

82
KF-BFO
80 KF-MGA

78
0 10 20 30 40 50 60 70 80 90 100
Sample No.
P hase,deg.

65
KF-BFO
60 KF-MGA

55
0 10 20 30 40 50 60 70 80 90 100
Sample No.
P hase,deg.

50
KF-BFO
45 KF-MGA

40
0 10 20 30 40 50 60 70 80 90 100
Sample No.

Fig. 4.8 Estimation of phase of Fundamental, 3rd and 5th harmonics

Figures 4.6 and 4.7 compare the estimation of fundamental, 3rd, 5th, 7th and 11th harmonics
components of signal respectively using both KF-BFO and KF-MGA algorithms. Comparison

86
shows that KF-MGA outperforms over KF-BFO in each case of estimation. Figures 4.8 show
a comparative estimation of phases of fundamental, 3rd and 5th, harmonics components signal
using both KF-BFO and KF-MGA algorithms. KF-MGA gives more correct estimation
compared to KF-BFO in this Figure 4.9.

1.2
KF-BFO
KF-MGA
1

0.8
MSE

0.6

0.4

0.2

0
0 100 200 300 400 500
Sample No.

Fig.4.9 Comparison of MSE in estimation of signal

Fig.4.9 shows the comparative estimation of Mean Square Error (MSE) of signal using the
two algorithms. From the figure, it is found that, MSE performance in case of KF-MGA is
comparatively better than KF-BFO

Table 4.3 compares the simulation results obtained by the proposed KF-MGA algorithm with
KF-BFO, Genetic Algorithm (GA) and BFO [70]. The final harmonics parameters obtained
with the proposed approach exhibit the best estimation precision where the largest amplitude
deviation is 2.6 occurred at the 11th harmonics estimation and the largest phase angle
deviation is 1.53580 occurred at the 3rd harmonics estimation. The computational time of
estimation using KF-MGA is the smallest (8.642 sec.) compared to other three algorithm

87
Table 4.3 Comparative assessment of Harmonic Estimation

Methods Parameter Fundamental 3rd 5th 7th 11th Comp

Time(s)
Actual F(Hz) 50 150 250 350 550
A(V) 1.5 0.5 0.2 0.15 0.1
Ø(0) 80 60 45 36 30
A(v) 1.476 0.487 0.182 0.157 0.0945
GA Deviation (%) 1.6 2.6 9 4.6 5.5
Ø(0) 80.58 62.46 47.05 34.36 26.9 15.425
Deviation (0) 0.58 2.46 2.05 1.64 3.1
A(v) 1.488 0.512 0.1947 0.1568 0.1044
BFO Deviation (%) 1.8 2.4 2.65 4.53 4.4
10.458
Ø(0) 80.47 57.9 45.83 34.46 29.12
Deviation(0) 0.47 2.1 0.83 1.54 0.88

A(v) 1.485 0. 51 0.195 0.1536 0.1036

KF- Deviation (%) 1 2 2.5 2.4 3.6
BFO 10.562
Ø(0) 80.44 58.2 45.78 34.62 29.324
Deviation 0.44 1.8 0.78 1. 38 0.676

A (V) 1.494 0.4973 0.2026 0.1535 0.0974

KF- Deviation () 0.4 0.54 1.3 2.3 2.6
MGA 8.642
() 80.3644 58.4642 45.7342 34.7245 29.4436
Deviation () 0.3644 1.5358 0.7342 1.2755 0.5564

4.7.2 Harmonics Estimation of Signal in presence of inter and sub-harmonics

To evaluate the performance of the KF-MGA algorithm in the estimation of a signal in the
presence of sub-harmonics and inter-harmonics, a sub-harmonic and two inter-harmonics
components are added to the original signal. The frequency of sub- harmonic is 20 Hz, the
amplitude is set to be 0.505 p.u. and the phase is equal to 75 degrees. The frequency,
amplitude and phase of one of the inter-harmonic is 130 Hz, 0.25p.u. and 65 degrees
respectively. The frequency, amplitude and phase of the other inter-harmonic is 180 Hz,
0.35p.u. and 20 degrees respectively.

88
 0.6
Ampl., P.U.
0.5 KF-BFO
KF-MGA
0.4
0 10 20 30 40 50 60 70 80 90 100
Sample No.
0.4
Ampl., P.U.

KF-BFO
0.3 KF-MGA

0.2
0 10 20 30 40 50 60 70 80 90 100
Sample No.
0.4
Ampl., P.U.

KF-BFO
0.35 KF-MGA

0 10 20 30 40 50 60 70 80 90 100
Sample No.

Fig.4.10 Estimation of amplitudes of a sub-harmonic and two inter-harmonics

Fig.4.10 shows the estimation of amplitudes of a sub-harmonic and two inter-harmonics

using the two algorithms. Using KF-MGA, the estimation is very much perfect with most of
the sample converge towards the reference value in each case of estimation.

In the simulation studies the performance index  is estimated by

N 
 ( y(k ) - y(k )) 2
  k 1
N
 100
 y(k )
k 1
2

89
 
Where y (k ) and y (k ) are actual and estimated signals, respectively. In this case, the
significance of the performance index  is that it provides the accuracy of the estimation
algorithm. Small value of  corresponds to more accurate estimation and vice versa.

Table 4.4 Comparision of Performance Index

SNR GA BFO KF-BFO KF-MGA

No noise 0.1564 0.1186 0.1062 0.0894
40dB 0.1754 0.1372 0.1146 0.0985
20dB 1.2056 0.8071 0.8032 0.7936
10dB 10.4386 5.2556 5.1634 4.6384

4.8 Experimental Validation

In view of real time application of the algorithm for estimating harmonics in a power system,
data is obtained in a laboratory environment on running a DG set on normal working day of
the laboratory as per the experimental setup shown in Fig. 4.11.

i) D-G set:(a) Alternator-3 phase, 50 Hz, Y connected, 415 volt, 1500 rpm, 55.8 A, 40
KVA (b)Diesel Engine- Borestroke=110116, 37.2 KW, 1500 rpm
ii) Rheostats: 100 ohm, 5 A (3 in no.)
iii) Non-linear load: 3 Phase diode bridge rectifier with a 10  resistor in series with
a 120mH inductor at the d.c side.
iv) Digital Storage Oscilloscope: Band Width-200MHz, Sample rate-2GS/s,
Channels-2, Record length-2500 data points, PC Connectivity- USB Port and Open
Choice PC Communication software, Probe-P2220
v) PC: 1.46 GHz CPU and 1GB RAM, Notebook PC

The waveform is stored in a Digital Storage Oscilloscope (Tektronix Ltd.) across almost 10-
ohm resistance (measured using multi-meter) and then through Open Choice PC
Communication software, data is acquired to the personal computer. In Experimental
validation, the sampling rate of 25 kHz is taken and there are 600 samples used for harmonics
estimation using KF-BFO and KF-MGA. Fig. 4.12 shows the estimation of voltage

90
signal using KF-BFO and KF-MGA algorithms from the real data obtained
from the experiment. In the said Fig., the estimated waveform approaches
the actual waveform over the cycle, having more accuracy in estimation in
case of KF-MGA as compared to KF-BFO.

Rheostat
415 V Rectifier Load
100, 5A (Three Phase
1500 rpm
DiodeBridge)
DG Set

Digital Storage PC (1.46 GHz, 1 GB RAM) with

Oscilloscope (200 PC Communication software
MHz)

Fig. 4.11Experimental setup for online data generation

20
actual
15 KF-BFO
KF-MGA
Amplitude in volt.

10

-5

-10

-15

-20
0 100 200 300 400 500 600
Sample No.

Fig. 4.12 Estimation of signal from real data

91
The performance of the proposed hybrid algorithm is very much dependent on the initial
choice of maximum and minimum value of unknown parameters taken. In this work,
maximum and minimum values as 20% deviation from their actual values were taken. By
using an optimal choice of parameters of the algorithms, faster convergence towards the actual
value of signal parameter can be achieved. Both the algorithms track the fundamental and
harmonic signals very well with the performance of tracking using KF-MGA is better than
KF-BFO but both the methods can be used for online and offline estimation.

4.9 Chapter Summary

This chapter presents two new hybrid algorithms which can accurately estimate the amplitudes and
phases of the fundamental as well as harmonic components contained in a power system signal
contaminated with noise. In the estimation process, the algorithms first apply MGA/BFO to estimate
the unknown parameters used for determining amplitudes and phases. Then final amplitudes and
phases of fundamental and harmonics components are estimated after updating the unknown parameter
using KF algorithm. Due to the hybridization of the KF algorithm with optimization algorithms such
as MGA and BFO, it is found that there is improvement in percentage error in estimation,
processing time in computation and performance in presence of inter and sub-harmonic
components in power system signal. Experimental results obtained from laboratory prototype
setup also prove the efficacy of the proposed KF-MGA algorithm as compared to KF-BFO
algorithm.

92
 Chapter 5
Development of Control Strategies for Harmonics
Elimination using Shunt Active Power Filter
5.1 Introduction
In recent decades, there are many studies about the issue of Harmonics and its mitigation
technique in power system. Due to the Vast application of non linear load such as Ballasts
used in Fluorescent tubes, static power converters, AC voltage governors (low power) used for
light dimmers and small induction motors correct phase angle, Metal diminution operation,
HVDC utilize large power converters, Switch mode power supplies(SMPS), small
uninterruptible power supplies(UPS) units, Three phase loads (Variable speed drives, ,Large
UPS Units) etc, causes harmonics in the distribution system [97-98][120] and the harmonics
currents causes over heating of transformer, over loading of neutral, un necessary tripping of
circuit breaker, over-stressing of power factor correction capacitors, skin effects etc. [100],
and hampers the quality of power, So the improvement of power quality is necessary at the
present age. Initially Passive filters, capacitors banks are used for harmonic elimination and
power factor improvement respectively. But it has associated with some disadvantages like
large size, bulky nature, resonance phenomenon etc. To overcome this disadvantages power
engineers developed several types of APF [100][122]. A brief discussion of several types
harmonics filter used for harmonics mitigation and its principle of operation discussed here.
5.2Types of Harmonics Filter
The power system needs harmonics filter to avoid the harmonics and to increase the system
efficiency. Harmonics filters are mainly classified in to three types.
1. Passive filter
2. Active filter
3. Hybrid filter

5.2.1 Passive filters:

The main constituents to form passive filters are Capacitor, inductor and/or resistor. The
passive filters are mainly classified into two types.

93
 (i) Tuned filters

(ii) High pass filters.

Tuned filters:

To filter out the particular harmonics frequency from the system Tuned filters are used

Fig.5.1. A single tuned Filter

Series connection of a capacitor, an inductor, a resistor and a separate out single frequency
harmonic form the structure of single tuned filter shown in Fig.5.1, providing low impedance
path to two harmonic frequencies is the main properties of a double tuned filter and it has
associated with the advantage of low loss at the lower frequencies. A double tuned filter is
shown in fig.5.2.

94
 Fig.5.2. Double tuned filter

High pass filter: High pass filter posses the characteristics of offering low impedance path to

all the high frequencies.

Fig.5.3 High pass filter

Fig. 5.4 shows a C-type high pass filter in which a capacitor is connected in series with the

inductor to provide low impedance path for low frequencies. It helps to reduce the loss at

low frequencies.

Fig.5.4 C-type high pass filter

95
Passive filters are connected in parallel with nonlinear loads such as diode/thyristor rectifiers,
ac electric arc furnaces, and so on. Among them, the combination of four single-tuned filters
to the fifth, seventh, 11th and 13th-harmonic frequencies and a second-order high-pass filter
tuned around the 17th-harmonic frequency has been used in a high-power three-phase
thyristor rectifier. The drawback of passive filters is that they create resonance condition at
particular frequencies they are intended to work for. This raises the magnitude of harmonic
voltages at that particular frequency.

5.2.2 Active Power Filter :

Pure active filters can be classified into two types according to their circuit configuration

I. Shunt (parallel) active filters

II. Series active filters

5.2.2.1 Shunt (parallel) active filters

Fig.5.5. Schematic diagram of a shunt active filter

Shunt APF is connected in parallel at the Point of Common Coupling (PCC) in between
source and nonlinear load. Main function of shunt APF design is to compensate current

96
harmonics caused by nonlinear load by supplying equal amount of harmonics at PCC but with
opposite polarity. Block diagram of shunt APF is shown in Fig.5.5. Active filter is an inverter
with some controllable switches. A voltage source inverter (VSI) or a current source inverter
(CSI) can be used as an active power filter.

In CSI based APF, a CSI is connected at PCC through second order low pass filter made up of
LF and CF as shown in Fig. 5.6. Current source of the inverter is replaced by a high DC side

inductor [123]. In CSI all the semiconductor switches must support unipolar current and
bipolar voltage. Earlier researchers were using Gate Turn-Off (GTO) thyristor with reverse
blocking capabilities. But recently to enhance research on CSI, Insulated Gate Bipolar
Transistor (IGBT) with a series diode is used.

Fig.5.6 Block diagram of VSI based shunt active power filter

Fig.5.7 Block diagram of VSI based shunt active power filter

97
In VSI the AC side of the inverter is connected at PCC through a coupling inductor. In the DC
side of VSI, DC source of inverter can be replaced by a large capacitor as there is no resistive
element used in the AC side of the inverter. As all the circuit elements are not ideal, to
overcome the internal resistance of the circuit elements some steps should be taken while
using VSI as APF. All switches must support bipolar current and unipolar voltage. So
semiconductor switches with anti-parallel diode are generally used. Block diagram of VSI
based shunt APF is shown in Fig. 5.7

Coming into the benefits and drawbacks of both VSI and CSI based shunt
APF, CSI fed APF is very efficient in low power conditions. But it is
comparatively heavier than the VSI fed APF. Also CSI fed APF has high DC
link losses due to use large inductor in the DC side of CSI. The major
drawback associated with VSI fed APF is that there is switching ripple in
the source current after compensation. There is no such significant
difference in harmonic compensation characteristics of both VSI and CSI
fed APF. But considering ease of implementation of VSI fed APF, its
performance development is considered in this thesis.

5.2.2.2 Series active filter circuit configuration

The series active filter is series connected with the power supply. This filter controls on the
basis of “Feedback” manner. The block diagram of a series active power filter is shown in the
fig.5.8.

Fig.5.8. Schematic diagram of series active filter

98
5.2.3 Hybrid Filters

Hybrid filters are based on the combination of active filters and passive filters. Such a
combination with the passive filter makes it possible to significantly reduce the rating of the
active filter. The task of the active filter is not to compensate for harmonic currents produced
by the thyristor rectifier, but to achieve “harmonic isolation” between the supply and the load
As a result, no harmonic resonance occurs, and no harmonic current flows in the supply.
Series connection of active filter and passive filter to form a hybrid filter is shown in the
Fig.5.9 and series connection of active filter and shunt passive filter is shown in Fig.5.10. To
provide viable and effective solutions to harmonic filtering of high-power rectifiers However,
they have difficulty in finding a good market because of the necessity of the transformer and
the complexity of the passive filter. A schematic diagram of the Harmonics mitigation filter
and principle operation of Shunt active power filter is shown in the Fig.5.11 and Fig.5.12
respectively.

Fig.5.9 Series connection of an active filter and a passive filter

99
Fig. 5.10 Combination of series active filter and shunt passive filter

100
 Harmonics Mitigation Filter

Low pass Filter Active power filter

Series APF Hybrid APF

Tuned Filter High Pass Filter

C-Type High
pass filter
Shunt APF
Single Tuned Double Tuned

Current Source Inverter Voltage Source Inverter

Combination Combination Combination

APF in
Of Shunt OfSeries Of Shunt
Series with
APF and APF and APF and
Shunt PF
Series APF Shunt APF Shunt APF

Fig.5.11 Schematic diagram of Harmonics mitigation filter classification

101
 1-Phase or 3- Interfacing Inductor/Transformer Non linear load
Phase Supply

System Variable
Power Circuit Interference
Detection

Getting Signal Generator

Reference
Over all System Controller
Signal
Estimator

Fig 5.12 Generalised block diagram of APF

The basic principle of APF is to utilise power electronics technologies to produce specific
currents components that cancel the harmonic currents components caused by the nonlinear
loads.Fig.5.12 shows the components of a typical APF system and their connections. The
information regarding the harmonic currents and other system variables are passed to the
compensation current/voltage reference signal estimator. The compensation reference signal
from the estimator drives the overall system controller. This in turn provides the control for
the gating signal generator. The output of the gating signal generator controls the power
circuit via a suitable interface. Finally, the power circuit in the generalised block diagram can
be connected in parallel, series or parallel/series configurations depending on the interfacing
inductor/transformer used.

In this Chapter, First a novel controller that uses an indirect current controller in conjunction
with Proptional Integral (PI) controller or Fuzzy Logic Controller (FLC) with different fuzzy
rule has been adapted.The indirect current controller with FLC is used to estimate peak
reference current besides maintaining the DC side capacitor voltage of the inverter nearly
constant.Current Controller based pulse width modulation (CCPWM) is used to generate the
switching signals of voltage source inverter.A two-level Hysteresis Current Control (HCC) in

102
conjuction with synchronous detection technique is applied to SAPF in a three-phase
distribution system. In order to mitigate harmonics produced by the nonlinear load connected
in a three-phase distribution system, a three-phase insulated gate bipolar transistor (IGBT)
based current controlled voltage source inverter (CC VSI) with a dc bus capacitor is used as
an Active Power Filter. The first step is to calculate SAPF reference currents from the sensed
nonlinear load currents by applying the synchronous detection method and then the reference
currents are fed to the proposed controller for generation of switching signals. Various
simulation results are presented to verify the good behavior of the SAPF with proposed two
levels HCC. For verification of Shunt Active Power Filter in real time, the proposed control
algorithm has been implemented in laboratory developed setup in dSPACE platform

5.3 Proposed Shunt Active Filter For Single Phase System

D2
Ls D1

source

D4
D3

Fig.5.13 Proposed Shunt active Power Filter

The basic components of a shunt active power filter are a single-phase full-bridge voltage-
source PWM Inverter, a DC bus Capacitor and an Inductor .The Inverter is connected to the
power supply network through the inductance of the inductor which ensures first the
controllability of the currents in the active filter and secondly it acts as a first-order passive

103
filter attenuating high frequency ripples generated by the inverter. The proposed model for our
work is given in Figure 5.13 and the magnitude of the parameters associated with the
proposed filter is given in Table 5.1.The SPSAPF is connected parallel with the single phase
line where the non –linear load is connected, due to the non-linearity behaviour of the load
harmonics are generated in the system and the filter cancels the effect of harmonics by
generating the same magnitude of harmonics current but having opposite phase with it.

5.4 Control Strategy applied to shunt active filters

The Controller is the most important part of the active power filter and recently a lot of
research is being conducted in this area[99], [102],[105.Conventional PI and proportional
integral derivative (PID) Controllers have been used to estimate the peak reference currents
and control the DC Capacitor voltage of the inverter. Mostly a Proportional- Integral (PI)
controller is used to control the DC link capacitor voltage as well as to estimate peak value of
reference source current for both single phase and three phases APF [103-105],[110]. When
the Source supplies a non-linear or reactive load it is expected to supply only the active
fundamental component of the load current and the compensator supplies the harmonics or
reactive component. The outer capacitor voltage loop will try to maintain the capacitor voltage
nearly constant which is also a mandatory condition for the successful operation of the active
power filter. The System losses are provided by the source in steady state. The Compensator
supplies the harmonics power, which manifests itself only on the reactive component of
power, however the conventional PI Controller requires precise linear mathematical model of
the system which is difficult to obtain under parameter variations and non-linear load
disturbance. Another drawback of the system is that the proportional and integral gains are
chosen heuristically [111].To overcome the demerits associated with PI Controller, recently
fuzzy logic controller (FLC) is developed to use as controller for many power electronics
system and active power filter application [106-108].

5.4.1 Reference source current generation

104
Generation of Reference source current is very much essential to operate the Acyive power
filter Sucessufully. Two methods namely time domain and frequencydomain has been adopted
by several researchers for reference current generation [110].

V*dc

Sin θs +
PI or Fuzzy
Logic Vdc
Controller
-

+
i*
Multiplier Low pass filter 2 Multiplier s

iL
+

Triangular Wave

-
S1
i*
s
Not S4
+
Current Control Limiter
- S3
+ -1
- Not S2
+

is

Fig. 5.14 Block Diagram of Control Mechanism

The frequency domain methods include, Discrete Fourier Transform (DFT), Fast
FourierTransform (FFT), sliding DFT (SDFT) and Recursive Discrete FourierTransform
(RDFT) based methods. The frequencydomain methods require large memory and
computation power .On theother hand, the time domain methods require less calculation and
are widely followed for computingthe reference current. Mostly used timedomain methods are
synchronous reference (d-q-0) theory, instantaneous real-reactive power (p-q) theory [109],
Multiplication with sine function [108] etc. The systematic steps adopted for indirect control
technique in conjunction with PI controller and Fuzzy Logic Controller used for Extraction of
current references shown in the fig.5.14.
105
Let us assume that the load current is expressed by the equation


I L   S    I Lh .Sin h S - h  (5.1)
h 1

This equation can be written in


I L ( S )  I L .Sin( S - h ) +  I Lh .Sin(h s - h )
h2

(5.2)

Now we can see that the load current consists of two parts Fundamental load current and load
current with harmonics again the fundamental load current consists of two parts

I L  S   I Lf + I Lh (5.3)

I Lf =Fundamental Part of the Load current, I Lh = Harmonic Part of the Load Current,
I Lf Consists of two parts i,e active part and reactive part

I Lf  I Lfa + I Lfr (5.4)

I Lfa , I Lfr Active
current and Reactive current component of fundamental Signal respectively
The Fundamental active Current

I Lfa  I L1Cos1Sin S (5.5)

The fundamental Reactive current
I Lfr  I L1Sin1Cos S (5.6)
For cancellation of harmonics and Compensation of Reactive power, we chose the reference
current I S which is equals to the fundamental active current I Lfa

I S  I Lfa  I L -  I Lh + I Lfr  (5.7)

For filtering the load current the active component of fundamental current is transformed into
the DC component. Which we can able to get by multiplying “sin ϴ S” both sides of equation-
5.2


I L ( S )Sin S  I L1 / 2.Cos1 - I L1 / Cos(2 s - 1) + Sin S +  I Lh.Sin(h S - h ) (5.8)
h2

106
The filtered output current therefore given by

I L  ( Sin S ) filtered  I Lf  I L1/ 2  Cos1 (5.9)

The design of dc bus voltage Vdc is used in control loop and should be regulated at a set
Value Vdcref .The dc side capacitor voltage is sensed, then compared with a reference
value .The signal error of comparison between the sensed voltage ( Vdc ) and its reference (
Vdcref ) is the input signal of the fuzzy regulator.The output of this regulator is considered as

the peak value of line current ( I SMax ).

5.4.2 PI Controller
A Proportional- Integral (PI) controller is used to control the DC link capacitor voltage as well
as to estimate peak value of reference source current for the proposed Shunt active power
filter (SAPF). The DC side capacitor voltage is sensed and compared with a reference
voltage. This error e  Vdc ,ref - Vdc at the nth sampling instant is used as input for PI
controller. The error signal is passed through Butterworth design based Low Pass Filter (LPF).
The LPF filter has cutoff frequency at 50 Hz that can suppress the higher order components
and allows only fundamental components. Its transfer function is represented as
H  S   K P + K I / S , where, K P is the proportional constant that determines the dynamic

response of the Dc-side voltage control and K I is the integration constant that determines the
settling time. The steady state error in the DC-side voltage is eliminated by the Controller.

5.4.3 Fuzzy logic controller

Fuzzy logic control is derived from fuzzy set theory introduced by Zadeh in 1965. It is one of
the soft computing tools that can take automatic decisions like human beings. In order to
implement the control algorithm of a shunt active filter in closed loop, the dc side capacitor
voltage is sensed and then compared with a reference value.

The obtained error ( e  Vdcref - Vdc ) and Change of error signal ce( n)  e( n) - e( n - 1) at
the nth sampling instant are used as input for the fuzzy processing. The output of the fuzzy
controller after a limit is considered as the amplitude of the reference current I S max .

107
For controlling the DC voltage fuzzy logic based controller has been adopted for our work,
Shunt active power filter dc bus capacitor voltage is an Important parameter to be controlled,
if this control is not done properly,source current will deteriorate and lapse from sinusoidal
wave form and affect the system seriosely, Since fuzzy control rules are derived from a
heuristic knowledge of system behaviour,[106], [107] Mathematical modelling and complex
computation are not needed,[128], [140] Simplicity of fuzzy control is based on using human
like linguistic terms in the form of IF-THEN rules to capture the non-system dynamics,[140]
The fuzzy logic controller which is used to regulate the dc bus capacitor voltage is sensed
using a voltage sensor and compared with the set reference voltage (v dc Ref),Input variables
of the fuzzy controller are capacitor voltage error (e),and change in voltage error (∆e) at the k th
sampling time, The rule table for the designed fuzzy controller is given in the Table 5.1&5.2
respectively, which is used in our works followed by [160]. For two inputs we took seven
membership functions, derived 49 rules and for the same two inputs we took five membership
functions and derived 25 rules. The block diagram of the systematic operation of Fuzzy Logic
Controller is shown in the Fig.5.15.

Knowledge Base Rule Base

Reference
Output
e Fuzzification Inference Defuzzification Plant

Sensor

Fig. 5.14 Block Diagram of Systematic operation of Fuzzy Logic Controller

Below Figures 5.15-5.17 shows the Input variable, Input variable error, output variable and
fuzzy structure for 7/7 rule

108
 Fig5.15 7/7 input variable

Fig 5.16 7/7 input variable error

Fig.5.17 7/7 Output variable

The Rules are manipulated as follows

Table 5.1 Fuzzy rule base for voltage control (7/7)

∆e
NB NM NS ZE PS PM PB
e
NB PB PB PM PM PS PS ZE
NM PB PM PM PS PS ZE NS
NS PM PM PS PS ZE NS NS
ZE PM PS PS ZE NS NS NM
PS PS PS ZE NS NS NM NM
PM PS ZE NS NS NM NM NB
PB ZE NS NS NM NM NB NB

109
Below Figures 5.18-5.20 shows the Input variable, Input variable error, output variable of
fuzzy structure for 5/5 rule.

Fig.5.18 5/5 input variable

Fig.5.19 5/5 input variable error

Fig.5.20 5/5 Output variable error

Table 5.2 Fuzzy rule base for voltage control (5/5)

∆e NB NS ZE PS PB
e
NB NB NB NB NS ZE

NS NB NB NS ZE PS

ZE NB NS ZE PS PB

PS NS ZE PS PB PB

PB ZE PS PB PB PB

110
5.5 Simulation Results of The Proposed Shunt Active Power Filter
This section presents the details of the simulation carried out to demonstrate the effectiveness
of the proposed control strategy of the shunt active filter for harmonics elimination and
reactive power compensation, the test system consists of a single-phase supply connected to a
non-linear load with series RL, SPSAPF is connected to the test system through an inductor
L, the values of the circuit elements used in simulations are given in Table 5.1.
MATLAB/SIMULINK is used to simulate the test system and the proposed active filters. The
simulation was conducted under three different conditions that is Without APF, with
Hysteresis current control and hysteresis fuzzy combined control along with different fuzzy
rule,the Comprehensive simulation results are presented below.

Table 5.3 Simulation Parameter of single phase system

Sl no Circuit Element Values

1 Nominal Frequency 50Hz

2 Source Voltage 230V/440V

3 DC-Link Capacitor 1000 µF

4 Capacitor Voltage 350V

5 Filter Inductance 1 mH

6 Line Impedance 3Ω, 0.010H

5.5.1 Results without active power filter

111
The non-linear load is connect to the single phase supply and the simulation results obtained is
shown in Fig.5.21 and the corresponding FFT analysis shown in Fig.5.22, THD Without
Active filter is 23.6% of fundamental. The load current and

40
Il(A m p)

20

-20

-40
0 0.01 0.02 0.03 0.04 0.05 0.06

200

100
V s(V olt)

-100

-200
0 0.01 0.02 0.03 0.04 0.05 0.06
Time in Sec
Fig.5. 21 without active power filter

Fig.5. 22 FFT analysis of Source Current without active power filter

5.5.2 Simulation results of shunt active power filter with PI current control
112
The load current I L , Source current I S and Supply voltage V S are shown in Fig.5.23.The
harmonic spectra of the supply current of shunt active filter with hysteresis current control
technique is shown in Fig.5.24, the THD is reduced from 20.64% to 7.63%.

20
il [A]

-20
40 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05

20
is [A]

0
-20

-40
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
200
Vs [V]

-200
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time [s]

Fig.5.23 SAPF with PI current control

Fig.5.24 FFT Analysis of source current of SAPF with PI current control

5.5.3 Simulation results of shunt Active Power Filter (SAPF) with Fuzzy Logic
control technique with 5/5 Fuzzy Rule
113
The load current I L , Source current I S and Supply voltage V S are shown in Fig.5.25.The
harmonic spectra of the supply current of shunt active filter with fuzzy hysteresis combined
current control technique along with 5/5 fuzzy rule is shown in Fig.5.26, The THD is reduced
from 20.64% to 5.24%.

20
il [A ]

-20
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05

20
is [A ]

-20

0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
200
V s [V ]

-200
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time [s]

Fig.5.25 SAPF with Fuzzy Logic Controller (5/5) Rule

Fig.5.26 FFT Analysis of Source current of the simulated wave form of SAPF with Fuzzy
Logic Controller (5/5) rule

5.5.4 Simulation results of shunt Active Power Filter (SAPF) with hysteresis
fuzzy combined control technique (7/7) fuzzy rule.

114
The load current I L , Compensating current I C , Supply current I S and Supply voltage V S
are shown in Fig.5.27.The harmonic spectra of the supply current of shunt active filter with
fuzzy hysteresis combined current control technique along with 7/7 fuzzy rule is shown in
Fig.5.28, The THD is reduced from 23.6% to 1.48%.

20
il [A]

-20
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
20
is [A]

-20
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
200
Vs [V]

-200
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time [s]

Fig.5.27 Simulated wave form of SAPF with Fuzzy Logic Controller (7/7) Rule

Fig.5.28 FFT Analysis of Source current of the simulated wave form of SAPF with Fuzzy
Logic Controller (7/7) Rule
Table 5.4: Comparison of THD reduction for each order of harmonics

115
Harmonic Order Without APF SAPF With SAPF with FLC SAPF with FLC
(5/5) Rule (7/7) Rule
PI Control
3rd 3.7644 1.896 0.8484 0.1849

5th 1.967 1.456 0.6222 0.1542

7th 1.213 0.5996 0.4391 0.1295

9th .845 0.4355 0.3065 0.1103

11th .3761 0.3249 0.2101 0.0927

13th 0.2133 0.2393 0.1379 0.0792

15th 0.1226 0.1741 0.088 0.0664

17th 0.0925 0.1253 0.0552 0.056

19th 0.0901 0.0881 0.039 0.0471

Table 5.5 Comparison of THD in Percentage of Fundamental

Different Control Techniques THD %

implemented
Without APF 23.6
SAPF with HCC 7.63
SAPF with Fuzzy (5/5) 5.24
SAPF with Fuzzy (7/7) 1.48

5.6 Description of the Three Phase Studied System

116
The active power filter consists of six power transistors (IGBT),Six power diodes, a dc
capacitor(Cdc),Three filter inductor (LC) and reference value of DC side capacitor
voltage(VDC,ref). The filter capacitors and reactors have the function of suppressing the
harmonic currents caused by the switching operation of the power transistors,reduction of
current harmonics is achieved by injecting equal but opposite current harmonics component at
the point of common coupling (PCC),there by canceling the original distortion and improving
the power quality on the connected power system. Besides it has the capability of damping
harmonics resonance between an exiting passive filter and the supply impedance. The shunt
active filter is shown in the figure 1

Fig.5.29 Three- phase shunt active power filter

5.6 Control Mechanisim

117
5.7.1 Hysteresis Current Controller
Hysteresis current control scheme is more advantageous for the control of active power filter
can be used to generate the switching signals of the inverter. It creates an environment for
fastest control with minimum hardware and has excellent dynamics. Conventional hysteresis
current controllers produce bad harmonic performance, by varying the hysteresis band its
performance can be improved to get a fixedswitching frequency.
There are different typesof Hysteresis current controllers available like two-level hysteresis
current controller and three-level hysteresis current controller. This work covers two-level
hysteresis current controller for the proposed active power line conditioner. The
hysteresis current control (HCC) scheme [116-117] is based on a nonlinear control as shown
in Fig.5. 30. The actual currents ica , icb , icc injected by the inverter are compared with the
* * *
reference currents (ica , icb , icc ) Resulting a error current, which is fed to the hysteresis
comparator in order to produce six switching gate pulses as shown in Fig.5.31, The error
current swings between the upper and lower hysteresis bands as presented in Fig.5.32, During
the up and down movement of the actual current, when it touches the upper hysteresis limit of
comparator-a, let the comparator output is 0 and the switching state of phase-a is being
assumed as S1= 0 and S2 =1. Similarly when the actual current reaches the lower hysteresis
limit of the same comparator, the switching state of the phase-a reverses, similar procedure
can be followed for switching state of other two-phases as shown in Fig.5.30.

The switching frequency of this control strategy can be determined as follows.

Referring to Fig.6.3, at any point of time if we resolve the rate of change of phase current can
be written as

dI I 2V IL
   dc  t  (5.10)
dt t L  2Vdc

In the equation (5.10)  2Vdc depends on switching state of inverter, rate of change of
inverter current is represented as I ,rate of change in current in time period is t .In the
hysteresis band complete switching cycle is from 0  t1  T

118
So the equation (5.10) can be written as for 0  t1  i.et  t1 

+ IL
t1  (5.11)
2Vdc

Again for the period t1  T (i.e.t  T - t1 )

The equation becomes

- IL
T - t1  (5.12)
- 2Vdc

By combining both the equations, total switching time can be written as:

2
1 V V
fs   dc  f max  dc (5.13)
T ILV dc IL

Where the inverter maximum switching frequency can be represented as f max

Fig.5.30 Block diagram of the hysteresis current control

119
 Fig.5.31 Diagram of two-level hysteresis current control

5.7.2 Synchronous Detection Method

The synchronous detection method is implemented for calculation of compensating currents in
which three phase source is feeding a highly nonlinear load. The balanced three phase source
currents can be obtained after compensation. The following steps are used for generation of
reference signal.
The three-phase Source voltage can be given as:

Vsa (t )  Vma sin t (5.14)

Vsb (t )  Vmb sin(t - 120 ) 0
(5.15)
Vsc (t )  Vmc sin(t - 240 0 ) (5.16)

3-phase current drawn by load given by


I La (t )  I
n 1
an sin(t -  an )
(5.17)

I Lb (t )   I bn sin(t -  bn - 120 0 )
(5.18)
N
I Lc (t )   I cn sin(t -  cn - 240 0 )
n 1 (5.19)

120
Step 1:
The 3-phase instantaneous power (P3∅) in the proposed system can be written as

 
v sa i La + v sb i Lb + v sc i Lc  p a + p b + p c  Vma sin t  I an sin(t -  an ) + Vmb sin(t - 120 0 ) I bn sin(t -  bn - 1
n 1 n 1

+ Vmc sin(t - 240 0 ) I cn sin(t -  cn - 240 0 )
n 1
(5.20)
Step-2:
The instantaneous power is passed through low pass filter (LPF), which blocks higher order
frequency component and only fundamental component is obtained from the output of LPF.

Pfund  Vma sin tI a1 sin(t -  a1 ) + Vmb sin(t - 120 0 ) I b1 sin(t -  b1 - 120 0 ) + Vmc sin(t - 240 0 ) I C1
Vma I a1
sin(t -  c1 - 240 0 )   cos a1 - cos(2t +  a1 ) + Vmb I b1  cos b1 - cos(2t +  b1 ) + Vmc I c1
2 2 2
 cos c1 - cos(2t +  c1 
(5.21)
Step 3:
The average fundamental power in 3-phase is given by
T
1 Vma I a1 V I V I
Pav 
T P
0
fund dt 
2
cos  a1 + mb b1 cos  b1 + mc c1 cos  c1
2 2
(5.22)

For 3-phase balanced nonlinear load the followings can be written as

3VI
Vma  Vmb  Vmc  V , I a1  I b1  I c1  I ,  a1   b1   c1   , P  cos 1 (5.23)
2

Step 4:
Using equation (5.23), the average power per phase can be written as
VI
 Pav  ph cos 1 (5.24)
2
Let I cos 1  I m  Maximum amplitude per phase fundamental current

121
 2 Pav  ph
Im  (5.25)
V

Step 5:
The Load current contains Fundamental, reactive and harmonic parts; if the active power filter
can be able to compensate the total reactive and harmonics parts then the source current
waveform will be Sinusoidal.
The fundamental component of the load current can be given as
I Fa  I m sin t (5.26)
I Fb  I m sin(t - 120 )
0
(5.27)
I Fc  I m sin(t - 240 0 ) (5.28)

* * *
The expression of reference current for shunt active power filter in each phase ica , icb , icc
*
ica  I La - I Fa (5.29)
icb  I Lb - I Fb
*
(5.30)
icc  I Lc - I Fc
*
(5.31)
After getting the reference current ,it is compared with the actual current by using hysteresys
current comparator to generate six switching pulses,which are used to control the IGBT either
by turning ON or OFF
5.7 Simulation Results and Analysis of three phase system

The system model as described in Fig.5.29 is developed in the MATLAB/Simulink

environment and input simulation parameters are given in Table 1.The objective of the
simulation is to examine the performance of SAPF considering balanced as well as
unbalanced nonlinear load applied to the studied system.
Table.5.6 Simulation Parameter for Three-phase Studied System
System parameters Values
Line Voltage 380V
Line Frequency 50Hz
Supply Impedance R=0.5Ω,L=1mH
DC Voltage 750V
Load Active Power=10kw,Inductive power=100Var
APF Inductor 30mH

5.8.1 Simulation Results using Shunt Active Power Filter for balanced case

Simulation waveforms for SAPF considering balanced nonlinear load are depicted in Fig.5.32.
The source voltage and load current are shown in Fig.5.32 (a) and Fig5.32 (b) respectively.
The load current as shown in Fig.5.32 (b) contains harmonics because of nonlinear load
122
connected to the system. The compensating current produced by SAPF and the source current
after compensation are shown in Fig.5.32(c) and Fig.5.32 (d) respectively. Fig.5.33 and
Fig5.34 shows the source current spectrum of phase-A without SAPF and with SAPF
respectively THD is reduced from 29.29% before compensation to 3.76% after compensation..

400
Phase A
300 Phase B
Phase C

200
Source Voltage (V)

100

-100

-200

-300

-400
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)

Fig.5.32 (a) Source Voltage Waveform

123
 6
Phase A
Phase B
Load C urrent (A ) 4 Phase C

-2

-4

-6
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)

Fig.5.32 (b) Load Current Waveform

Fig.5.32 (c) Compensating current Waveform

Fig.5.32 (d) Source current after compensation

80
Phase A
60 Phase B
Phase C
S ource C urrent (A )

40

20

-20

-40

124
-60

-80
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)
 Fig.5.33 FFT Analysis of Source current spectrum without SAPF of phase A

THD= 3.76%

2.5

2
Mag(%ofFundamental)

1.5

0.5

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.34 FFT Analysis of Source current spectrum with SAPF of phase A

Table 5.7Comparison of THD in Percentage for Balanced Case

Before Compensation After Compensation

29.29 3.76

5.8.2 Simulation results using Shunt active power filter for unbalanced case
The simulation results of a three phase source supplying a unbalanced bridge rectifier feeding
a resistive and inductive load along with a three phase VSI-based SAPF is discussed here. For
the studied system, Fig.5.35 (a-d) represent the simulated waveforms The source current
spectrum of different phases in the proposed system without and with SAPF is shown in
Fig.5.36 (a-c) and Fig5.38(a-c) respectively. It has been observed from Fig.5.36 (a) and
Fig.5.38 that the THD for phase A is reduced to 2.41% after compensation as compared to
22.97% before compensation. Similarly, the THD of source current for phase B is reduced to
3.02% after compensation comparison to 24.74% before compensation. For phase C, the THD
is reduced to 2.75% after compensation from 23.56% before compesation.

125
 500
Phase A

S o u rce V o lta g e s (V )
400 Phase B
phase C
300

200

100

-100

-200

-300

-400

-500
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)

Fig.5.35 (a) Source voltage under unbalanced condition

20
L o a d C u rre n t ( A )

Phase A
15 Phase B
phase C
10

-5

-10

-15

-20
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)

Fig.5.35 (b) Load current under unbalanced condition

126
 80
Phase A
C ompensating C urrent (A ) 60 Phase B
Phase C
40

20

-20

-40

-60

-80
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)

Fig.5.35 (c) Compensating under unbalanced condition using SAPF

100
Phase A
80
Phase B
60 Phase C
S ource C urrent (A)

40

20

-20

-40

-60

-80

-100
0 0.02 0.04 0.06 0.08 0.1
Time (sec.)

Fig. 5.35 (d) Source current under unbalanced condition after compensation using SAPF

127
 THD= 22 . 97%

4
ag(%ofFundamental)

3
M

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.36 (a) FFT Analysis of Source current spectrums without SAPF of Phase A

THD= 24.74%
9

6
ental)

5
Mag(%ofFundam

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.36 (b) FFT Analysis of Source current spectrums without SAPF of Phase B

128
 THD= 23 .56%

12

10

8
Mag(%ofFundamental)

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.36 (c) FFT Analysis of Source current spectrums without SAPF of Phase C

THD= 2 .41%

1.8

1.6

1.4

1.2
Mag(%of Fundamental)

0.8

0.6

0.4

0.2

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.37 (a) FFT Analysis of Source current spectrums with SAPF of Phase A

129
 THD= 3 .02%

2.5

2
Mag(%of Fundamental)

1.5

0.5

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.37 (b) FFT Analysis of Source current spectrums with SAPF of Phase B

THD= 2 .75%
2

1.8

1.6

1.4
Mag(%of Fundamental)

1.2

0.8

0.6

0.4

0.2

0
0 2 4 6 8 10 12 14 16 18 20
Harmonic order

Fig.5.37 (c) FFT Analysis of Source current spectrums with SAPF of Phase C

Table 5.8Comparison of source current in THD for unbalanced load

Phases Before Compesation After Compensation

A 22.97 2.41
B 24.74 3.02
C 23.56 2.75

130
5.9 Experimental Results and Analysis
To validate the effectiveness of the shunt active power filter along with Fuzzy Logic
Controller operated by 7/7 fuzzy rule as control mechanism is presented here, some
experimental results are provided in this section. The block diagram of complete hardware
structure is shown in Fig.5.40. The prototype of experimental setup is shown in Fig.5.41, Part
number SQL100A1600V rectifier with a 470  F electrolytic capacitor and a rheostat is used as
a nonlinear load. An auto transformer is used to gradually increase the voltage level up to 110
volts. A SEMIKRON inverter is connected in parallel in between source and load at the point
of common coupling through a ferrite core inductor having inductance 5mH. Out of three legs
of the three phase inverter, only two legs have been used in this experiment. The SEMIKRON
inverter comprises of dual Insulated Gate Bipolar Transistor (IGBT) drivers SKHI 22 AR,
IGBT switches (SKM 75 GB 123 D) and filter capacitor of capacitance 1100  F .Two voltage
sensors LV 25-P are used for sensing the source voltage and filter capacitor voltage. Two
current sensors LA 55-P are used for sensing the desired currents. Sensor circuits are made
according to their datasheet specifications. The output of the sensors is fed to the dSPACE
1104 control board. The required pulse is given to the driver input with proper amplification.
Waveforms are taken by oscilloscope through digital to analog converter port of dSPACE to
avoid noises.

R
L

Nonlinear C
Load V
C
I
O
I
L
Inductor
IGBT Switches
L
I
S (SKM 75 GB 123 D)

AC 4 IGBT Gate signals

I Source IGBT Gate Drivers
(SKHI 22 AR )

V
S Control signals
Voltage sensor
V
C
circuit
( LV25 -P)
I
S dSPACE 1104
Current sensor Control Board
I
O circuit
(LA55 -P)

131
 Fig.5.38block diagram of complete hardware structure

The FLC controlled Shunt active filter Simulink model is implemented using a dSPACE
system that allows to implement a real time controller directly in MATLAB/Simulink
environment [59] [76]. The simulink file is automatically converted in a C-Code file by the
real time workshop (RTW) of dSPACE System. The c-code file becomes source for the Real-
Time Interference (RTI) of dSPACE System, which with the help of a C Compiler/Linker,
produces and downloads the machine code in the controller board.

Non Linear Load dSPACE

Auto Transformer Inverter

Current Sensor Voltage Sensor DSO

Fig.5.39 The prototype of the experimental setup

5.11 Results of Experimental Setup

Fig.5.40 Load Current and Source Voltage

132
 Fig.5.41 Source Current and Source Voltage

Fig. 5.42Filter capacitor voltage with Fuzzy logic Controller

Experimental waveforms of source voltage and load current are shown in Fig.5.42 Source
voltage and source current waveforms of Fuzzy logic controller with 7/7 fuzzy rule is shown
in Fig.6.43 and the filter capacitor voltage with Fuzzy Logic Controller 7/7 rule is shown in
Fig.6.44. It is found that source current is approximately sinusoidal and in same phase to that
of source voltage, which implies the proposed control strategy is appropriate.
5.10Chapter Summary
Performance of the shunt active filter using different control techniques is evaluated through
Indirect current controller with FLC, which is used to estimate peak reference current besides
maintaining the DC side capacitor voltage of the inverter nearly constantShunt active filter is
investigated under steady state condition with PI controller and Fuzzy logic controller with

133
different fuzzy rules such as 7/7 and 5/5. Comparative analysis of the performance of shunt
active power filter for balanced and unbalanced cases are also presented. Simulation results
show that the shunt active power filter is able to minimize the harmonic contents in the phase
A, phase B and Phase C in three phase system. THD of supply current is minimized in all the
three phases by using the shunt active power filter with synchronous detection technique.
Prototype experimental setup is developed and proposed FLC with 7/7 rule is implemented for
verifying its effectiveness in mitigating harmonic.

134
 Chapter 6
Summary and Conclusions
6.1 Summary of the Work
The thesis has mainly investigated on frequency and harmonics estimation of signal and
development of SAPF with control technique for power conditioning. The novelty of the
present work is the introduction of IRNTA to frequency estimation, VLLMS, and hybrid
algorithms for harmonics estimation. Finally the various control techniques such PI and soft
computing approach such as fuzzy logic controller in conjuction with indirect controller and
synchronous detection technique applied to SAPF for enchancement of power quality.

 A brief review of problems and short coming of earlier techniques to power system
frequency and harmonics estimation are discussed.
 Estimation of power system signal frequency is investegated using RLS, KF and
newly proposed IRNTA algorithms under several critical conditions such as Jump in
frequency, sudden change in amplitude and phase of a signal, in the presence of
harmonics, inter harmonics and sub-harmonics.The estimation performance of the
proposed algorithm is compared with two existing recursive techniques such as
Recursive Least Square (RLS) and Kalman filter (KF). Both simulation and
experimental studies have been performed to study the efficacy of the proposed
algorithm and a comparision made beween the three algorithms (IRNTA, RLS and
KF). Moreover, it has been observed that the proposed algorithm (IRNTA) not only
superior in minimizing the error but also superior in terms of convergence rate and
computational aspect as compared to RLS and KF.
 Harmonics estimation of power system signal based on VLLMS algorithm is
developed.Power system signal is modelled in a parametric form, then VLLMS
algorithm is applied on this signal. The performance of such a formulation is compared
with other existing algorithms such as RLS and LMS for several critical cases that
often arise in a power system, e.g., sudden change in amplitude of signal, in presence
of sub and inter-harmonics in the signal and also for dynamic signal. Finally,
harmonics estimation of laboratory generated signal i.e. across the output of a single

135
 phase PWM inverter, using arduino due microcontroller is also investigated. In all the
cases, the performance of the proposed VLLMS algorithm is found to be superior
compared with that of RLS and LMS algorithms.
 Two new hybrid algorithms (KF-MGA and KF-BFO) for power system harmonics
estimation are suggestedfor the improvement in percentage error in estimation,
processing time in computation and performance in presence of inter and sub-
harmonic components. First input signal is fed to either MGA or BFO algorithm.
Unknown parameters (weight vectors before initialization) are optimized using any
one of the said algorithm. Optimized output of MGA / BFO is taken as the initial
values of weights for KF algorithm. Then weights are updated using the steps of KF
algorithm. Fundamental as well as harmonic components are estimated from final
updated weights of KF. For estimation of harmonics amplitudes and phases, both the
algorithms effectively able to track the fundamental and harmonic signals, the
performance of tracking using KF-MGA is better than KF-BFO but both the methods
can be used for online and offline estimation. A laboratory prototype is developed for
generation of power system data in order to validate the efficacies of the proposed
algorithm.
 An indirect current controller with soft computing approach such as Fuzzy Logic
Controller (FLC) with different fuzzy rules has been implemented in single phase
shunt active power filter for iomprovement of power quality.The indirect current
controller with FLC is used to estimate peak reference current besides maintaining the
DC side capacitor voltage of the inverter nearly constant.Current Controller based
pulse width modulation (CCPWM) is used to generate the switching signals of voltage
source inverter.The filtering performance of FLC based indirect curent controller with
different fuzzy rules such as 7/7 and 5/5, is compared with PI controller. From the
simulation results, it is found that FLC controller having 7/7 rule base shows better
filtering performance as compared to 5/5 rule base. the effectiveness of the hysteresis
current control technique applied to Shunt Active Power Filter (SAPF) for elimination
of harmonics in a three-phase distribution systemis presented. In order to mitigate
harmonics produced by the nonlinear load connected in a three-phase distribution
system, SAPF with two-level HCC is proposed. A three-phase insulated gate bipolar
transistor (IGBT) based current controlled voltage source inverter (CC VSI) with a dc

136
 bus capacitor is used as an Active Power Filter.Synchronous detection method is
applied for reference current generation and then the reference currents are fed to the
proposed controller in order to generate switching signals for VSI. Various simulation
results are presented to verify the good behavior of the SAPF with proposed two levels
HCC.Finally, a prototype experimental setup is developed in laboratory, implementing
FLC based indirect current controller with 7/7 rule base in dSPACE 1104.
Experimental results validate the efficacy of the proposed FLC based indirect current
controller for improvement of power quality.

6.2 Book Contributions

The contributions of the Thesis are as follows.

137
 Critical analysis of tracking time, computational time and estimation error of various
estimation techniques applied to frequency and harmonics estimation of power system
signal.
 Development of improved recursive Netwon type algorithm (IRNTA) for power
system frequency estimation with introduction of parameter updation technique to
avoid short comings such as poor convergence rate, more estimation error and more
computational burden of the conventional Recursive Least square algorith (RLS) and
Kalman Filter (KF).Finally, frequency estimation of the laboratory data collected from
normal working day of a Laboratory and also industrial setup is presented to prove the
effectiveness of the proposed algorithm.
 Development of a Variable Leaky Least Mean Square (VLLMS) based algorithm for
power system harmonics estimation. In the proposed algorithm a leak adjustment
technique is incorporated to avoid the parameter drifting or the stalling problem to
overcome short comings such as poor convergence rate, more estimation error and
more computational burden of the conventional LMS and RLS. Finally, harmonics
estimation of the proposed algorithm is invesitgated in real time application of a
laboratory generated signal i.e. across the output of a single phase PWM inverter,
using arduino due microcontroller .
 Developed two hybrid algorithms by hybridizing a signal processing technique such as
KF with optimization techniques such as BFO and MGAfor improvement in
percentage error in estimation, processing time in computation and performance in
presence of inter and sub-harmonic components during harmonics estimation of a
power system signal.A comparative study of two proposed strategy for estimation of
harmonics is made to show the superiority in estimation performance in case of KF-
MGA. A laboratory prototype is developed for generation of power system data in
order to validate the efficacies of the proposed algorithms.
 An indirect current controller with a soft computing technique such as FLC has been
developed for power quality improvement in a single phase system. The performance
of FLC based controller is compared with conventional PI controller in simulation
results. AShunt Active Power Filter (SAPF) with two-level Hysteresis Current Control (HCC)
technique along with synchronous detection method for improvement of power quality in a

138
 three-phase distribution system is developed. The efficacy of the proposed Fuzzy Logic
Controller is also verified experimentally implementing it in a dSPACE1104 platform.

6.3 Conclusions

139
The research studies conducted resultsthe following conclusions

 Three numbers of recursive algorithms such as RLS, KF and IRNTA have been
applied to power system frequency estimation. Out of these algorithms, simulation and
experimental results confirm that IRNTA provides better estimation results compared
to other two.
 For achiving more accuracy in harmonic estimation variation of leakage factor is
introduced in the new proposed VLLMS algorithm. It is found that estimation
accuracy, computational and tracking time in case of VLLMS better than conventional
LMS and RMS. Estimation performance of VLLMS algorithm is also verified
experimentally in implemenating it on Arduino duo microcontroller.
 Two hybrid algorithms such as KF-MGA and KF-BFO are also proposed for
harmonics estimation. The estimation performance of the proposed algorithm is also
verified using both simulation and experimental data. From the results it is conformed
KF-MGA provides better estimation results in each case.

 A FLC based shunt active power filter using an indirect current controller design for
improvement of power quality in single phase system. The performance of the
proposed controller is also compared with the conventional PI controller in simulation
studies. To prove the effeciveness of the proposed controller, an experimental set-up is
developed in the laboratory impementing it in dSPACE 1104.AShunt Active Power
Filter (SAPF) with two-level Hysteresis Current Control (HCC) technique along with
synchronous detection method is propsed for improvement of power quality in a three-
phase distribution system.

6. 4 Future Scope of Work

 Estimation of Frequency and Harmonics problems can also be extended in nonlinear

frame work where techniques such as Nonlinear Least Square, H ∞(Nonlinear setting)

140
 and EKF (Nonlinear setting) can be applied with further comparative assessment of the
methods presented in the thesis.

 Integrating Distributed Generation (DG) into electricity distribution network settles

new needs for power quality monitoring. In future this work can be extended in
designing active power filters for DG system handling uncertainties in nonlinear filter
dynamics

 Proposed research work can be extended to design and develop a novel Shunt Active
Power Filter (SAPF) with a FPGA based digital controller for effective filteration
perfomance and better reactive power compensation.

References
1. V. Backmutsky, V. Zmuzdikov, A. Agizim, and G. Vaisman, “A New DSP Method for
Precise Dynamic Measurement of the Actual Power-Line Frequency and its Data
Acquisition Application,” Measurement, vol. 18, no. 3, pp. 169–176, 1996

141
2. Djuric M.B., Djurisic, Z.R. “Frequency Measurement of Distorted Signals Using
Fourier and zero Crossing Techniques,” Electric. Power Syst. Res, vol.78, no.8, pp. 1407–
1415,2008
3. Duric, M.B and Durisic, Z.R “ Frequency Measurement in Power Network in the
Presence Of Harmonics using Fourier and Zero crossing Technique,” Power Tech, 2005
IEEE Russia, pp.1-6, 27-30 June 2005
4. M.V.V.S.Yalla, A digital Multifunction protective relays, IEEE Trans. Power Del,
vol.7, no.1,pp.193-201,1992
5. N.T. Stringer “The effect of DC offset on current –operated relays,” IEEE Trans. on
Ind. App, vol.34, no.1, pp.30-34,1998
6. P.J.Moore, J.H. Allmeling and A.T.Johns “Frequency Relaying Based on Instantaneous
FrequencyMeasurment,” IEEE Trans.Power Del, vol.11, no.4, pp.1737-742, 1996
7. T.S.Sidhu and M.S. Sachdev “An Iterative technique for fast and accurate
measurement of power system frequency,” IEEE Trans.on Power Del, vol.13, no.1,
pp.109-115,1998
8. G.Benmouyal “An adaptive sampling interval generator for digital relaying”
IEEETrans.Power Del, vol.4, no.3, pp.1602-1609, 1989
9. T.A.George and D.Bones “Harmonics power flow determination using Fast Fourier
Transform,” IEEE Trans. Power Del, vol.6, no.2, pp.530-535, 1991
10. L.L.Lai, C.T.Tse, W.L.Chan, and A.T.P.So, “ Real-Time Frequency and Harmonics
Evaluation Using Artificial Neural Networks” IEEE Trans.on Power Del, vol.14, no.1,
pp.52-59,Jan.1999
11. A.A. Girgis and F.M.Ham “A qualitative study of pitfalls in FFT,” IEEE
Trans.Aerosp.Electron.Sys, vol.16, PP.434-439, 1980
12. T.S.Sidhu “Accurate Measurement of Power System Frequency Using a Digital Signal
Processing,” IEEE Trans. on power Del ,vol.14, no.1, pp.75-81, 1999
13. T.Lobos and J.Rezmer “Real-time Determination of power system frequency” IEEE
Trans. On Instrumentation and Measurement, vol.46, no.4, pp.877-881, 1997
14. Jin Kwon Hwang, Penn N.Markham “Power system frequency estimation by reduction
of noise using three digital filter,” IEEE Trans. on instrumention and Measurement
,vol.63, no.2, pp.402-409, 2014
15. Tomas Radil, PedroM.Ramos, A.Cruz Serra “New Spectrum Leakage Correction
Algorithm for Frequency Estimation of Power System Signals,” IEEE Trans.on
Instrumentation and Measurement ,vol.58, no.5, pp.1670-1679, 2009

142
16. Arghya sarkar and SamarjeetSengupta “Bandpass Second-degree Digital –integrator
based power system frequency estimation under non sinusoidal conditions,” IEEE
Trans.on Instrumentation and measurement, vol.60, no.3, pp.846-851, 2011
17. H.Karini, M. Ghartemani, M.R.Iravani “ Estimation of frequency and its rate of
change for applications in power systems”Power Engineering Society General
Meeting,IEEE , vol.2, pp. 13-17, 2003
18. P.J.Moore, R.D.Carranza, A.T.Johns “ Model system Test on a new Numeric Method
of power system frequency measurement,” IEEE Trans. on Power Del, vol.11, no.2,
pp.696-701,1996
19. Mohsen Mojiri and Alireza Bakhahai “Robust Adaptive Frequency Estimation of
Three-Phase Power Systems,” IEEE Trans.On Instrumentation and Measurement ,vol.59,
no.7, pp. 1793-1802, 2010
20. WuJiekang, Long Jun, WangJixiang “High-Accuracy Wide-Range Frequency
estimation methods for power system signals under nonsinusodial Conditions” IEEE
Trans.on power Del, vol.20, no.1, pp. 366-374. 2005
21. YiliXia, DaniloP.Mandie “Widely Linear Adaptive Frequency estimation of
unbalanced Three-phase power systems,” IEEE Trans. on Instrumentation and
Measurement, vol.61, no.1, pp.74-83, 2012
22. Bernard Widrow “Thinking about Thinking: The Discovery of the LMS Algorithm,”
IEEE Signal processing Magazine, pp.100-103, 2005
23. A.K.Pradha, A.Routray, Abir Basak “Power System Frequency Estimation Using Least
Mean Square Technique,” IEEE Trans. On Power Del, vol.20, no.3, pp.1812-1816, 2005
24. P.K.Ray, B.D.Subudhi, S.Ghosh “Variable leaky least mean-square algorithm based on
power system frequency estimation,” IET Sci.Meas.Technol, vol.6, no.4, pp.288-297,
2012
25. Bernard Widrow, John McCool, and Michael Ball “The Complex LMS Algorithm”
Proceeding of the IEEE , Vol.63, no.4, pp.719-720, 1975
26. P.K.Dash, A.K.Pradhan and G.Panda “Frequency Estimation of Distorted Power
System Signals Using Extended Complex Kalman Filter,” IEEE Trans. on Power Del,
vol.14, no. 3, pp. 761-766, 1999
27. B.Subudhi,P.K.Ray,S.R.Mohanty, A.M.panda “ A Comparative Study of Different
Power System Frequency Estimation Technique,” Int. J. Automation and control,vol.3,
no.2, pp.202-214,2009
28. Solim Abdel –Hady, A.M. A1-Kandari, R.A.Alammari “Linear Kalman Filter
Algorithm with Clarke Transformation for Power System Frequency Estimation,” Global

143
 Journal of research in Engineering Electrical and Electronics Engineering, vol. 12, no. 1,
pp. 524-534, 2012
29. Liangliang Li, Wei Xia, Dongyuan Shi and Jianzhuang Li “Frequency Estimation on
Power System Using Recursive-Least-Squares Approach,” Proceedings of the
International Conference on Information Technology and Software Engineering, DOI:
10.1007/978-3-642-34522-7_2, Springer
30. P.K.Das, K.R.Krishnanand, M.Padhee, “Fast recursive Gauss-Newton adaptive filter
for the estimation of power system frequency and harmonics in a noisey environment,”
IET Generation,Transmission,Distribution, vol.5, no.12, pp.1277-1289, 2011
31. M.Gupta, S.Srivastava and J.R.P.Gupta “Power system frequency estimation using
neural network and genetic algorithm,” Proceedings of Joint International Conference on
Power System Technology and IEEE Power India Conference, POWERCON, pp. 12-15
2008
32. M.Gupta “ A Faster Estimation Algorithm Applied to Power Quality Problems,”
International Journal of Engineering science and Technology ,vol.2, no.9, pp. 4448-4461,
2010
33. A.Soundarrajan, S.Sumathi,G.Sivamurugan “ Hybrid Evolutionary Algorithm for
frequency and voltage Control in power generating System,” ICTACT Journal on soft
Computing, vol.2, pp.88-97, 2010
34. P.K.Ray, B.Subudhi, A.M.Panda “Hybrid Signal processing and soft computing
approaches to power system frequency estimation,” International Journal of computer
and Communication Technology (IJCCT), vol.2, no.4, pp.16-20, 2011
35. A. K. Deb, Jayadeva, M. Gopal, and Suresh Chandra “SVM-Based Tree-Type Neural
Networks as a Critic in Adaptive Critic Designs for Control,” IEEE Trans. On Neural
network, vol.18, no.4, pp.1016-1030, 2007
36. MaamarBettayeb and UvaisQidwai “A Hybrid Least Squares-GA-Based Algorithm for
Harmonic Estimation” IEEE Transactions on Power Delivery, vol.18, no.2, pp.377-382,
2003
37. Y. R. Sood, N. P. Padhy and H. O. Gupta “ Discussions of Optimal Power flow by
Enhanced Genetic algorithm,” IEEE Trans. on Power System, vol. 18, no. 3, pp. 1219-
1219, 2003.
38. P.K.Ray, B. Subudhi “BFO optimized RLS algorithm for power system harmonics
estimation, Applied soft computing, vol.12, no.8, pp.1965-1977, 2012
39. P.K.Ray “Signal Processing and soft Computing approaches to power signal frequency
and harmonics estimation,” Ph.D Thesis.

144
40. S.A. Soliman , R.A. Alammari , M.E. El-Hawary ‘Frequency and harmonics
evaluation in power network using fuzzy regression technique’ Electric Power System
Research, vol. 66, pp. 171-177., 2003
41. Liu K “Identification of linear time-varying systems” J Sound Vib, vol.206, no.4,
pp.487-505,1997
42. Sp Spiridonakos MD, Fassois SD. “Parametric identification of a time-varying
structure based on vector vibration response measurements” Mechan SystSignal Process,
vol. 23, no.6, pp.2029–2048. 2009
43. Vladimir. V. Terzija “Improved Recursive Newton-Type Algorithm for Frequency and
Spectra Estimation in Power systems,” IEEE Trans. On Instrumentation and
Measurement, vol.52, no.5, pp.1654-1659, 2003
44. Girgis AA, Peterson WL “Adaptive estimation of power frequency deviation andits
rate of change for calculating sudden power system overloads” IEEE Trans.Power Del,
vol.5, pp.585-594, 1990
45. Goursat M, Dohler M, Mevel L, Andersen P “ Crystal clear SSI for operational modal
analysis of aerospace vehicles” In: Structural dynamics, proceedings of the 28th
international modal analysis conference (IMAC), vol. 3. pp. 1421–30, 2010
46. R. Chudamani, K. Vasudevan, and C. S. Ramalingam, "Real-Time Estimation of
Power System Frequency Using Nonlinear Least Squares," IEEE Trans. on Power Del,
vol. 24, no. 3, pp. 1021-1028, 2009.
47. P. K. Dash, S. Hasan, and B. K. Panigrahi, "Adaptive complex unscented Kalman filter
for frequency estimation of time-varying signals," IET Science, Measurement &
Technology, vol. 4, no. 2, pp. 93-103, 2010.
48. C. Gherasim, J. Van den Keybus, J. Driesen, and R. Belmans, "DSP implementation of
power measurements according to the IEEE trial-use standard 1459," IEEE Trans. on
Instrumentation andMeasurement, vol. 53, no. 4, pp. 1086-1092, 2004.

49. R.Trapero, H.Sira-Ramirez and V.Feliu Batlle “On the algebraic identification of
thefrequencies, amplitudes and phases of two sinusoidal signals from their noisy sum,”
International Journal of Control, vol.81, No.3, pp.507-518, 2008.
50. Math H.J and Irene Yu-HuaGu “Signal processing of power quality Disturbances,”
A.JohnWiely&Sons, 2006.
51. SachinK.Jain and S.N.Singh “Harmonics Estimation in Emerging power system: Key
Issues and Challenges,” Electric power syste.Res, vol.81, pp.1754-1766, 2011

145
52. A.A. Girgis and F.M.Ham “A qualitative study of pitfalls in FFT,’ IEEE
Trans.Aerosp.Electron.Sys, vol.16, no.5, PP.434-439, 1980
53. T.A.George and D.Bones, “Harmonics Power Flow Determination Using Fast Fourier
Transform,” IEEE Trans. Power Del, vol.6, no.2, pp.530-535, 1991
54. H. C. Lin and C. S. Lee “Enhanced FFT-based Parameter Algorithm for Simultaneous
Multiple Harmonics Analysis,” in Proc. Inst. Elect. Eng., Gen. Transm. Distri, vol. 148,
pp. 209–214, 2001
55. S.Winograd “On Computing The Discrete Fourier Transform” Proceeding Natl.
Acad.Sci. U.S.A, vol.73, pp.1005-1006, 1976
56. Cichocki. A. and Lobos.T “Artificial neural network for real time estimation of basic
waveforms of voltage and currents” IEEE Trans. power system, vol.9, no.2, pp.612-618,
1994
57. Bakamidis.S, Dendrinos. M and Carayannis G “SVD Analysis by synthesis of
harmonics signals,” IEEE Trans. Signal process, vol.39, no.2 pp.472-477. 1991
58. Karen Kennedy, Gordon Lightbody, Robert Yacamini, “Power System Harmonic
Analysis Using the Kalman Filter” IEEE Power Engineering Society General Meeting,
Vol.2, pp.752-757, 2003.
59. Mustapha Sara, Kamal Djazia, Abdelmadjid Chaoui and Fatech Krim “ Three-phase
active power filter with Integrator-Proptional Cntrol,” Journal of Electrical System, no.1,
pp.79-84,
60. Haili Ma, Adly A. Girgis “Identification and Tracking of Harmonic Sources in a Power
System Using a Kalman Filter” IEEE Trans. on Power Deli, vol.11, no.3, pp.1659-1665,
1996
61. P. K. Dash, A. K. Pradhan , G. Panda, R. K. Jena, S. K. Panda “On Line Tracking of
Time Varying Harmonics Using an Integrated Complex Kalman Filter and Fourier Linear
Combiner,” Proc. IEEE Conference on Power Engineering Society, Singapore, vol.3,
pp.1575-1580, 2000
62. J.A.Rosendo Macias and A.Gomezexposito “ A Comparative between Kalman filter
and STDFT for Harmonics Estimation in Power systems,” Proc.of the 5th
WSEAS/IASME Int.Conf on electric power systems,Highvoltages,
Electricmachines,Tecenfie,Spain, pp.574-578, 2005
63. MaamarBettayeb and UvaisQidwai “Recursive Estimation of Power System
Harmonics,” Electric Power System Research, vol.47, pp. 143-152, 1998

146
64. S.Osowski “SVD Technique for Estimation of Harmonics Components in a Power
System: A Statisticial Approach” IEE Proc-Gener. Trans. Distrib, vol.141, no.5, pp.473-
479, 1994
65. T.Lobos, T.Kozina and H.J.Koglin “Power system harmonics estimation using linear
least squares method and SVD,” IEE Proc. Generation, Transmission & Distributions,
Vol.148, no.6, pp.567-572, 2001.
66. Huaiwei Liao “Power system harmonic state estimation and Observality Analysis via
Sparsity Maximization, IEEE Trans. on Powersystem,vol.22, no.1,pp.15-23,2007
67. Ahmet S. Yilmatz, Ahmed Alkan, Musa H. Asyali “Application of parametric spectral
estimation methods on detection of power system harmonics” Electric Power System
Research, vol-78, pp. 683-693, 2008.
68. Jan Mandel “Efficient Implementation of the Ensemble Kalman Filter,” Center for
Computational Mathematics report, University of Colorado at Denver and
HealthSciences Center, Denver, pp.1-7, 2006.
69. M. Joorabian, S.S. Mortazavi, A.A. Khayyami “ Harmonics estimation in a power
system using a novel-hybrid Least Square –Adaline algorithm” Electric Power System
Research, vol.79, no.1, pp. 107-116, 2009
70. Ray, P.K. and Panda, G “ Harmonics estimation using KF-Adaline algorithm and
elimination using hybrid active power filter in distorted power system signals,”
International Journal of Modeling, Identification and Control, vol. 16, no. 2, pp.149–158,
2012
71. Tadeus Lobos, AndrzejCichocki, PawelKostyla, ZbigniewWaclawek “Adaptive On-
Line Learning Algorithm for Robust Estimation of Parameters of Noisy Sinusoidal
Signals,” Artificial Neural Networks-ICANN, Springer Berlin/Heidelberg, vol.1327/1997,
pp.1193-1198,
72. S. Ghodratollaht, M.Razzaz, M.Moghaddasian and M. Monadi“ Harmonics Estimation
in Power System Using Adaptive Perceptions based on a Genetic Algorithm,” WSEAS
Trans. on Power Systems, vol.2, no.11, 2007
73. M.Bettayeb and U. Qidwai “A Hybrid Least Squares-GA-Based Algorithm for
Harmonic Estimation,” IEEE Transactions on Power Del, vol.18, no.2, pp.377-382, 2003
74. S.Mishra “A Hybrid Least Square-Fuzzy Bacterial Foraging Strategy For Harmonic
Estimation,” IEEE Transactions on Evolutionary Computation, vol.9, no.1, pp.61-73,
2005

147
75. De Arruda, N.Kagan and P. Ribeiro “Harmonics Distortion State Estimation Using an
Evolutionary Strategy,” IEEE Transactions on. Power Del, vol.25, no.2, pp.831-842,
2010
76. Aziz Boukadoum and Tahar Bahi “ Fuzzy Logic Controlled Shunt Active Power Filter
for harmonics Compensation and power quality improvement,” Journal of Engineering
Science and Technology Review, vol.7, no.4, pp. 143-149, 2014
77. V. Ravikumar Padhi and B.k.Panigrahi “Comparative Study of Evolutionary
Computing Methods for Parameter Estimation of Power Quality Signal,” International
Journal of Applied Evolutionary Computation, vol.1, no.2, pp.28-59, 2010
78. S.A. Soliman , R.A. Alammari , M.E. El-Hawary ‘Frequency and Harmonics
Evaluation in Power Network Using Fuzzy Regression Technique,” Electric Power
System Research, vol. 66, pp. 171-177, 2003

79. Fujita H. Akagi H “A Practical approach to harmonic Compensation in power systems-

Series connection of passive and active filters,” IEEE Trans. On Ind Appl, vol.27,
no.6,pp.1020-1025, 1991

80. Girgis. A, Chang, W.B and Makram.E.B “A digital recursive measurement scheme for
on-line trackingof power system harmonic,”IEEE Trans.power Del, vol.6, no.3,
p p.1153-1160, 1998

81. Mori.H,Uematsu.Hand Tsazuki.S “An Artificial Neural Network based method for
predicting power system voltage harmonics,” IEEE Trans.Power Del, vol.7, no.1,
pp.402-409, 1992

82. Benventiste A. “Design of adaptive algorithms for the tracking of time varying
systems,” Int J Adaptive Control Signal Process, vol.1,pp.31–39, 1987

83. Cai Tao, DuanShanxu, Ren Ting and Liu Fangruli “ A robust parametric method
forpower harmonics estimation based on M-estimators,” Measurement, vol.43, no.1,pp.
67-77, 2010

84. J. C. Das, “Passive Filters – Potentialities andLimitations,” IEEE Trans. on Industry

Applications ,vol. 40, no. 1, pp. 232-241, 2004

148
85. K. K. C. Yu, N. R. Watson, and J. Arrillaga, "An adaptive Kalman filter for dynamic
harmonic state estimation and harmonic injection tracking," IEEE Transactions on Power
Del, vol. 20, no. 2, pp. 1577-1584, 2005.

86. D. J. Simon, Optimal State Estimation: Kalman, “H Infinity, and Nonlinear

approaches” Hoboken, N.J: Wiley-Interscience, 2006.

87. V. A. Bavdekar, A. P. Deshpande, S. C. Patwardhan “Identification of process and

measurement noise covariance for state and parameter estimation using Extended
Kalman filter,” Journal of Process Control, vol.21, pp. 585-601, 2011

88. Julio Barros and Enrique Pérez “Automatic Detection and Analysis of Voltage Events
in Power Systems,” IEEE Transactions on Instrumentation and Measurement, vol. 55,
no. 5, pp. 1487-1493, 2006

89. T. P. Tsao, R. C. Wu, and C. C. Ning “The optimization of spectral analysis for signal
harmonics,” IEEE Trans. Power Del, vol. 16, pp. 149–153, 2001

90. P.K.Ray and B. Subudhi “ Ensemble Kalman Filter based Power System Harmonics
Estimation,” IEEE Trans. on Instrumentation and Measurement, vol. 61, no. 12, pp.
3216-3224, 2012

91. Dong Hwa Kim, Ajith Abraham, Jae Hoon Cho “A hybrid genetic algorithm and
bacterial foraging approach for global optimization,” Information Science, vol.177,
pp.3918-3937, 2007

92. K. Mayyas K. and T. Aboulnasr, “Leaky LMS Algorithm: MSE Analysis for Gaussian
Data,” IEEE Trans. Signal Processing, vol. 45, no.4, pp. 927-934, 1997

93. M. Kemenetsky and B. Widrow “A Variable Leaky LMS Adaptive Algorithm,”

Proceedings, 38th IEEE Asilomar Conference on Signals, Systems and Computers, pp.
125-128,.2004

94. T. Hoya, Y. Loke, J. A. Chambers and P.A. Naylor “Application of Leaky Extended
LMS (XLMS) Algorithm in Stereophonic Acoustic Echo Cancellation,” Signal
Processing, vol. 64, no. 1, pp. 87-91, 1998

149
95. A.Klepka and T. Uhl, “Identification of modal parameters of non-stationary systems
with the use of wavelet based adaptive filtering,” Mechanical Systems and Signal
Processing, vol. 47, no. 1, pp. 21-37, 2014

96. M.D. Spiridonakos and S.D. Fassois, “Parametric identification of a time-varying

structure based on vector vibration response measurements,” Mechanical Systems and
Signal Processing, vol. 23, no. 6, pp. 2029-2048, 2009

97. Hirofumi Akagi, “New Trends in Active filters for power conditioning,” IEEE Trans.
On Industry Application, vol.32, no.6, pp. 1312-1322, 1996

98. H.Fujita and H.Akagi “The Unified Power Quality Conditioner: the Integration of
Series and Shunt Active Filters,” IEEE Trans. Power Electronics, vol.13, no. 2, pp. 315-
322, 1998

99. B Singh, K.A1-Haddad, A.Chandra “Performance Comparison of Two Current

Control Technique Applied to an Active Filter,” Eigth International Conference on
Harmonics and quality of Power, vol.1, pp. 133-138, 1998

100. B Singh, K.A1-Haddad, A. Chandra “A Review of Active Filters for Power Quality
Improvement,” IEEE Trans. Ind. Electron, vol.46, pp. 960-971, 1999

101. Salem Rahmania,∗, Kamal Al-Haddad a, Hadi Youssef Kanaan b1 “A comparative

study of shunt hybrid and shunt active power filters for single-phase
applications:Simulation and experimental validation,” Mathematics and Computers in
Simulation, vol.71, pp.345–359, 2006

102. Ambrish Chandra, Bhim Singh, B. N. Singh and Kamal Al-Haddad, “An Improved
Control Algorithm of Shunt Active Filter for Voltage Regulation, Harmonic Elimination,
Power-Factor Correction, and Balancing of Nonlinear Loads,” IEEE Trans. On Power
Electronics, vol.15, no. 3, pp.495-507, 2000

103. D. A. Torrey and A. M. A. M. Al-Zamel “Single-Phase Active Power Filters for

Multiple Nonlinear Loads,” IEEE Trans. Power Electron, vol. 10, no. 3, pp. 263–272,
1995

150
104. J. Miret, M. Castilla, J. Matas, J. M. Guerrero, Luis Garcia de Vicuna “Design of an
Analog Quasi-Steady-State Nonlinear Current Mode Controller for Single Phase Active
Power Filter,” IEEE Trans. Ind. Electron, vol. 56, no. 12, pp. 4872-4881, 2009

105. H. Komurcugil and O. Kukrer “A New Control Strategy for Single-Phase Shunt Active
Power Filters Using a Lyapunov Function,” IEEE Trans. Ind. Electron, vol. 53, no. 1, pp.
305–312, 2006.

106. P. Kumar and A. Mahajan, “Soft computing techniques for the control of active power
filter,” IEEE Trans. Power Delivery, vol. 24, no. 1, pp. 452–461, 2009

107. J. Dixon, J. Contardo, L. Moran “DC link fuzzy control for an active power filter,
sensing the line current only,” in Proc. IEEE Power Eng. Soc. Com, pp. 1109–1113,
1997.

108. L. H. Tey, P.L.So and Y. C. Chu “Improvement of power quality using adaptive shunt
active power filters,” IEEE Trans. on power Del, vol.20, pp. 1558-1568, 2006.

109. H. Akagi, Y. Kanazaya, and A. Nabae “Instantaneous reactive power compensators

compromising switching devices without energy storage component,” IEEE Trans.
IndustryApplication, vol. 20, no. 3, pp. 625–630, 1984

110. J. Matas, Luis Garcia de Vicuna, J. Miret, J. M. Guerrero, M. Castilla, “Feedback

linearization of single phase active power filter via sliding mode control,” IEEE Trans.
power. Electron, vol. 23, no. 1, pp. 116-125, Jan 2008

111. M. Kale, E. Ozdemir, “ An adaptive hysteresis band current controller for shunt active
power filter” Electric Power System Research, vol. 73, no.2, pp. 113–119, 2005.

112. M. Angulo, D.A. Ruiz-Caballero, J.Lago, M. L. Heldwein, S.A. Mussa “Active power
filter control strategy with implicit closed-loop current control and resonant controller,”
IEEE Trans. Ind. Electron, vol.60, no. 7, pp. 2721-2730, 2013.

113. Wu CJ, Chiang JC, Yen SS, Liao CJ, Yang JS, Guo TY. Investigation and Mitigation of
harmonics amplification problems caused by single-tuned filter, IEEE Trans Power Del
1998:800-6.

151
114. Abbas Ketabi, Mohammad Farshadnia,MajidMalekpour, ReneFeuilet “ A new control
strategy for active power line conditioner (APLC) using adaptive notch filter,” Electrical
Power and energy systems,vol.47, pp31-40, 2013

115. Sushreesangita Patnaik and Anup Kumar Panda “Optimizing current harmonics
compensation in three-phase power systems with an Enchanced Bacterial Foraging
approach,” Electrical Power and Energy system, vol. 61, pp. 386-398, 2014

116. R. Pavlanin1, P. Spanik1 and B. Dobrucky “ Comparison of Multi-Resonant- and

Hysteresis Band Controllers used in Current Control Loop of Shunt Active Power Filter,”
International Conference on Renewable Energies and Power Quality (ICREPQ’12)
Santiago de Compostela (Spain),pp.1-5 2012

117. N. Gupta1*, S. P. Singh1, S. P. Dubey2 “DSP based adaptive hysteresis-band current

controlled active filter for power quality conditioning under non-sinusoidal supply
voltages,” International Journal of Engineering, Science and Technology vol. 3, no. 4,
pp. 236-252, 2011

118. Abdelhafid Semmah , Ahmed Massoum , Patrice Wira “Improvement of PWM

Rectifier Performances Using Direct Power Control and Fuzzy Logic Theory,”
Australian Journal of Basic and Applied Sciences, vol.7, no.4, pp.297-303, 2013

119. L.Moran, P.Godoy, R.Wallace, and J.Dixon. “ A new current control strategy for active
power filters using thrre PWM Voltage source inverter,” IEEE PESC 93 Seattle,24th
annual IEEE, pp.3-6, 1993

120. R. C. Dugan, M. F. McGranaghan, S. Santoso, and H. W. Beaty “Electrical Power

Systems Quality,” McGraw-Hill, 2002.

121. P . K. Dash, D. P. Swain, A. Routray, A. C. Liew “Harmonic Estimation in a Power

System using Adaptive Perceptrons,” IEE Proceedings of Generation, Transmissions &
Distributions, vol.143, no.6, pp.565-574, 1996

122. Saini, M.K. and Kapoor, R “ Classification of power quality events – a review,” Intl.
Journal of Electric Power and Energy Systems, vol. 43, no. 1, pp.11–19, 2012

152
123. M. Routimo, M. Salo, H. Tuusa “Comparison of voltage-source and current-source
shunt active power filters,” IEEE Trans. Power Electron, vol. 22, no. 2, pp. 636–643,
2007

124. J.L.Dominguez, J.Arguelles, M.. Arrieta, B. Jaurieta, M. Benito and I. Zugazaga“ New
Quick Convergence Invariant Digital Filter for Phasor Estimation,” Electric Power
System Research, vol.79, no.5, pp. 705-713, 2009

125. P K Dash, A C Liew, D P Swain and B Mishra “Fast tracking of transient power
system signals using fuzzy LMS algorithm,” Electrical Power & Energy Systems, vol.
20, no. 8, pp. 555–561, 1998

126. D. A. Gonzalez and J. C. McCall “Design of Filtersto Reduce Harmonic Distortion in

Industrial PowerSystems,” IEEE Trans. on Industry Applications, vol.1, no.23,pp. 504-
512, 1987

127. A. Ludbrook “Harmonic Filters for Notch Reduction,” IEEE Trans. on Industry
Applications, vol. 24, no.5, pp. 947-954, 1988

128. L. Saribulut, A. Teke, M. Tumay “Artificial neural network based discrete fuzzy logic
controlled active power filter,” IET Power Electronics, vol. 7, no. 6, pp. 1536–1546,
2014.

129. S. Bhattacharaya, D. M. Divan, B. Banarji, “Synchronous reference frame harmonic

isolator using series active filter,” Proc. 4th EPE, Florence, vol. 3, pp. 030–035, 1991

130. S. Reyes, P. Salmeron, H. Kim “Instantaneous reactive power theory applied to active
power filter compensation: different approaches, assessment and experimental results,”
IEEE Trans. Ind. Electron, vol. 55, no. 1, pp. 184–196, 2008

131. G. H. Bode, D. G. Holmes “Implementation of three level hysteresis current control

for single phase voltage source inverter,” Power Electron Specialists conference, 2000.
PESC 00. 2000 IEEE 31st Annual vol.1, pp.33-38, 2000.

153
132. Kuo-Kai Shyu, Ming-Ji Yang, Yen-Mo Chen, Yi-Fei Lin, “Model reference adaptive
control design for shunt active power filter system,” IEEE Trans. Ind. Electron., vol. 55,
no. 1, pp. 97–106, 2008.

133. Siew-Chong Tan, Y.M. Lai, C.K. Tse “A unified approach to the design of PWM-base
sliding mode controller for basic DC-DC converters in continuous conduction mode,”
IEEE Trans. circuits and systems, vol.53, no.8, pp. 1816-1827, 2006.

134. A.Abrishamifar, A.A.Ahamad, M, Mohamadian “Fixed switching frequency sliding

mode control for single-phase unipolar inverter,” IEEE Trans. power Electron, vol.27,
no.5, pp. 2507-2514, may 2012.

135. S. Oucheriah, L. Guo “ PWM based adaptive sliding mode control for boost DC-DC
converters,” IEEE Trans. Ind. Electron, vol. 60, no. 8, pp. 3291–3300, 2013

136. F. Betin, D. Pinchon, Gerard-Andre Capolino “A time-varying sliding surface for

robust position control of a DC motor drive,” IEEE Trans. Ind. Electron, vol. 49, no. 2,
pp. 462–472, 2002.

137. E. Bianconi, J. Calvente, R. Giral, E. Mamarelis, G. Petrone, C. A. Ramos-Paja,

M.Vitelli “ A fast current based MPPT technique employing sliding mode control,” IEEE
Trans. Ind. Electron, vol. 60, no. 3, pp. 1168-1178, 2013

138. M. A. Mahmud, H. R. Pota, M. J. Hossain “Dynamic stability of three phase grid

connected photovoltaic system using zero dynamic design approach” IEEE Journal of
Photovoltaics, vol. 2, no. 4, pp. 564–571, 2012

139. Wu CJ, Chiang JC, Yen SS, Liao CJ, Yang JS, Guo TY “Investigation and Mitigation
of harmonics amplification problems caused by single-tuned filter,” IEEE Trans Power
Del . vol.13, no.3, pp.800-806, 1998

140. Smitha Krishnan’ Lathika B.S “Shunt Active Power Filter Based on Fuzzy-Hysteresis
Controller,” for Electric Traction System 10th National Conference on Technological
Trends (NCTT09) , pp.161-166, 2009

154
141. S. W. Sohn, Y. B. Lim, J. J. Yun, H. Choi, and H. D. Bae “A Filter Bank and a Self-
Tuning Adaptive Filter for the Harmonic and Interharmonic Estimation in Power
Signals,” IEEE Trans. Instrum. Meas, vol. 61, no. 1, pp. 64-73, 2012

142. S. K. Jain and S. N. Singh, “Fast Harmonic Estimation of Stationary and Time-Varying
Signals Using EA-AWNN,” IEEE Trans. Instrum. Meas, vol. 62, no. 2, pp. 335-343,
2013

143. S. K. Jain and S. N. Singh, “Exact Model Order ESPRIT Technique for Harmonics and
Interharmonics Estimation,” IEEE Trans. Instrum. Meas, vol. 61, no.7, pp. 1915-1923,
July 2012

144. Ray, P.K. and Panda, G “ Harmonics estimation using KF-Adaline algorithm and
elimination using hybrid active power filter in distorted power system signals,”
International Journal of Modeling, Identification and Control, vol. 16, no. 2, pp.149–
158, 2012

145. P.S.Puhan, P.K.Ray & G.Panda “ Development of real-time implementation 5/5 rule
based fuzzy logic Controller Shunt active Power Filter for Power quality Improvement,
International journal of Emerging Electrical Power System, vol.17.no.6,pp.607-617,2016

146. P.K.Ray, P.S.Puhan & G.Panda “Real time Harmonics Estimation of Distorted Power
System Signal” International journal of Electrical Power and Energy System. Vol.78,
pp.91-98, 2016

147. P.S.Puhan, P.K.Ray & G.Panda “AComparative Analysis of Shunt Active Power filter
and Hybrid Active power filter with Different Control Technique applied forHarmonic
Elimination in a Single phase System” International Journal of Modeling, Identification
and Control, vol.24, no.1, pp.19-18,2015

148. P.K.Ray, P.S.Puhan & G.Panda “Improved Recursive Newton Type Algorithm based
power system frequency estimation” International journal of Electrical Power and
Energy System. vol.65, pp.231-237, 2015

155
149. G.Panda, P.K.Ray & P.S.Puhan “Harmonic estimation of distorted power system
signals employing two hybrid strategies” International Journal of Modeling,
Identification and Control, vol.22, no.1, pp.20-32, 2014

150. G.Panda, P.K.Ray, P.S.Puhan & S.Das “Novel schemes used for estimation of power
system harmonics and their elimination in a three-phase distribution system International
Journal of Electrical Power and Energy Systems” vol.53, pp.842-856, 2013

156