You are on page 1of 12

258 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 3, NO.

2, APRIL 2022

Advanced Control Strategy With Voltage Sag


Classification for Single-Phase Grid-Connected
Photovoltaic System
Mohammed Ali Khan , Student Member, IEEE, Ahteshamul Haque , Senior Member, IEEE,
V. S. Bharath Kurukuru , Student Member, IEEE, and Saad Mekhilef , Senior Member, IEEE

Abstract—This article develops a fault ride-through control RPM Reduced power mode.
strategy for achieving low-voltage ride through in single-phase PLL Phase-locked loop.
grid-connected photovoltaic (PV) systems (GCPVS). The proposed PD Proportional derivative.
control system adapts a neural network classifier for islanding
classification and model predictive control for achieving the control LF Low frequency.
of the two-stage PV system. This control scheme takes advantage VCO Voltage controlled oscillator.
of the nonlinear nature of the power converters and develops a cost PI Proportional integrator.
function-based approach to achieve fast and efficient control. In PR Proportional resonant.
addition, the proposed controller provides voltage support to the
grid during voltage sags by injecting minimum reactive current
within the threshold. The operation of the proposed control strategy I. INTRODUCTION
is verified by performing simulation tests on a 4 kW GCPVS
UE to matured photovoltaic (PV) technology and con-
by creating a sag type of fault in the utility. Further, laboratory
experiments were carried out with the developed controller. The
results ensure that the proposed control system adheres to the grid
D stant drop in installation cost, grid-connected PV systems
(GCPVS) are proliferating. This created challenges related to
requirements by enabling voltage support during grid faults.
voltage control, load shedding, frequency regulation, phase
Index Terms—Fault ride through (FRT), grid codes, grid- mismatch, and others [1]. In the past, distributed generation
connected photovoltaic (PV) system (GCPVS), power electronic (DG) such as PV was allowed to isolate from the grid when
converter, reactive current control.
a fault occurs and reconnect within several seconds once the
fault had been cleared. Nevertheless, due to high PV penetration,
NOMENCLATURE nowadays, the same methods cannot be used because it will
FRT Fault ride through. further deteriorate the power quality and potentially end up in a
LVRT Low-voltage ride through. power blackout.
PV Photovoltaics. Considering this, many grid standards are published by differ-
GCPVS Grid-connected photovoltaic systems. ent international committees, transmission operators, and distri-
NN Neural network. bution operators [2]. These grid codes regulate the interaction
MPC Model predictive control. between the utility and the DG systems by monitoring the
DG Distributed generation. complex aspects such as fault ride through (FRT), islanding
MPPT Maximum power point tracking. detection, antiislanding deenergization [3], and active/reactive
FCS- MPC Finite control set model predictive control. power control. Further, they expect the inverters to act more
actively and smartly with specifications such as grid support,
Manuscript received August 15, 2020; revised October 7, 2020 and November low-voltage ride through (LVRT), and fault detection, while op-
17, 2020; accepted November 21, 2020. Date of publication December 2, 2020; erating in a grid-connected environment [4]. Also, the inverters
date of current version March 25, 2022. (Corresponding author: Mohammed Ali must be capable of injecting some reactive current when there
Khan.)
Mohammed Ali Khan, Ahteshamul Haque, and V. S. Bharath Kurukuru is a requirement for LVRT in the grid.
are with the Advance Power Electronics Research Laboratory, Department To achieve this, various techniques were widely discussed in
of Electrical Engineering, Jamia Millia Islamia (A Central University), the literature. An inverter testbench with back-to-back topology
New Delhi 110025, India (e-mail: mak1791@gmail.com; ahaque@jmi.ac.in;
kvsb272@gmail.com). is developed to assess the LVRT performance of PV inverters
Saad Mekhilef is with the Power Electronics and Renewable Energy Re- using an active power control scheme [5]. The adapted scheme
search Laboratory, Department of Electrical Engineering, University of Malaya, has limitations with transient current overshoots and slow active
Kuala Lumpur 50603, Malaysia, and also with the School of Software and
Electrical Engineering, Faculty of Science, Engineering and Technology, current recovery time. Besides, it is identified that the active
Swinburne University of Technology, Victoria, VIC 3122, Australia (e-mail: power control schemes result in high voltage and current over-
saad@um.edu.my). shoots for transient voltage sags. Further, a reactive current
Color versions of one or more figures in this article are available at https:
//doi.org/10.1109/JESTIE.2020.3041704. injection control scheme with varying positive and negative
Digital Object Identifier 10.1109/JESTIE.2020.3041704 sequences is developed in [6] to provide voltage support for
2687-9735 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ADVANCED CONTROL STRATEGY WITH VOLTAGE SAG CLASSIFICATION FOR SINGLE-PHASE GCPVS 259

GCPVS under voltage sags. It is observed that the implemented predictions of the controlled variables in the GCPVS are cal-
active and variable reactive power scheme has a quick recovery culated using the MPC. Besides, the control command is chosen
time but has drawbacks due to delay in reaction to voltage sag, based on an optimization criterion; thus, the selected control
resulting in oscillations in active and reactive power. In [7], a action guarantees the minimum error of the desired controlled
multimode control strategy is developed to achieve LVRT by variable [19]. The MPC approach provides several advantages
continuously extracting active power from PV arrays through over other methods, such as: 1) it can handle multiple inputs
interleaved boost converters. It is identified that the developed and outputs simultaneously; 2) the selected control command is
approach has a low peak current and has drawbacks due to chosen optimally while considering its future effects; and 3) it
voltage and current oscillations. In [8], the LVRT is achieved can incorporate constraints [20].
for GCPVS by injecting the maximum rated current, which Further sections of the article are organized as follows. Sec-
maximizes the power capability of the inverter during voltage tion II describes the GCPVS and its control. Section III describes
sag. The developed technique has drawbacks due to the slow the proposed approach for controlling the PV system using
rate of active current recovery, reduced active power, and high FCS-MPC. Numerical simulations and experimental analysis
oscillations in reactive power. Further, a smart LVRT controller are presented in Section IV, and Section V provides the con-
is developed by coordinating the operation of MPPT and ride cluding remarks based on the observations made during the
through the controller, and identifying the nature of PV power simulation and experiment.
and current in [9]. The developed controller achieved reactive
power support with active power curtailment, especially for the II. GRID-CONNECTED PHOTOVOLTAIC SYSTEMS
LV grid with a low X/R ratio. Moreover, it is identified that
The two-stage operation of the GCPVS along with the control
enhancing active power injections and reducing active power
layout is illustrated in Fig. 1. The two-stage operation provides
oscillations are highly prioritized instead of providing voltage
an additional advantage with the boost converter operating in
support. In [10] and [11], the LVRT capability of a hybrid
three different modes. These modes were selected based on the
power system is enhanced by reinforcement learning techniques
amount of active power to be injected for three voltage regions:
through convertible static compensators. The connection of ad-
1) the dc/dc converter using incremental conductance MPPT
ditional static compensation devices is considered as a solution
[21] operates only on the grid voltage normal operation region;
for achieving voltage support but incurs additional cost. Also,
2) the reduced power mode (RPM) operates when the voltage
islanding detection is a major challenge while operating with
sag in between 0.9 ≤ Vg ≤ 0.5 p.u., where the PV power is
these controllers. Generally, the islanding detection is performed
tracked to maintain the inverter power operation; and 3) finally,
either using remote techniques or local techniques [12] while
the short-circuit current mode is activated for severe voltage
dealing with classical controllers. Most of the methods cause
dips. The short-circuit current mode receives its name because
active frequency shifts or inject current harmonics [13]. Hence,
the authors instantiate a short-circuit between the PV’s terminals
to overcome these drawbacks, and achieve efficient islanding de-
when the system is required to deliver only reactive power.
tection, various intelligent techniques [14] and machine learning
algorithms [15], [16] are developed.
A. Control of Grid-Connected Inverter
Despite the fact of employing efficient islanding detection
techniques, the LVRT imposes technical challenges, such as For controlling a single-phase grid-connected inverter, the fast
grid current control and power balance between the generator and accurate detection of utility voltage and current information
side and the grid while dealing with classical controllers. All of is important since it is used in generating the reference signals.
these must be solved during the duration of the event to ensure A detailed overview of the three stages of controlling is given
that the grid code requirements are met. Although some authors as follows:
have reported advances in the field, normally they focus on the 1) Grid Synchronization: The phase-locked loop (PLL) tech-
controllability or performance improvement of a specific part niques are investigated for modern synchronous control systems.
of the PV system. As stated in [17], the controllability of dc/ac The implementation of each element and the basic PLL structure
converters to remain connected to the grid and provide voltage is shown in Fig. 2 [22]. The proportional derivative (PD) is
support under fault or disturbances is proven to be one of the implemented with a sinusoidal multiplier and a low-pass filter.
greatest challenges. The low frequency (LF) and voltage-controlled oscillator (VCO)
In light of the above requirements, a finite control set model are separately implemented based on a proportional integrator
predictive control (FCS-MPC), also known as direct MPC, is (PI) controller. The operation of the PD block is effectively an
adopted in this research. This control scheme takes advantage of emulation of the active power calculation process. Zero output
the nonlinear nature of the power converters and offers superior from the PD block denotes a quadrature relationship between
steady-state results and a fast transient response [18]. To achieve the grid voltage v and fictitious current i , and in this case, the
an accurate response of the adapted controller for two important estimated angle θ equals the grid angle θ.
qualities that are needed to accomplish LVRT requirements, From Fig. 2, the fictitious power p calculated as p = v ×
a neural network (NN)-based islanding classifier is employed i = V sin θ cos θ . Further, by applying the product to a sum
along with the controller. The classifier is trained with all trigonometric identity, the fictitious power p can be rewritten as
possible cases of islanding in a single-phase grid-connected V V
environment to identify the state of the system. Further, p = sin (θ − θ ) + sin (θ + θ ) . (1)
2 2

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
260 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 3, NO. 2, APRIL 2022

Fig. 1. Schematic of the control structure for a single-phase two-stage grid-connected photovoltaic system.

2) Current Control Loop: Fig. 3 shows the structure of the


proportional resonant (PR) controller, which consists of a pro-
portional term and a resonant term. The proportional gain Kp
determines the system dynamics in terms of bandwidth and
phase, while the resonant term provides an infinite gain at
frequency ω0 . Conceptually, this is similar to the integrator in
the PI controller whose infinite dc gain yields a zero dc error.
Fig. 2. Basic structure of PLL.
An ideal PR controller is capable of eliminating the steady-state
error or harmonics at the frequency ω0 , while producing no phase
shift or gain at other frequencies.
Based on Fig. 3, the transfer function of the PR controller is
given as
Mi (s) Kr s
GP R (s) = = Kp + 2 . (3)
εi (s) s + ω02
Fig. 3. PR controller structure representation.
However, the realization of the ideal resonant transfer function
cannot be achieved owing to the component tolerance in the
physical circuit and finite precision levels in digital systems.
Considering θ = ωt + ∅, θ = ω  t + ∅ and assuming
Even though it is theoretically possible, the infinite gain would
ω = ω  , for a phase difference of ϕ, the fictitious power can be
potentially lead to system stability issues. Thus, the nonideal
expressed as
form of a PR controller is given as follows, where ωc is the
V V bandwidth around the resonant frequency ω0 :
p = sin (ϕ) + sin (2ω  t + ϕ) . (2)
2 2 Mi (s) 2Kr ωc s
GP R (s) = = Kp + 2 . (4)
From (1) and (2), it can be seen that the information of phase εi (s) s + 2ωc s + ω02
difference is contained in the dc term. Therefore, a low-pass
filter is typically used to filter out the undesired double frequency III. FAULT RIDE THROUGH
harmonics. The remaining PD error is then tuned to zero via the The power profile references (Pg∗ , Q∗g ) shown in Fig. 1 are
LF and VCO, and the phase and frequency information from the controlled to generate the grid current reference during the LVRT
PLL is synchronized to the input signal. operation of the inverter. For a grid voltage sag, these references

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ADVANCED CONTROL STRATEGY WITH VOLTAGE SAG CLASSIFICATION FOR SINGLE-PHASE GCPVS 261

are modified to provide dynamic grid voltage support through the minimum error that is directly applied later. As the generator-
inverters based on different grid integration standards [23]–[25]. side converter has two main control requirements, MPPT during
Further, the LVRT control action is supposed to achieve the normal operation and RPM during voltage sags, the FCS-MPC
regulation of overcurrent protection for the transient voltage requires the model of the power converter to estimate the future
sag operation and discard the momentary disconnection of PV behavior of the system, as well as the integration of the control
system from the grid, provided that the transient voltage sag lies requirement in its strategy. For a boost converter operating in the
within the permissible limits of the grid supporting protocols two-stage grid-connected system, the boost current iLB , when
[26], [27]. In light of the above requirements, initially, the both states of the converter are considered is given by
voltage per unit Vpu levels are calculated at the point of common diLB 1
coupling (PCC). As the voltage sag fault gets detected, the FRT = (VP V + (SBoost − 1) Vdc ) (6)
dt LB
control action computes the required active and reactive power
references based on the voltage sag levels using the following: where Vdc represents the voltage at the dc link, and SBoost
denotes the switching state of the boost converter that can be
⎧  ∗
⎪ Qg = 0 either 1 or 0, which indicates the states ON and OFF, respectively.

⎪ 0.9 < V ≤ 1.0 ⇒
⎪ ∗ LB is the boost inductance. The continuous-time model of
 Pg∗ = Prated
pu


⎨ Qg =  2Prated (1 − Vpu ) boost converter inductance current in (6) can be converted to
0.5 < Vpu ≤ 0.9 ⇒ ∗ 2 (5) discrete-time by approximating the derivative using the forward

⎪  ∗ P = Prated − Q2rated


g
Euler method [28]

⎪ Qg = Prated

Pg∗ = 0 dx x (k + 1) − x (k)
= (7)
dt TsB
where Vpu is the per unit voltage measured at the PCC, is Pg∗
where TsB represents the discretization time sample. By substi-
the active power reference, Q∗g is the reactive power reference,
tuting (7) into (6), the discrete-time model of the current iLB is
Prated is the rated active power, and Qrated is the rated reactive
given by
power.
From (5), it is identified that the generation of active and iLB (k + 1) − iLB (k)
reactive power reference signals is divided into three conditions TsB
based on the identified Vpu [27]. For the first condition, Vpu lies 1
between 0.9 and 1. Here, only the active power is injected into = [VP V (k) + (SBoost (k) − 1) Vdc (k) . (8)
LB
the grid. For the second conditions, the Vpu ranges between 0.5
and 0.9. This condition injects both active power and reactive The predicted current in (k+1), can be simplified to
power with the inverter by considering the fixed apparent power. iLB (k + 1)
Further, in the third condition, Vpu is less than 0.5. Here, the
TsB
reactive power is only injected into the grid with no active power = [VP V (k) + (SBoost (k) − 1) Vdc (k) + iLB (k)] .
injection. LB
(9)
The presence of CP V in the system helps to maintain the PV
A. Control Synthesis for LVRT
voltage constant during operation, and if TsB is small enough,
Conventionally, the techniques described for controlling the it can be assumed that VP V (k) ∼= VP V (k + 1). Therefore, to
inverter and synthesizing the required voltage reference are obtain PP V , both sides of (9) are multiplied by VP V . The
based on various linear control approaches and involve space resulting equation is
vector and pulsewidth modulation methods. These techniques
TsB
needed an appropriate selection of the switching frequencies as PP V (k + 1) = VP V (k) [VP V (k)
they play a major role in sizing the converter. Compared with LB
classical linear control techniques applied for power convert- + (SBoost (k) − 1) Vdc (k)] + PP V (k) .
ers, the FCS-MPC does not require any modulation technique (10)
because the output command is directly applied to the power
The variables expressed in (10) consider only a one-step
electronics. Besides, FCS-MPC maintains the discrete nature of
prediction, and all of them can be obtained by measuring the
the power converters, fast dynamic response, flexibility in the
signals indicated in Fig. 1. Additionally, it does not consider the
definition of the control objectives, and provides various perfor-
power required for the charging/discharging of the capacitor.
mance improvements along with easy inclusion of nonlinearities
The predicted PV power can be modified to include capacitor
by proper selection of the cost function.
dynamics as
In light of the above advantages, the FCS-MPC approach
is analyzed in detail to achieve various LVRT requirements in TsB VP V (k)
PP V (k + 1) = [VP V (k)
GCPVS. The adapted approach is developed to achieve grid LB
current control, dc-link voltage control, grid code compliance, + (SBoost (k) − 1) Vdc (k)]
and fast transient control operation. The FCS-MPC uses the
limited number of the converter switching positions inside of the dVP V
+ PP V (k) − CP V VP V (11)
optimization function to find the converter state that provides the dt
Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
262 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 3, NO. 2, APRIL 2022

where the derivative can be discretized also using Euler methods. On the other hand, in the case of (14), the cost function is
However, the same equation can be rewritten to consider a given by
different prediction horizon h
gB (k) = [sBoost (k + h − 1) − sBooststates ]2 (16)
TsB
PP V (k + h) = VP V (k + h − 1) [VP V (k + h − 1) where sBooststates represents a vector of the states of the boost
LB converter (0 and 1).
+ (SBoost (k + h − 1) − 1) Vdc (k + h − 1)]
B. Grid-Side Converter
ΔVP V
+ PP V (k + h − 1) − CP V VP V (k + h − 1) . The SRF-PLL system provides fast and accurate synchro-
Δth
(12) nization information with a high degree of immunity and in-
sensitivity to various power quality problems, such as power
In general, the models in (9) to (12) correlate the PP V output quality disturbances, harmonics, imbalances, voltage sag, swell,
power to the state of the switching signal given by SBoost , and and notches. The basic principle of this theory is about the
the output voltage of the boost converter. Therefore, one can transformation of current variables in synchronously rotating
use (12) to predict the PV power, and in turn, regulate PP V dq frame. To generate the unit vectors, voltage signals are
accurately. Similarly, (12) can be solved in terms of SBoost as processed by the PLL. As per SRF theory, current signals are
 transformed into dq frame and then filtered. Then, compensating
1 LB current variables are transformed back from dq frame and fed to
SBoost (k+h) = 1 +
Vdc (k+h−1) TsB VP V (k+h−1) current controller which is used for generating switching pulses
× [PP V (k + h) − PP V (k + h − 1) + PCP V ] for inverter switches. This enhances the inverter operation for
dynamic voltage regulation by adapting current control mode.
− VP V (k + h − 1)} (13) Further, it overcomes the drawbacks of intermittent power output
in standalone photovoltaic systems. This results in a constant
where PCP V represents the capacitor charging/discharging
power supply by the PV system. Modeling the inverter shown
power. In the notation given by (13), the predicted PP V has
in Fig. 1 using PLL offers an advantage for the computationally
a different meaning. In this case, PP V (k + h) is no longer a
demanding FCS-MPC scheme. First, the single-phase variables
predicted signal, but the power reference to track. Therefore, to
are transformed from the natural frame to the dq frame. Second,
avoid confusion, (13) is rewritten as
dq variables are naturally dc quantities during line-to-ground

1 LB fault conditions, which in turn allow the use of approximations
SBoost (k+h) = 1 + such as x(k) ∼ = x(k + 1), where x represents a state variable.
Vdc (k+h−1) TsB VP V (k+h−1)
The control signals forming the switching states define the value
× [PP∗ V (k + h) − PP V (k + h − 1) + PCP V ] of the inverter terminal voltages as
−VP V (k + h − 1)} . (14) Vi = sVdc . (17)
Equation (14) shows the optimal state of the boost converter The grid currents can be expressed in terms of the inverter
that tracks, in this case, the PV power reference. voltage, the grid voltage, and the inverter filter in the natural
Cost function: The predictive control scheme utilizes an frame as
optimization function that executes an algorithm in charge of dig 1
= (Vi − Vg − Rg ig ) . (18)
finding a control action that delivers minimum error. Here, (12) dt Lg
and (14) provide two possible options to implement FCS-MPC;
The grid current dynamics in the natural frame can be con-
however, they differ in the cost function definition. When using
verted to the synchronous dq frame and expressed in state-space
the approach shown in (12), the main objective is to regulate the
form as shown in the following:
power delivered by the PV. On the other hand, when using (13),    
the cost function objective is changed to find which of the two d igd igd Vid Vgd
= A + Bi + Bg (19)
possible states is closer to the optimal state, which in turn will dt i gq i gq V iq Vgq
regulate the PV power. When using the model given by (12), the where
cost function is defined as R 1 −1
− Lgg ωg 0 0
gB (k) = [PP V (k + h) − PP∗ V (k + h)] 2 A = ; Bi = Lg
1 ; Bg = Lg
−1
(15) R
−ωg − Lgg 0 Lg 0 Lg

where PP∗ V represents the PV power reference, in which its (20)


h-state can be estimated via an extrapolation technique such as where ωg represents the grid electrical frequency.
Lagrange or vector angle methods [29]. It is important to mention The discrete-time representation for d- and q-axis currents can
that for a small enough sampling period, no extrapolation is be obtained from the continuous-time state-space representation
required if the h-state does not exceed 2 [30]. Moreover, in this in (18) for a one-step prediction, as follows:
   
case, no extrapolation is necessary because the signal reference igd (k+1) igd (k) Vid (k) V (k)
=Φ + Γi + Γg gd
is constant. igq (k+1) igq (k) Viq (k) Vgq (k)

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ADVANCED CONTROL STRATEGY WITH VOLTAGE SAG CLASSIFICATION FOR SINGLE-PHASE GCPVS 263

where FCS-MPC approach were given. In this section, both controllers


Φ =e ATs −1
Γi = A (Φ − I2×2 ) Bi are combined in a single control scheme that allows the LVRT
(21) for PV systems. Equation (24) shows the prediction function
Γg = A−1 (Φ − I2×2 ) Bg
used to obtain the converter terminal voltages based on the in-
where I represents the identity matrix. As it is shown in (20), verter current references, grid variables, and system parameters.
the predicted behavior of the grid-side currents depends on the However, the control strategy utilized here is based on the power
present grid variables and the inverter voltages (Vid (k) and balance of the dc and ac side of the inverter using the dc-link
Viq (k)). voltage control. Therefore, the inverter current references have
Similar to the approach followed for generator-side control, to be calculated based on the active and reactive power required
(20) can be expressed in terms of the control actions, which, in during the fault event
this case, is the inverter voltage. Therefore, (20) can be rewritten   
as P Vgd Vgq igd
    = . (28)
Q Vgq −Vgd igq
Vid (k) i (k + 1) i (k) V (k)
= Ψ gd − Υ1 gd − Υ2 gd
Viq (k) igq (k + 1) igq (k) Vgq (k) The system described in (28) can be utilized to calculate the
(22) inverter active current reference for the next sampling time by
where inverting the matrix
Ψ = Γ−1  ∗   
i
igd (k + 1) 1 Vgd (k) Vgq (k)
Υ1 = Γ−1
i Φ (23) =
Υ2 = Γ−1 i∗gq (k + 1) 2 (k) + V 2 (k)
Vgd Vgq (k) −Vgd (k)
i Γg
gq
 ∗
where the superscript notation (−1) denotes the matrix inverse. Pg (k + 1)
× (29)
The predicted d- and q-axis grid currents in (20) have a different Q∗g (k + 1)
meaning in (22). In this notation, the predicted currents repre-
sent the d- and q-axis reference currents. Therefore, to avoid where Pg∗ (k + 1) is given by the dc-link voltage compensator
confusion, (27) is rewritten as and Q∗g (k + 1) is obtained following the grid code requirement
 ∗   and voltage depth.
Vid (k) igd (k + 1) igd (k) V (k) At the same time, the generator-side controller must operate
= Ψ ∗ − Υ1 − Υ2 gd
Viq (k) igq (k + 1) igq (k) Vgq (k) the PV array at a specific location in the PP V versus VP V curve
(24) to avoid grid overcurrent and eventual disconnection from the
where superscript ∗ indicates a reference signal. grid. Therefore, the PV power reference is calculated based on
Equation (23) shows that the inverter control signal at k the future grid current as
depends on the current state variables and the reference current
to track. Moreover, the matrices Ψ, Υ1 , and Υ2 can be calculated PP∗ V (k+1) = [Vgd (k)igd_T (k + 1) + Vgq (k)igq_T (k + 1)]
beforehand to improve the algorithm execution speed. (30)
Cost function: Similar to the cost function defined previously where igd_T and igq_T are the active/reactive currents obtained
for the generator-side controller, (20) and (24) allow FCS-MPC as per the requirement.
implementation for inverter control that delivers the same result.
When using the model given by (20), the cost function is defined IV. IMPLEMENTATION AND RESULTS
as By considering all the design parameters discussed in the
2 previous section, a 4 kW p GCPVS is designed using MAT-
gg (k) = igd (k + 1) − i∗gd (k + 1)
LAB/Simulink environment. The design parameters and values
2
+ igq (k + 1) − i∗gq (k + 1) (25) of the system are depicted in Table I.
where the reference and the grid current are compared. On the
other hand, if (24) is used instead, the cost function is given by A. Islanding Classifier
the state-based approach as By integrating the PV system with the grid, it is necessary
 2  2 to continuously monitor the system and ensure that the system
gg (k) = Vid (k) − VINVStatesd + Viq (k) − VINVStatesq operates smoothly. If any of the system parameters tend to vary,
(26) it is essential to detect the abnormality and clear the fault in a
where VINVStatesd and VINVStatesq represent all possible d and q certain time limit as specified by different standards [31]. To
voltage vectors delivered by the inverter on dq frame, given by achieve this, a fault detection method using a wavelet transform
 and NN classifier is developed. The detailed process of classifier
VINVStatesd
= [Tdq ] [vi ] . (27) training for islanding classification is discussed in a previous
VINVStatesq
work of the authors [32]. Further, this research identifies seven
different operating conditions corresponding to normal opera-
C. Control Structure on Symmetrical Faults
tion, power and frequency mismatch faults, harmonics, failures
In Sections III-A and III-B, the details for the control of the during grid synchronization, and different symmetrical faults
generator- and grid-side components of the PV system using the on the grid side of the developed system to train the classifier.

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
264 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 3, NO. 2, APRIL 2022

TABLE I
OVERALL SYSTEM DESCRIPTION FOR 4 kWp SOLAR INVERTER

Initially, the characteristics of these faults and operating con-


ditions are measured through the voltage/current measurements
at PCC of the grid-connected system for few cycles. The mea-
Fig. 4. Confusion matrix for the overall process of NN classifier.
sured characteristics are subjected to frequency and time series
representation using wavelet transform and the corresponding (0)
features of the signal are extracted. In this research, the discrete the zeroth element is the bias unit a0 with a value of 1. Further,
wavelet transform with Daubechies 5 mother wavelet is used from the perspective of forward propagation
to extract eight different features of the voltage measurement z (0) = θ(k) a(k−1) (31)
for different operating and fault conditions at the PCC. These
z (k) ∈ R1×(m|θ )
(k)
∈R m×n
features include energy, entropy, peak-to-peak ratio, skewness, (32)
kurtosis, signal-to-noise ratio, harmonics, and power spectral  
density. a(k) = g z (k) (33)
Further, these eight features form a feature matrix of size where g(x) corresponds to the activation function evaluated
8 × 4526 which is used with the NN approach for pattern elementwise, and the dimensions of a(k) and z (k) are equal.
recognition and classifier development. The back-propagation Further, it is observed that each layer has a matrix multi-
algorithm is executed for initializing the classification technique plication and an activation function computation. Each naïve
consisting of eight neurons in the input layer, 12 neurons in the matrix multiplication has a runtime of O(n3 ) and the activation
hidden layer, and seven neurons in output layers. The target function with elementwise multiplication has a runtime of O(n).
classes are defined for all the input data by tabulating the classes Moreover, by analyzing the dimensions of the NN it is observed
as binary data (0000001b to 1000000b). Besides, 70% of the that
feature matrix data is used for training whereas 30% of the (0)
feature matrix data is used for validation and testing purposes. θ(0) ∈ Rn ×1
(34)
To train the network efficiently, a total of 362 iterations were (1) (0)
θ(1) ∈ Rn ×n
(35)
performed. The gradient after six validation checks is observed
to be 0.0040456. The confusion matrix for the trained classifier is θ(2) ∈ Rn
(2)
×n(1)
. (36)
shown in Fig. 4. From the results, it is identified that out of 4526
samples, 4462 samples were trained, validated, and classified This can be generalized as
accurately, whereas 64 samples are incorrectly classified. The  (k)
Rn ×1 if k=0
training accuracy of 98.6% for fault classification with the help θk = (k) (k−1) (37)
of input and target data is achieved. Rn ×n if k>0

where n(k) corresponds to the number of neurons along with the


B. Complexity Calculation bias unit in layer k.
The pattern recognition and classification approach of NN in- Further, from the analytical reasoning of matrix multiplica-
volves a forward propagation which has an asymptotic complex- tion, it is identified that
ity. Generally, this can be assessed by calculating the runtime a = n(k) (38)
complexity of matrix multiplication. To begin with, the input
vector of forward propagation is described as x ∈ Rn , where z = n(k−1) (39)
the first element of the vector is a bias unit x0 = 1. Here, the
input is treated as an activation function with index x = a(0) and d = n(k−2) . (40)

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ADVANCED CONTROL STRATEGY WITH VOLTAGE SAG CLASSIFICATION FOR SINGLE-PHASE GCPVS 265

From the above equations, the number of multiplications


(nmul ) and number of activation function applications (ng ) are
calculated as
nlayers    

nmul = nk n(k=1) n(k−2) + n(1) n(0) 1 (41)
k=2

 
nlayers 
ng = n(k) . (42)
k=1

Besides, the naïve assumption that a(k) has same dimension


(k)
as θ(k) , which is not an obvious case for a(k) ∈ Rn , provides
time = nmul + ng ⇔ (43)
 
nlayers   
time = n(k) n(k−1) n(k−2) + n(1) n(0) 1
k=2

 
nlayers 
+ n(k) . (44)
k=1
Fig. 5. LVRT simulation at voltage sag between 0.5 ≤ Vpu ≤ 0.9: (a) voltage
Further, while analyzing the matrix algorithm, it is assumed and (b) current at PCC.
that the matrices are quadratic; this provides
nmul = nlayers · n3 . (45)
Besides, by reconsidering the assumption that there are the
same numbers of neurons in each layer and this is equal to the
number of layers, it is identified that
   
nmul = O n · n3 = O n4 . (46)
The same can be carried out for activation functions as
  Fig. 6. DC-link voltage for voltage sag between 0.5 ≤ Vpu ≤ 0.9.
ng = nlayers · n = O n2 . (47)
Therefore, the total runtime becomes
 
O n4 + n2 ⇔ (48)
 
O n4 ∵ ∀n ≥ 1|n4 + n2 ≤ 2n4 . (49)
From the above discussion, the NN structure with three layers
having 8, 12, and 7 neurons in each layer, respectively, and
operating with feedforward propagation takes 0.02 μs to classify
an input vector with 98.6% accuracy.

C. Simulation Testing
Fig. 7. Active and reactive power during LVRT simulation for voltage sag
To assess the robustness of the developed classifier and the between 0.5 ≤ Vpu ≤ 0.9.
proposed FRT controller, a line-to-ground (LG) fault is imple-
mented on the grid-connected system with design parameters
discussed in Table I using MATLAB/Simulink software. Ini- the condition as islanding, indicating the controller to act ac-
tially, the LG fault is implemented to create a voltage drop at cordingly. Further, with the action of the developed FCS-MPC
0.5 ≤ Vpu ≤ 0.9 on the normal operation of the power system controller, the reactive current is increased up to the threshold
network. The voltage (Vpcc ) and current (Ipcc ) at the PCC during limit as per the requirement to stabilize the dc-link voltage
the action of LG fault are shown in Fig. 5. From the figure, it can and provide voltage support in the network. The instantaneous
be identified that at t = 0.2 s, a voltage sag occurs disrupting saturation limit and latch limit strategies are adapted in this re-
the normal operation of the system. At this instant, the dc-link search to limit the increasing current value beyond the threshold
voltage experiences a transient as shown in Fig. 6 and the active limits. This process is based on the process of clipping the crest
power decreases to 2.6 kW as shown in Fig. 7. of the current for ensuring correct operation during the FRT
During this condition, the classifier monitors the voltage and [33]. Besides, the inverter control operates with constant peak
current at PCC and as the fault persists, the classifier classifies current control through the duration of the event to ensure safe

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
266 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 3, NO. 2, APRIL 2022

Fig. 10. Active and reactive power during LVRT simulation for voltage sag
Vpu < 0.5 p.u.

Fig. 8. LVRT simulation at voltage sag at Vpu < 0.5 p.u.: (a) voltage and
(b) current at PCC.

Fig. 11. Experimental setup for testing of the developed algorithm.

codes. The variations in voltage and current at PCC, dc-link


voltage, and active and reactive power during the postfault and
prefault conditions and actions of the controller are shown in
Figs. 8–10, respectively.
In both the conditions, the system is stabilized within 0.2 s
Fig. 9. DC-link voltage sag operation Vpu < 0.5 p.u.
by restoring the grid voltage and controlling the dc-link volt-
age. This identifies the effectiveness of the developed control
approach.
operation and avoid overcurrent loading. This resulted in voltage
restoration by t = 0.4 s, which less than the time prescribed by
the grid codes [34], [35]. The variations in voltage and current at D. Experiment
PCC, dc-link voltage, and active and reactive power during the To assess the performance of the developed FCS-MPC con-
postfault and prefault conditions and actions of the controller troller, the experimental tests are carried out on a 4 kW p single-
are shown in Figs. 5–7, respectively. phase two-stage GCPVS. The PV array is realized through the
Similarly, the LG fault is implemented to create a voltage sag Keysight solar simulator which exhibits the operation of PV sys-
Vpu < 0.5 p.u. on the normal operation of the power system tem under varying irradiance profile. Further, the boost conver-
network. The Vpcc and Ipcc during LG fault are shown in Fig. 8. sion stage and single-phase inverter are developed with the de-
From the figure, it can be identified that at t = 0.2 s, a volt- tails provided in Table I. Besides, the developed controller is dig-
age sag occurs, disrupting the normal operation of the system. itally implemented using Altera Cyclone IV EP4CE115F29C7N
During this condition, the FRT scheme generates the power field-programmable gate array. The grid integration of the two-
references to inject the full reactive current, which is calculated stage single-phase system is achieved by connecting the inverter
to be 1.5 times the rated current of the inverter [36]. To achieve to a low-voltage grid simulated using Chroma regenerative grid
the overcurrent protection, the instantaneous saturation limit simulator. The complete experimental setup of the testing system
strategy is activated at 0.2 s to monitor the increase of current is shown in Fig. 11. The experiment is performed for a similar
levels beyond the rated current value. At this instant, the dc-link fault scenario as discussed in simulation analysis.
voltage experiences high transient, as shown in Fig. 9, and the In Fig. 12, a fault which enables a deflection of voltage
active power decreases to 1 kW as shown in Fig. 10. Further, between 0.5 ≤ Vpu < 0.9 at Vpcc is injected after 220 ms. At this
the classifier monitors Vpcc and Ipcc and classifies the condition point, the islanding classification algorithm classifies the state
as islanding, indicating the FCS-MPC controller to perform the of the system as an islanding scenario. Further, the FRT scheme
necessary control operation. This resulted in voltage restoration injects reactive current by calculating the power reference. As
by t = 0.4 s, which is less than the time prescribed by the grid the voltage levels are identified between 0.5 ≤ Vpu < 0.9, the

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ADVANCED CONTROL STRATEGY WITH VOLTAGE SAG CLASSIFICATION FOR SINGLE-PHASE GCPVS 267

TABLE II
COMPARISON BETWEEN DIFFERENT REACTIVE POWER CONTROLS

Fig. 12. Voltage and current at PCC for LVRT experiment.

Fig. 13. Active and reactive power during the LVRT experiment.

maximum allowable current is equal to the rated current of


the inverter. Further, the inverter reacts, maintaining the dc-link
voltage and injecting the required reactive power, as shown in
Fig. 13.
The proposed FCS-MPC shows a good performance through TABLE III
the duration of the event to meet the LVRT requirement. The COMPARISON OF CLASSIFICATION AND REACTIVE
POWER CONTROL STRATEGIES
peak current control shows that the inverter is operated within
its safety limits and its maximum nominal capacity at all times.
Therefore, unintentional inverter shutdown is avoided during
voltage sag operation, which demonstrates FRT capability. It
is also important to mention that the amount of reactive power
injected into the grid is always lower than the inverter capacity.
Injecting full inverter capacity reactive power during voltage
sags will cause a safety tripping due to overcurrent conditions.
The advantages of the proposed reactive power control technique
FRT are provided in Table II.
A brief comparison of the developed classification and reac-
tive power control strategy with the mechanisms available in the
literature is shown in Table III.

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
268 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 3, NO. 2, APRIL 2022

V. CONCLUSION [16] H. R. Baghaee, D. Mlakic, S. Nikolovski, and T. D. Dragicevic, “Support


vector machine-based islanding and grid fault detection in active distribu-
An FCS-MPC was developed in this article for achieving tion networks,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 8, no. 3,
FRT in single-phase two-stage GCPVS. The developed control pp. 2385–2403, Sep. 2020.
[17] K. Ma, M. Liserre, and F. Blaabjerg, “Power controllability of three-phase
strategy was successfully able to achieve LVRT during voltage converter with unbalanced AC source,” in Proc. 28th Annu. IEEE Appl.
sags in the system. Initially, the control strategy aimed at de- Power Electron. Conf. Expo., 2013, pp. 342–350.
tecting the grid faults in a GCPVS. This process is achieved by [18] C. Bordons and C. Montero, “Basic principles of MPC for power convert-
ers: Bridging the gap between theory and practice,” IEEE Ind. Electron.
using pattern recognition characteristics of the NN. The strategy Mag., vol. 9, no. 3, pp. 31–43, Sep. 2015.
successfully classifies the faults with 98.6% accuracy under [19] J. Rodriguez and P. Cortes, Predictive Control of Power Converters and
2 ms. Once the fault was identified, the FCS-MPC controller Electrical Drives. Hoboken, NJ, USA: Wiley-IEEE Press, 2012.
[20] J. M. Maciejowski, P. J. Goulart, and E. C. Kerrigan, “Constrained control
provides reactive power support to enhance the FRT capability of using model predictive control,” in Advanced Strategies in Control Systems
the grid-connected inverter. The developed classifier and control With Input and Output Constraints, Berlin, Germany: Springer, 2003,
algorithm were assessed by performing numerical simulations pp. 273–291.
[21] A. Safari and S. Mekhilef, “Simulation and hardware implementation of
and experimental analysis. The results obtained confirm that the incremental conductance MPPT with direct control method using Cuk
MPC scheme provides an excellent control performance under converter,” IEEE Trans. Ind. Electron., vol. 58, no. 4, pp. 1154–1161,
voltage sag conditions and satisfies grid code requirements. Apr. 2011.
[22] S. Golestan, M. Monfared, F. D. Freijedo, and J. M. Guerrero, “Design and
tuning of a modified power-based PLL for single-phase grid-connected
power conditioning systems,” IEEE Trans. Power Electron., vol. 27, no. 8,
REFERENCES pp. 3639–3650, Aug. 2012.
[23] IEEE Standard for Interconnecting Distributed Resources With Electric
[1] K. N. Nwaigwe, P. Mutabilwa, and E. Dintwa, “An overview of solar
Power Systems, IEEE Std 1547:2003, 2003.
power (PV systems) integration into electricity grids,” Mater. Sci. Energy
[24] O. P. Mahela, N. Gupta, M. Khosravy, and N. Patel, “Comprehensive
Technol., vol. 2, no. 3, pp. 629–633, Dec. 2019.
overview of low voltage ride through methods of grid integrated wind
[2] Y.-K. Wu, J.-H. Lin, and H.-J. Lin, “Standards and guidelines for grid-
generator,” IEEE Access, vol. 7, pp. 99299–99326, 2019.
connected photovoltaic generation systems: A review and comparison,”
[25] P. Sochor, N. M. L. Tan, and H. Akagi, “Low-voltage-ride-through con-
IEEE Trans. Ind. Appl., vol. 53, no. 4, pp. 3205–3216, Jul. 2017.
trol of a modular multilevel single-delta bridge-cell (SDBC) inverter for
[3] Y. Zhang, “Anti-islanding modeling of grid-tied inverters,” in Proc. Int.
utility-scale photovoltaic systems,” IEEE Trans. Ind. Appl., vol. 54, no. 5,
Power Electron. Appl. Conf. Expo., 2014, pp. 644–647.
pp. 4739–4751, Sep. 2018.
[4] P. Piya, M. Ebrahimi, M. Karimi-Ghartemani, and S. A. Khajehoddin,
[26] H. D. Dehghani Tafti, A. I. Maswood, G. Konstantinou, J. Pou, and P.
“Fault ride-through capability of voltage-controlled inverters,” IEEE
Acuna, “Active/reactive power control of photovoltaic grid-tied inverters
Trans. Ind. Electron., vol. 65, no. 10, pp. 7933–7943, Oct. 2018.
with peak current limitation and zero active power oscillation during unbal-
[5] J. E.-G. Carrasco, J. M. Tena, D. Ugena, J. Alonso-Martinez, D. Santos-
anced voltage sags,” IET Power Electron., vol. 11, no. 6, pp. 1066–1073,
Martin, and S. Arnaltes, “Testing low voltage ride through capabilities of
May 2018.
solar inverters,” Elect. Power Syst. Res., vol. 96, pp. 111–118, Mar. 2013.
[27] Y. Yang, H. Wang, and F. Blaabjerg, “Reactive power injection strategies
[6] J. Miret, A. Camacho, M. Castilla, L. G. de Vicuna, and J. Matas, “Control
for single-phase photovoltaic systems considering grid requirements,”
scheme with voltage support capability for distributed generation invert-
IEEE Trans. Ind. Appl., vol. 50, no. 6, pp. 4065–4076, Nov./Dec. 2014.
ers under voltage sags,” IEEE Trans. Power Electron., vol. 28, no. 11,
[28] S. Kouro, P. Cortes, R. Vargas, U. Ammann, and J. Rodriguez, “Model
pp. 5252–5262, Nov. 2013.
predictive control—A simple and powerful method to control power
[7] C.-Y. Tang, Y.-T. Chen, and Y.-M. Chen, “PV power system with multi-
converters,” IEEE Trans. Ind. Electron., vol. 56, no. 6, pp. 1826–1838,
mode operation and low-voltage ride-through capability,” IEEE Trans. Ind.
Jun. 2009.
Electron., vol. 62, no. 12, pp. 7524–7533, Dec. 2015.
[29] M. Rivera, V. Yaramasu, J. Rodriguez, and B. Wu, “Model predictive
[8] J. L. Sosa, M. Castilla, J. Miret, J. Matas, and Y. A. Al-Turki, “Control strat-
current control of two-level four-leg inverters—Part II: Experimental
egy to maximize the power capability of PV three-phase inverters during
implementation and validation,” IEEE Trans. Power Electron., vol. 28,
voltage sags,” IEEE Trans. Power Electron., vol. 31, no. 4, pp. 3314–3323,
no. 7, pp. 3469–3478, Jul. 2013.
Apr. 2016.
[30] A. Calle-Prado, S. Alepuz, J. Bordonau, J. Nicolas-Apruzzese, P. Cortes,
[9] H. Khan, S. J. Chacko, B. G. Fernandes, and A. Kulkarni, “Reliable and
and J. Rodriguez, “Model predictive current control of grid-connected
effective ride-through controller operation for smart PV systems connected
neutral-point-clamped converters to meet low-voltage ride-through re-
to LV distribution grid under abnormal voltages,” IEEE J. Emerg. Sel.
quirements,” IEEE Trans. Ind. Electron., vol. 62, no. 3, pp. 1503–1514,
Topics Power Electron., vol. 8, no. 3, pp. 2371–2384, Sep. 2020.
Mar. 2015.
[10] L. Zhou, A. Swain, and A. Ukil, “Reinforcement Learning Controllers
[31] S. B. Kjaer, J. K. Pedersen, and F. Blaabjerg, “A review of single-phase
for Enhancement of Low Voltage Ride Through Capability in Hybrid
grid-connected inverters for photovoltaic modules,” IEEE Trans. Ind.
Power Systems,” IEEE Trans. Ind. Informat., vol. 16, no. 8, pp. 5023–5031,
Appl., vol. 41, no. 5, pp. 1292–1306, Sep./Oct. 2005.
Aug. 2020.
[32] M. A. Khan, V. S. B. Kurukuru, A. Haque, and S. Mekhilef, “Island-
[11] A. Z. Fatama, M. A. Khan, V. S. B. Kurukuru, A. Haque, and F. Blaabjerg,
ing classification mechanism for grid-connected photovoltaic systems,”
“Coordinated reactive power strategy using static synchronous compen-
IEEE J. Emerg. Sel. Topics Power Electron., early access, Apr. 7, 2020,
sator for photovoltaic inverters,” Int. Trans. Elect. Energy Syst., vol. 30,
doi: 10.1109/JESTPE.2020.2986262.
no. 6, Jun. 2020, Art. no. e12393.
[33] C. A. Plet, M. Graovac, T. C. Green, and R. Iravani, “Fault response of
[12] A. Khamis, H. Shareef, E. Bizkevelci, and T. Khatib, “A review of island-
grid-connected inverter dominated networks,” in Proc. IEEE Power Energy
ing detection techniques for renewable distributed generation systems,”
Soc. Gen. Meeting, 2010, pp. 1–8.
Renew. Sustain. Energy Rev., vol. 28, pp. 483–493, Dec. 2013.
[34] IEEE Standard for Interconnection and Interoperability of Distributed
[13] D. Voglitsis, N. P. Papanikolaou, and A. C. Kyritsis, “Incorporation of
Energy Resources With Associated Electric Power Systems Interfaces,
harmonic injection in an interleaved flyback inverter for the implementa-
IEEE STD 1547-2018, 2018.
tion of an active anti-islanding technique,” IEEE Trans. Power Electron.,
[35] Utility-Interconnected Photovoltaic Inverters - Test Procedure of Islanding
vol. 32, no. 11, pp. 8526–8543, Nov. 2017.
Prevention Measures, IEC 62116:2014, 2014.
[14] D. Mlakic, H. R. Baghaee, and S. Nikolovski, “A novel ANFIS-based
[36] Y. Yang, F. Blaabjerg, and H. Wang, “Low-voltage ride-through of single-
islanding detection for inverter-interfaced microgrids,” IEEE Trans. Smart
phase transformerless photovoltaic inverters,” IEEE Trans. Ind. Appl.,
Grid, vol. 10, no. 4, pp. 4411–4424, Jul. 2019.
vol. 50, no. 3, pp. 1942–1952, May/Jun. 2014.
[15] M. R. Alam, K. M. Muttaqi, and A. Bouzerdoum, “Evaluating the ef-
[37] Y. Yang, A. Sangwongwanich, H. Liu, and F. Blaabjerg, “Low voltage
fectiveness of a machine learning approach based on response time and
ride-through of two-stage grid-connected photovoltaic systems through the
reliability for islanding detection of distributed generation,” IET Renew.
inherent linear power-voltage characteristic,” in Proc. IEEE Appl. Power
Power Gener., vol. 11, no. 11, pp. 1392–1400, Sep. 2017.
Electron. Conf. Expo., 2017, pp. 3582–3588.

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ADVANCED CONTROL STRATEGY WITH VOLTAGE SAG CLASSIFICATION FOR SINGLE-PHASE GCPVS 269

[38] M. Elnozahy, E. El-saadany, and M. Salama, “A robust wavelet-ANN V. S. Bharath Kurukuru (Student Member, IEEE)
based technique for islanding detection,” in Proc. Power Energy Soc. Gen. received the B.Tech. degree in electrical and electron-
Meeting, 2011, pp. 1–8. ics engineering from Avanthi’s Research and Techno-
[39] M. Ahmadipour, H. Hizam, M. L. Othman, and M. A. Radzi, “Islanding logical Academy, Vizianagaram, India, in 2014, and
detection method using ridgelet probabilistic neural network in distributed the M.Tech. degree in power systems from Amity
generation,” Neurocomputing, vol. 329, pp. 188–209, 2019. University, Noida, India, in 2016. He is currently
[40] Y. Yang and F. Blaabjerg, “Low- oltage ride-through capability of a single- working toward the Ph.D. degree in intelligent mon-
stage single-phase photovoltaic system connected to the low-voltage grid,” itoring of photovoltaic systems with the Advance
Int. J. Photoenergy, vol. 2013, pp. 1–9, 2013. Power Electronics Research Laboratory, Department
of Electrical Engineering, Jamia Millia Islamia (A
Central University), New Delhi, India.
In 2019, he was a Visiting Research Fellow with the Center of Reliable Power
Electronics, Aalborg University, Aalborg, Denmark. He research interests in-
clude fault diagnosis, condition monitoring, and reliability of power electronics
converters in renewable energy systems and electric vehicles.
Mohammed Ali Khan (Student Member, IEEE) re-
ceived the B.Tech. degree in electrical and electronics
engineering from Karunya University, Coimbatore,
India, in 2013, and the M.Tech. degree in power
system from Amity University, Noida, India, in 2016.
He is currently working toward the Ph.D. degree
in Power electronics converter controlling with the
Department of Electrical Engineering, Jamia Millia Saad Mekhilef (Senior Member, IEEE) received
Islamia (A Central University), New Delhi, India. the B.Eng. degree in electrical engineering from the
In 2020, he has been a Visiting Researcher with University of Setif, Setif, Algeria, in 1995, and the
the Center of Reliable Power Electronics, Aalborg master’s degree in engineering science and the Ph.D.
University, Aalborg, Denmark. He was a Guest Faculty with the Department degree in electrical engineering from the University of
of Electrical Engineering, Jamia Millia Islamia. He has authored or coauthored Malaya, Kuala Lumpur, Malaysia, in 1998 and 2003,
many articles in peer-reviewed journals and presented his research articles in respectively.
several international conferences. His research interests include artificial intel- He is currently a Professor and the Director of the
ligence, power electronics, and their application in renewable energy systems, Power Electronics and Renewable Energy Research
power quality improvements, and smart grids. Laboratory, Department of Electrical Engineering,
University of Malaya. He is also the Dean of the
Faculty of Engineering, University of Malaya. He is also a Distinguished Adjunct
Professor with the Faculty of Science, Engineering and Technology, School of
Software and Electrical Engineering, Swinburne University of Technology, Mel-
bourne, VIC, Australia. His current research interests include power converter
topologies, control of power converters, renewable energy, and energy efficiency.
Prof. Mekhilef is an Editor for Renewable and Sustainable Energy Reviews
Ahteshamul Haque (Senior Member, IEEE) re- and an Associate Editor for the IEEE TRANSACTIONS ON POWER ELECTRONICS,
ceived the B.Tech. degree from Aligarh Muslim Uni- IEEE OPEN JOURNAL OF INDUSTRIAL ELECTRONICS, International Journal of
versity, Aligarh, India, in 1999, the master’s degree Circuit Theory and Applications, and Journal of Power Electronics. He is also
from the Indian Institute of Technology Delhi, New a Guest Editor for the IEEE JOURNAL OF EMERGING AND SELECTED TOPICS
Delhi, India, in 2000, and the Ph.D. degree from the IN INDUSTRIAL ELECTRONICS, IET Power Electronics, IET Renewable Power
Department of Electrical Engineering, Jamia Millia Generation, and International Transactions on Electrical Energy Systems.
Islamia (A Central University), New Delhi, India, in
2015, all in electrical engineering.
Prior to academics, he was with the research and
development units of world reputed multinational
industries and his work has been patented in USA and
Europe. He is currently a Professor (Assistant-Senior Scale) with the Department
of Electrical Engineering, Jamia Millia Islamia. He has established Advance
Power Electronics Research Laboratory, Department of Electrical Engineering,
Jamia Millia Islamia. He is a Principal Investigator of the MHRD-SPARC Project
and other research and development projects. He has authored or coauthored
around 100 publications in international journals and conference proceedings.
His current research interests include power converter topologies, control of
power converters, renewable energy, energy efficiency, reliability analysis, and
electric vehicle operations.
Dr. Haque was the recipient of the IEEE PES Outstanding Engineer Award
for the year 2019.

Authorized licensed use limited to: Konya Teknik Universitesi. Downloaded on December 16,2023 at 12:48:44 UTC from IEEE Xplore. Restrictions apply.

You might also like