\documentclass[10pt,twocolumn,a4paper]{article}
\usepackage{fullpage}
\usepackage{setspace}
\usepackage{parskip}
\usepackage{titlesec}
\usepackage[section]{placeins}
\usepackage{xcolor}
\usepackage{breakcites}
\usepackage{lineno}
\usepackage{hyphenat}
\renewcommand{\familydefault}{\sfdefault}
\PassOptionsToPackage{hyphens}{url}
\usepackage[colorlinks = true,
linkcolor = blue,
urlcolor = blue,
citecolor = blue,
anchorcolor = blue]{hyperref}
\usepackage{etoolbox}
\makeatletter
\patchcmd\@combinedblfloats{\box\@outputbox}{\unvbox\@outputbox}{}{%
\errmessage{\noexpand\@combinedblfloats could not be patched}%
}%
\makeatother
\renewenvironment{abstract}
{{\bfseries\noindent{\abstractname}\par\nobreak}\footnotesize}
{\bigskip}
\titlespacing{\section}{0pt}{*3}{*1}
\titlespacing{\subsection}{0pt}{*2}{*0.5}
\titlespacing{\subsubsection}{0pt}{*1.5}{0pt}
\usepackage{graphicx}
\usepackage[space]{grffile}
\usepackage{latexsym}
\usepackage{textcomp}
\usepackage{longtable}
\usepackage{tabulary}
\usepackage{booktabs,array,multirow}
\usepackage{amsfonts,amsmath,amssymb}
\providecommand\citet{\cite}
\providecommand\citep{\cite}
\providecommand\citealt{\cite}
% You can conditionalize code for latexml or normal latex using this.
\newif\iflatexml\latexmlfalse
\providecommand{\tightlist}{\setlength{\itemsep}{0pt}\setlength{\parskip}{0pt}}%
\AtBeginDocument{\DeclareGraphicsExtensions{.pdf,.PDF,.eps,.EPS,.png,.PNG,.tif,.TIF,.jpg,.JPG,.jpeg,.JPEG}}
\usepackage[utf8]{inputenc}
\usepackage[greek,english]{babel}
\usepackage{authblk}
\usepackage{marvosym}
\renewcommand{\floatpagefraction}{.9}
\author[1,\Letter]{Annu Govind}
\author[1]{Anupama Prakash}
\author[1]{Prakash Kumar}
\affil[1]{\small{Amity University Uttar Pradesh}}
\affil[\Letter]{\footnotesize{\href{mailto:annugovind@gmail.com}{annugovind@gmail.com}}}
\usepackage{lastpage}
\usepackage{fancyhdr}
\usepackage{eso-pic}
\usepackage{tabulary}
\newcommand*{\doi}{}
\newcommand*{\firstpage}{78}
\newcommand*{\volume}{101}
\renewcommand*{\year}{2021}
\newcommand*{\issue}{1}
\newcommand{\journal}{Journal of Power Technologies\space \volume \space (\issue) \space (\year) \space \firstpage--\pageref{LastPage} }
\fancypagestyle{plain}{%
\headsep 20pt
\fancyfoot{}%
\fancyhead{}
\fancyhead[L]{\textsf{\journal{}}}
}
\setcounter{page}{\firstpage}
\pagestyle{fancy}
%\fancyfoot[C]{\vspace{1ex} ---\space\thepage\space---}
\fancyfoot{}
\fancyhead{}
\fancyhead[L]{\textsf{\journal{}}}
\fancyfoot[R]{\textsf{\thepage\space $|$\space \pageref{LastPage}}}
\renewcommand{\floatpagefraction}{1}
\headsep 10pt
\newcommand\JPTlogo{%
\put(351,755){%
\setlength{\unitlength}{0.3cm}
\begin{picture}(20,9)
\color{blue!75}
\thicklines
\put(20, 9){\line(-1,0){7}}
\put(13, 9){\line(0,-1){1.65}}
\put(13, 7.35){\line(1,0){7}}
\put(20, 7.35){\line(0,1){1.65}}
\put(15.675, 6.15){\line(0,-1){5.15}}
\put(15.675, 1){\line(1,0){1.65}}
\put(17.325, 1){\line(0,1){5.15}}
\put(17.325, 6.15){\line(-1,0){1.65}}
\put(9.75, 9){\line(0,-1){1.65}}
\put(9.75, 3.5){\line(0,1){1.65}}
\qbezier(9.75, 9)(12.5, 9)(12.5, 6.25)
\qbezier(12.5, 6.25)(12.5, 3.5)(9.75, 3.5)
\qbezier(9.75, 7.35)(10.85, 7.35)(10.85, 6.25)
\qbezier(10.85, 6.25)(10.85, 5.15)(9.75, 5.15)
\put(8.55, 9){\line(0,-1){8}}
\put(8.55, 1){\line(-1,0){1.65}}
\put(6.9, 1){\line(0,1){8}}
\put(6.9, 9){\line(1,0){1.65}}
\put(5.4, 9){\line(0,-1){5.75}}
\qbezier(5.4, 3.25)(5.4, 1.3)(3.75, 1)
\put(3.75, 1){\line(0,1){8}}
\put(3.75, 9){\line(1,0){1.65}}
\qbezier(0.2, 4.2)(0.3, 1.5)(2.55, 1)
\qbezier(1.85, 4.2)(1.85, 2.85)(2.55, 2.65)
\put(0.2, 4.2){\line(1,0){1.65}}
\put(2.55, 2.65){\line(0,-1){1.65}}
\end{picture}
}
}
\providecommand\AddToShipoutPictureBG{}
\AddToShipoutPictureBG{\JPTlogo}
\begin{document}
\title{Performance Enhancement of Shunt Active Power Filter Application using
Adaptive Neural Network}
\vspace{-1em}
\date{}
\begingroup
\let\center\flushleft
\let\endcenter\endflushleft
\maketitle
\endgroup
\selectlanguage{english}
\begin{abstract}
Adaptive neural network (ANN) topology-based control is proposed in this
paper for three phase three wire shunt active power filter (SAPF)
application. The proposed controller improves power quality and
compensates harmonic components. The system includes a current
controlled voltage source inverter (CC-VSI) using three phase insulated
gate bipolar transistors (IGBT), a DSP module for generating regulated
pulse width modulated (PWM) pulse and reference DC bus. The increase in
nonlinear load applications has raised power quality issues. SAPF has
emerged as one of the best solutions to improve power quality.
Application of ANN in SAPF eliminates the need for unit template
generation and the tuning requirement of phase locked loop (PLL), as
required in traditional SAPF. The proposed ANN based SAPF can be
dynamically regulated for minimum harmonic contamination. The results
were obtained and verified in Matlab/ Simulink platform.%
\end{abstract}%
\sloppy
\textbf{Keyword:} DSP controller; harmonic minimization; neural network;
power quality; shunt active power filter\textbf{}
\section{Introduction}
{\label{667956}}
In recent decades power quality issues have been raised due to the
increasing application of power electronic controllers, which include
AC/DC/AC conversion and digital computational equipments used in
interconnected electrical power networks~\cite{Gupta_2012,Na_2018,Kumar_2018}. These
applications have nonlinear characteristics and cause harmonic pollution
in power transmission/ distribution networks~\cite{Huangfu_2018,Palwalia_2008}. ~
The power-electronics converter \& nonlinear loads are the main source
of harmonics \& reactive power, affectng power system
performance~\cite{Lin_2019,Kumar_2016,Singh_2018,Jayaswal2020}. Voltage harmonics and related power
distribution problems arise due to the current harmonics produced by
nonlinear loads such as controlled/uncontrolled converter, electronic
power supplies, lighting ballasts, arc furnaces, adjustable speed
drives, electric oven and uninterruptible power supplies
(UPS)~\cite{Ji_2018,Kumar_2015a,Singh_1999}. Distorted voltage due to harmonics affects
power quality for both power system operators and the consumers
connected at the point of common coupling (PCC)~\cite{Wang_2018,Palwalia_2010}.~
Passive and active filters are used extensively for improving the
quality of power by overcoming current/voltage harmonics \& compensating
reactive power~\cite{Palwalia_2011,Agrawal_2016,Kumar_2015}. Usually, passive filtering has been
preferred for harmonic compensation in the electrical power system due
to its low cost, simplicity, reliability and efficient operation, but it
has many disadvantages such as bulk, unsuitability for changing system
conditions, source impedance strongly influencing the filtering
characteristics and possibilities of anti-resonance with network
impedance resulting in harmonic amplification~\cite{Kumar_Sharma_2015,Chebabhi2018SelfTF,Subudhi2014ACA,palwalia2015}. These
drawbacks of passive filters and increasing power quality concerns need
to be overcome by enhanced approaches towards mathematical analysis,
robust design and ease of implementation of APFs~\cite{Nie_2018,Wang_2018a}. In
series APFs voltage compensation signals are produced, whereas in shunts
APFs current signal are obtained. APFs have better control and faster
response than conventional passive filters. This paper deals with shunt
APF control for minimization of the harmonic components involved. ~
\section{System Configuration}
{\label{351487}}
The proposed shunt APF consists of a DSP control unit for processing
voltage error signal and neural network logic for processing output
parameters from the source. Fig.~{\ref{792840}} shows
the different components of the proposed system configuration for
reactive power compensation and eliminating harmonics current
components. The 3-phase power source was connected to a 3-phase
uncontrolled converter having inductive-resistive~\emph{(RL)} /
capacitive-resistive~\emph{(RC)} loads. The CC-VSI connected to a
self-supported DC busbar
(\emph{L}\textsubscript{\emph{C}},~\emph{R}\textsubscript{\emph{C}})
consists of standard self commutating IGBT switches. Series inductors
(\emph{L}\textsubscript{\emph{C}},~\emph{R}\textsubscript{\emph{C}})
were used to avoid high~\emph{di/dt}~(Akagi et al., 2007). ~
\par\null\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/1/1}
\caption{{DSP-based three-phase SAPF
{\label{792840}}%
}}
\end{center}
\end{figure}
To eliminate switching spikes, three smoothing inductors
(\emph{L}\textsubscript{\emph{sm}}, \emph{R}\textsubscript{\emph{sm}})
were coupled with a series nonlinear load. SAPF compensate the harmonic
components \& maintain the supply current sinusoidal by obtaining
harmonic current of similar value and opposite phase.The capacitor
connected on the dc side provides minor ripple in steady-state dc
voltage. This capacitor stores energy and provides necessary real power
during the transient period.
\section{Control Strategy}
{\label{857211}}
The control technique used for reference current generation is simple,
robust and has fast dynamic response under fixed and variable load
conditions. It should also perform its best under non sinusoidal supply
voltage condition. To attain it, the proposed control stratagem was
divided into two steps. These two steps include estimation of reference
source current using instantaneous p-q theory and fundamental signal
extraction using the ANN algorithm.
\subsection{Instantaneous p-q Theory}
{\label{272941}}
Fig.~{\ref{644013}} shows a basic block diagram using
the p-q theory control algorithm. It was designed for calculation of the
reference source current.
~\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/2/2}
\caption{{Schematic representation of reference current signal generation
{\label{644013}}%
}}
\end{center}
\end{figure}
s per p-q theory, instantaneous reactive power compensator encompassing
devices like switching devices -- which are passive components and do
not need energy storage units. It can be used for compensation of a
fundamental imaginary power component from harmonic current involved in
instantaneous reactive load power under both steady state and transient
conditions. ~Clarke Transformation was used for p-q theory. 3-phase
voltage source voltage (\emph{v}\textsubscript{\emph{sa}
~},~\emph{v}\textsubscript{\emph{sb}}
,~\emph{v}\textsubscript{\emph{sc}}) and load currents
(\emph{i}\textsubscript{\emph{La}~},~\emph{i}\textsubscript{\emph{Lb}}
,~\emph{i}\textsubscript{\emph{Lc}}) were changed into an \selectlanguage{greek}α-β-\selectlanguage{english}0
reference frame and calculated as {[}refer Equation (1){]}
\[\begin{equation}
\begin{aligned}
\begin{bmatrix}
v_\alpha \\ v_\beta \\ v_0\\
\end{bmatrix}= \begin{bmatrix}
C
\end{bmatrix}
\begin{bmatrix}
v_{sa} \\ v_{sb} \\ v_{sc}\\
\end{bmatrix}
and
\begin{bmatrix}
i_{L\alpha} \\ i_{L\beta} \\ i_{L0}\\
\end{bmatrix}= \begin{bmatrix}
C
\end{bmatrix}
\begin{bmatrix}
i_{La} \\ i_{Lb} \\ i_{Lc}\\
\end{bmatrix}\\
\\
where
\begin{bmatrix}
C
\end{bmatrix}=\sqrt{2/3}
\begin{bmatrix}
1 -\dfrac{1}{2} -\dfrac{1}{2}\\
0 \dfrac{\sqrt{3}}{2} -\dfrac{\sqrt{3}}{2}\\
\dfrac{1}{\sqrt{2}} \dfrac{1}{\sqrt{2}} \dfrac{1}{\sqrt{2}}\\
\end{bmatrix}
\end{aligned}
\end{equation}\]~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~
For balanced voltage condition, zero-sequence voltage is absent and
current is available. In order to represent zero-sequence current and
voltage components,~\emph{p} and~\emph{q} can be given as {[}refer Eq.
(2){]} ~ ~
~ ~ ~ ~~\[\begin{equation}
\begin{aligned}
\begin{bmatrix}
p \\ q
\end{bmatrix}
= \begin{bmatrix}
v_\alpha v_\beta \\ -v_\beta v_\alpha
\end{bmatrix}
\begin{bmatrix}
i_{L\alpha} \\ i_{L\beta}
\end{bmatrix}
\\
p = v_\alpha . i_{L\alpha} + v_\beta . i_{L\beta}
\\
q = v_\alpha . i_{L\beta} - v_\beta . i_{L\alpha}
\end{aligned}
\end{equation}
\]~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~
Thus, the instantaneous real and reactive power (q) can be calculated.
Net power can be expressed as~\emph{P}\textsubscript{\emph{net}}
represented as the sum of real and reactive power. This real and
reactive power can further split in terms of DC (average) and AC
(oscillating) components, given as {[}refer Eq. (3){]} ~
\[\begin{equation}
\begin{aligned}
p_{net} = p + q = \overline{p} + \tilde{p} + \overline{q} + \tilde{q}
\end{aligned}
\end{equation}
\]
Here,~\(\overline{p}\) and~\(\overline{q}\)~and shows DC component
of real and reactive power respectively.~\(\tilde{p}\)
and~\(\tilde{q}\) are oscillating values of active and reactive
power. These oscillations of power occur between load and source without
contributing to system energy. The \selectlanguage{greek}α-β \selectlanguage{english}factors of load current~ are
given as {[}refer Eq. (4){]}~~
\[\begin{equation}
\begin{aligned}
\begin{bmatrix}
i_{L\alpha} \\ i_{L\beta}
\end{bmatrix}
= \dfrac{1}{v^2_\alpha + v^2_\beta}
\begin{bmatrix}
v_\alpha -v_\beta \\ v_\beta v_\alpha
\end{bmatrix}
\begin{bmatrix}
p \\ q
\end{bmatrix}
\end{aligned}
\end{equation}
\]
Load current \selectlanguage{greek}α-β \selectlanguage{english}component can be expressed on the basis of the average
component and an instantaneous oscillating component of the real and
imaginary power of the load.
\[\begin{equation}
\begin{aligned}
i_{L\alpha} = \dfrac {v_\alpha}{v^2_\alpha + v^2_\beta}.\overline{p} + \dfrac {v_\alpha}{v^2_\alpha + v^2_\beta}.\tilde{p} + -\dfrac {v_\beta}{v^2_\alpha + v^2_\beta}.\overline{q} + -\dfrac {v_\beta}{v^2_\alpha + v^2_\beta}.\tilde{q}\\
\\
\\
i_{L\beta} = \dfrac {v_\beta}{v^2_\alpha + v^2_\beta}.\overline{p} + \dfrac {v_\beta}{v^2_\alpha + v^2_\beta}.\tilde{p} + -\dfrac {v_\alpha}{v^2_\alpha + v^2_\beta}.\overline{q} + -\dfrac {v_\alpha}{v^2_\alpha + v^2_\beta}.\tilde{q}
\end{aligned}
\end{equation}
\]~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~
Equation (5) gives various components of load currents. To make the
supply current smooth and distortion-free, currents should first flow
from source to load; and compensate the other remaining reactive
(\(\overline{q}\) ) \& oscillating (\(\tilde{p}\)
,~\(\tilde{q}\)~)~ components; thereby improving overall power
quality. Thus, the current control topology can be sub-divided into
direct and indirect current control.~
(a) In the d\emph{irect current control (DCC) topology}, reference
current (i*\textsubscript{ca}~,~ i*\textsubscript{cb}~ ,~
i*\textsubscript{cc}~)~ is compared with filter current (
i\textsubscript{ca}~ , i\textsubscript{cb}~ , i\textsubscript{cc}~).~
(b) In i\emph{ndirect current control (ICC)}, reference source current (
i*\textsubscript{sa}~,~ i*\textsubscript{sb}~ ,~ i*\textsubscript{sc}~)
is compared with source current ( i\textsubscript{sa}~,~
i\textsubscript{sb}~ ,~ i\textsubscript{sc} ).~
These two current controlling techniques differ by their technique for
reducing switching ripples. In the DCC topology, {switching ripples are
more as compared to ICC topology}; as a reference, APF current operates
on feed-forward control and varies rapidly compared to reference supply
current, which operates on feedback control.~
ICC topology has been used in this paper to generate a reference signal.
To obtain average active power, \selectlanguage{greek}α-β \selectlanguage{english}components are calculated from
active power using load current and supply voltage. The zero/ dc power
component is obtained from LPF with a tuned time constant. Now, the
obtained~\emph{V}\textsubscript{\emph{dc}}
and~\emph{V}\textsubscript{\emph{dc,ref}}~are compared and fed to the PI
controller. The PI controller connected to the DC-link PCC compensates
filter losses and regulates bus voltage. The PI controller output is
processed error signal/loss power~\emph{(loss)~\cite{Saidu_2020,Meena_2020,Jayaswal_2018}}.
Therefore, the net active power of source (\emph{p'}) is the sum of zero
components of active load power (\(\overline{p}\) ) and obtained error
signal~\emph{(p}\textsubscript{\emph{loss}}\emph{)~ \emph{)}{[}refer Eq.
(6)-(8){]}} ~
\[\begin{equation}
\begin{aligned}
p' = \overline{p} + p_{loss}
\end{aligned}
\end{equation}
\]
\[\begin{equation}
\begin{aligned}
i^*_{s\alpha} = \dfrac {v_\alpha}{v^2_\alpha + v^2_\beta}.p'
\end{aligned}
\end{equation}
\]
\[\begin{equation}
\begin{aligned}
i^*_{s\beta} = \dfrac {v_\beta}{v^2_\alpha + v^2_\beta}.p'
\end{aligned}
\end{equation}
\]
Equation (9)~gives two-phase to three-phase transformation, using this
three-phase instantaneous reference supply currents (
i*\textsubscript{sa}~,~ i*\textsubscript{sb}~ ,~ i*\textsubscript{sc} )
are to be computed.~~\[\begin{equation}
\begin{aligned}
\begin{bmatrix}
i^*_{sa} \\ i^*_{sb} \\ i^*_{sc}
\end{bmatrix}
=\sqrt{2/3}
\begin{bmatrix}
1 0 \\
-\dfrac{1}{2} \dfrac{\sqrt{3}}{2} \\
-\dfrac{1}{2} -\dfrac{\sqrt{3}}{2}
\end{bmatrix}
\begin{bmatrix}
i^*_{s\alpha} \\ i^*_{s\beta}
\end{bmatrix}
\end{aligned}
\end{equation}
\]
A hysteresis control based PWM converter was used to generate the
desired pulses to obtain the regulated output.
\subsection{Adaptive Neural Network
Architecture}
{\label{173714}}
ANN architecture was used to extract a fundamental component from SAPF
by minimizing harmonic pollution. It can be assured by estimating the
reference current correctly in ideal and real time load applications for
a given source voltage. In the context of increasing dependency on
nonlinear loads in common households as well as industrial applications,
minimization of harmonic pollution has been a prime challenge. When
distorted voltage at PCC involves harmonic contamination in the source
current and is further affected by additional harmonic contamination due
to load current, ANN has been found more suitable than p-q theory in
these cases of imprecise voltage.
This paper deals with a DSP based ANN controller for extraction of
fundamental signal using the ICC scheme shown in Fig.
{\ref{967329}}. Source voltages
(\emph{v}\textsubscript{\emph{sa}} , \emph{v}\textsubscript{\emph{sb}} ,
\emph{v}\textsubscript{\emph{sc}}) are first transformed into a two
phase (\emph{v}\selectlanguage{greek}\textsubscript{\emph{α}} \selectlanguage{english}\&
\emph{v}\selectlanguage{greek}\textsubscript{\emph{β}}\selectlanguage{english}) component. An adaptive linear
(ADALINE) network based ANN was established to obtain primary voltage
components \emph{v}\selectlanguage{greek}\textsubscript{\emph{α,f}}\selectlanguage{english} \&
\emph{v}\selectlanguage{greek}\textsubscript{\emph{β,f}}\selectlanguage{english} from distorted
\emph{v}\selectlanguage{greek}\textsubscript{\emph{α}} \selectlanguage{english}\& \emph{v}\selectlanguage{greek}\textsubscript{\emph{β}}
\selectlanguage{english}components. Primary voltage components were used to calculate the active
power transmitted from source to load in a 3-phase system. The ANN
(ADALINE) used in this paper has two neural layers in feed forward
control topology, having ``n'' inputs and single output. The output
signal depends on training data and pattern; thereby providing a trained
ANN block for generation of regulated source voltage. ~\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/3/3}
\caption{{Reference Current Generation using ANN
{\label{967329}}%
}}
\end{center}
\end{figure}
Voltage input weight factors in the proposed ADALINE network was given
as \emph{V} = (\emph{v}\textsubscript{\emph{1n}},
\emph{v}\textsubscript{\emph{2n}},
\emph{v}\textsubscript{\emph{3n}},\ldots{}\emph{v}\textsubscript{\emph{mn}})\textsuperscript{T}
and \emph{W}= (\emph{w}\textsubscript{\emph{1n}},
\emph{w}\textsubscript{\emph{2n}},
\emph{w}\textsubscript{\emph{3n}},\ldots{}\emph{w}\textsubscript{\emph{mn}})\textsuperscript{T},
respectively. Considering a linear transfer function with
\emph{Y}\textsubscript{\emph{0}} bias, the obtained outcome \emph{Y} in
terms of voltage input and weight for n\textsuperscript{th} function is
calculated as {[}refer Eq.(10){]}~
\[\begin{equation}
\begin{aligned}
Y_{(n)} = W_{(n)}^T . V_{(n)} = Y_0 + w_{1n}v_{1n} + w_{2n}v_{2n}+ w_{3n}v_{3n}+ ....+ w_{mn}v_{mn}\\
= Y_0 + \sum {w_{mn}v_{mn}}
\end{aligned}
\end{equation}
\]
Adaption of the weight was done by~\emph{Widrow-Hoff delta rule}, also
termed as Least mean square (LMS) in the presented ADALINE network. It
minimizes the mean square error between the expected and obtained
output~\emph{y~}(\emph{n}). It can be represented as~ {[}refer Eq.
(11){]}~~
\[\begin{equation}
\begin{aligned}
w_{(n+1)} = w_{(n)} + \alpha \dfrac{e_{(n)}v_{(n)}}{\lvert v_{(n)} \rvert^2}
\end{aligned}
\end{equation}
\]
Where, \emph{w}(\emph{n}+1) and \emph{w}(\emph{n}) represent the next
and present value of weight vector respectively, \emph{v}(\emph{n}) and
\emph{e}(\emph{n}) show the input (voltage) error signal. The value of \selectlanguage{greek}α
\selectlanguage{english}was calculated by the current decomposition technique.
\section{Simulation and Results ~}
{\label{442816}}
The presented system and ANN were validated on the Matlab/ Simulink
platform. The system under consideration is shown in
Fig.{\ref{903208}}
\par\null\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/4/4}
\caption{{Proposed ANN architecture for APF
{\label{903208}}%
}}
\end{center}
\end{figure}
The factors of source voltage, APF and load have been simulated as
listed in Table~{\ref{tab:my-table}}. Simulated results
have been obtained to verify the working of proposed neural network for
the APF with balanced and nonlinear load conditions. The three phase
source voltage and current used in this paper has been shown in the
Fig.~{\ref{999068}} and
Fig.~{\ref{231627}} respectively. ~\selectlanguage{english}
\begin{table*}[]\center
\caption{{The APF parameters used for simulation
}}
\label{tab:my-table}
\begin{tabular}{|l|l|}
\hline
\multicolumn{1}{|c|}{System parameters} & \multicolumn{1}{c|}{Simulated values} \\ \hline
Supply voltage (Vs) & 230 volts/ph, 50 Hz \\ \hline
Cdc, Vdc & 1.5 * 102 \selectlanguage{greek}μ\selectlanguage{english}F, 612 volt \\ \hline
Coupling inductors (Rc, Lc) & 0.2 \selectlanguage{greek}Ω, \selectlanguage{english}2.75 mH \\ \hline
Source impedance (Rs , Ls) & 0.1 \selectlanguage{greek}Ω , \selectlanguage{english}0.25 mH \\ \hline
Smoothing inductor (Rsm, Lsm) & 0.1 \selectlanguage{greek}Ω, \selectlanguage{english}1 mH \\ \hline
(RL1, LL1) and (RL2, LL2) & (50 \selectlanguage{greek}Ω, \selectlanguage{english}20 mH) , (100 \selectlanguage{greek}Ω, \selectlanguage{english}16 mH) \\ \hline
\end{tabular}
\end{table*}
\par\null\par\null\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--5-Voltage-source-output-voltage-in-volts/Fig--5-Voltage-source-output-voltage-in-volts}
\caption{{Voltage source output voltage in volts
{\label{999068}}%
}}
\end{center}
\end{figure}\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--6-Voltage-source-output-current-in-amps/Fig--6-Voltage-source-output-current-in-amps}
\caption{{Voltage source output current in amps
{\label{231627}}%
}}
\end{center}
\end{figure}
The obtained THD is nominal for source voltage and current as 1.12\% and
3.12\% respectively, as shown in Fig. {\ref{527241}}
and Fig. {\ref{564549}}. Now a load of 7.5 kW has
initially been connected to the source at 0.05 Sec. At this instant when
APF is switched ON, the obtained source current is changed to sinusoidal
as of stepped waveform. Thus, the obtained DC capacitor voltage achieves
steady state voltage magnitude, in contrast to the reference voltage
magnitude, in a few consecutive cycles as obtained in Fig.
{\ref{333508}}. To analyze the system behavior for
variable load condition, load has been changed from 7.5 kW to 10 kW at
t=0.34 s by changing the balanced load and again changed from 10 kW to
7.5 kW at t=0.4 s. ~\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--7-THD-of-voltage-source-output-voltage/Fig--7-THD-of-voltage-source-output-voltage}
\caption{{THD of voltage source output voltage
{\label{527241}}%
}}
\end{center}
\end{figure}
~\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--8-THD-of-voltage-source-output-current/Fig--8-THD-of-voltage-source-output-current}
\caption{{THD of voltage source output current
{\label{564549}}%
}}
\end{center}
\end{figure}\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--9-Voltage-across-the-capacitor-for-balanced-load/Fig--9-Voltage-across-the-capacitor-for-balanced-load}
\caption{{Voltage across the capacitor for balanced load
{\label{333508}}%
}}
\end{center}
\end{figure}
To check the robustness of the proposed system, distorted nonlinear
current and voltage was considered, as shown in
Fig.~{\ref{591632}} and
Fig.~{\ref{475056}} respectively. The obtained
V\textsubscript{DC} is highly distorted and compensating current, as
shown in Fig. {\ref{504465}}, has been used.
\par\null\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--10-Nonlinear-current-due-to-non-linear-load/Fig--10-Nonlinear-current-due-to-non-linear-load}
\caption{{Nonlinear current due to nonlinear load
{\label{591632}}%
}}
\end{center}
\end{figure}\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--11-Distorted-voltage-due-to-non-linear-load/Fig--11-Distorted-voltage-due-to-non-linear-load}
\caption{{Distorted voltage due to nonlinear load
{\label{475056}}%
}}
\end{center}
\end{figure}\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--12-Compensating-current/Fig--12-Compensating-current}
\caption{{Compensating current
{\label{504465}}%
}}
\end{center}
\end{figure}
In order to reduce higher order harmonics, ANN was incorporated to
generate reference voltage signal by extracting fundamental voltage
component from distorted voltage waveform. This reference voltage signal
is converted into alfa and beta components and then passed through a DSP
unit to generate the required PWM signal for obtaining line fundamental
voltage and current components as obtained in Fig.
{\ref{988937}}. The response of the trained neural
network has been seen in less than one cycle time period. Obtained line
voltage and current has been shown in Fig.
{\ref{982220}}. Thus obtained regulated PWM pulses can
be used to obtain desired regulated volage and current at the output
terminals. The obtained output has THD of 4.52\%, as shown in Fig.
{\ref{627821}}.\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--13-Generated-PWM-waveform-from-the-ANN-DSP-module/Fig--13-Generated-PWM-waveform-from-the-ANN-DSP-module}
\caption{{Generated PWM waveform from the ANN-DSP module
{\label{988937}}%
}}
\end{center}
\end{figure}\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--14-Obtained-line-voltage-and-current-waveform/Fig--14-Obtained-line-voltage-and-current-waveform}
\caption{{Obtained line voltage and current waveform
{\label{982220}}%
}}
\end{center}
\end{figure}\selectlanguage{english}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=200]{figures/Fig--15-Obtained-THD-in-output-line-voltage-and-current/Fig--15-Obtained-THD-in-output-line-voltage-and-current}
\caption{{Obtained THD in output line voltage and current
{\label{627821}}%
}}
\end{center}
\end{figure}
\section{CONCLUSION ~}
{\label{762807}}
A neural network-based controller is proposed in this paper for a
three-phase three-wire shunt APF. To validate the compatibility of the
proposed approach, an indirect current control theory-based controller
was developed for variable loads at different switching instant. To
generate the reference source current instantaneous p-q theory with
indirect current control technique is used and the switching pattern of
semiconductor devices used in the PWM converter. The performance of the
proposed APF was evaluated in Matlab/ Simulink platform. The obtained
THD is within the permissible limit of 10\% for both uniform and
nonlinear loading condition. Compared to the direct current control
technique, the indirect current control technique is superior for
eliminating switching ripples from the supply current. The obtained
results reveal satisfactory operation of the system at hand, under
sudden load and frequency changes conditions. ~
\par\null
\textbf{Acknowledgement}
\par\null
I would like to express my deep gratitude to Professor D.K.Palwalia for
his valuable \& enthusiastic encouragement for this research work. I
would also like to thank Dr. V.K.Tayal and Dr. Prakash Kumar, my
research supervisors, for their patient guidance and useful critiques
for this research work. My thanks are also extended to Mr. Kuldeep
Jayaswal for his valuable and constructive suggestions during the
planning and development of this article.
Finally, I wish to thank my husband and my kids for their support and
encouragement throughout my work.
\par\null\par\null
\selectlanguage{english}
\FloatBarrier
\bibliographystyle{unsrt}
\bibliography{bibliography/converted_to_latex.bib%
}
\end{document}