1

A Dual-Discrete Model Predictive Control-

Based MPPT for PV Systems

Abderezak Lashab, Student Member, IEEE, Dezso Sera, Senior Member, IEEE, and Josep M. Guerrero, Fellow, IEEE


Abstract--This paper presents a method that overcomes the under this peak simultaneously [2]. These algorithms are
problem of the confusion during fast irradiance change in the named as maximum power point trackers (MPPT). The
classical MPPTs as well as in model predictive control (MPC)- classical and the most well-known MPPT is the Perturb and
based MPPTs available in the literature. The previously Observe (P&O). In fact, this algorithm is simple and requires
introduced MPC-based MPPTs take into account the model of only the use of sensors for measuring the PV current and
the converter only, which make them prone to the drift during
voltage. But, as its name denotes, this algorithm continuously
fast environmental conditions. Therefore, the model of the PV
array is also considered in the proposed algorithm, which allows perturbs the voltage (in case of voltage control) by
it to be prompt during rapid environmental condition changes. It adding/subtracting a fixed voltage increment (ΔV) to/from the
takes into account multiple previous samples of power, and based PV voltage, which produces some oscillations in the output
on that is able to take the correct tracking decision when the PV power. Also, its speed convergence is limited, by reason
predicted and measured power differ (in case of drift issue). that the choice of the step size is linked to the steady state
After the tracking decision is taken, it will be sent to a second operation conditions [3], [4]. Incremental Conductance (INC)
part of the algorithm as a reference. The second part is used for also is a well-known MPPT [5], [6], and its operational
following the reference provided by the first part, where the principle is very congruous to that of P&O algorithm. It
pulses are sent directly to the converter, without a modulator or a
therefore provides tantamount static and dynamic perfor-
linear controller. The proposed technique is validated
experimentally by using a buck converter, fed by a PV simulator. mances as P&O according to the investigation reported in [7].
The tracking efficiency is evaluated according to EN50530 There exist other classical methods, such as fractional open-
standard in static and dynamic conditions. The experimental circuit voltage (FOCV) [8], and fractional short-circuit current
results show that the proposed MPC-MPPT is a quick and (FSCC) [2]. But, these methods do not converge to the true
accurate tracker under very fast changing irradiance, while MPP, and they suffer from power loss during the measurement
maintaining high tracking efficiency even under very low of the fractional variable. The relative merits of these
irradiance. numerous approaches are discussed and investigated in [2].
The fact that the classical MPPTs fail to pursue the MPP
Index Terms-- Buck converter, dc-dc power conversion, Drift, under rapidly changing atmospheric conditions, has raised
Double cost function, EN50530 standard, Maximum power point
concern of many researchers [9]-[14]. This issue is referred as
tracking, MPC, Photovoltaic systems.
drift in the literature [9]. For instance, the conventional P&O
fails to track the MPP during fast environmental condition
changes, because this algorithm and its rules are designed for a
I. INTRODUCTION
static PV curve, and if there is a fast change in the

P HOTOVOLTAIC (PV) electricity production system is

one of the most essential renewable energy systems, due to
its advantageous features, primarily the clean, free, and
irradiance/temperature, the rules of this algorithm are no
longer sufficient. As a scenario, if the result of the condition
Ppv(k)-Ppv(k-1)>0 is yes, P&O considers that the operating
unlimited resource. It is predicted that in 2035, the energy point is approaching the MPP, and subsequently, the same
generated by PV systems will increase by almost 20 times, decision as the previous one will be taken. However, there is
expanding to 846TWh [1]. another probability P&O is not designed to be aware of, which
Under each irradiance/temperature level, the PV array is, the increase of power caused by the increase of irradiance
provides different output power vs voltage characteristic P(v). during one perturbation period is larger than the increase in
This latter, is nonlinear and in normal conditions has only one power induced by the previous perturbation. In this case, the
peak, indeed it has a shape close to the intersection shape “∩”. operating point is may be going far away from the MPP. In
The peak of this curve is usually referred to as the maximum [9], a condition has been added to P&O by observing the
power point (MPP). Various algorithms have been proposed in change in current, which provides to P&O the knowledge
the literature for defining and making the PV module working when the operating point goes to the right side of the MPP.
But, the operating point may go to the right or left side of the
MPP, that depends upon the last action taken by P&O just
prior to the irradiance change. In [10], the change of power
The authors are with the Department of Energy Technology, Aalborg
University, 9220 Aalborg, Denmark (e-mail: [email protected]; [email protected];
resulted by the environmental condition changes is subtracted
[email protected]). from the overall resulted PV power, to allow the P&O
 2

discriminate the change in power resulted by incre- MPC-MPPT [25]-[29] and Digital Observer (DO)-MPC-
menting/decrementing the involved reference from the power MPPT [30]-[31].
caused by the insolation changes. In [11], maximum and In FCS-MPC-MPPT, the discrete-time model of the system
minimum boundaries have been set to limit the PV voltage is used to predict the behavior of the controlled variable up to
around the estimated MPP. This avoids undesirable excursions N horizon length. The switching state that entails a minimized
of the PV voltage during rapidly changing atmospheric cost function will be selected to be applied during the next
conditions. In [12], multi-sampling (MS) MPPT has been sampling time directly to the converter without the necessity
developed, in which a voltage step size is incremented and of a PI controller or a modulation stage. Among the merits of
decremented and incremented (+-+), and based on the behavior FCS in power electronics generally are its fast dynamic
of the PV power caused by these actions, the right decision response and its ability of handling nonlinearities, as well as
will be set. This approach provides a good dynamics. including multi-variables in the cost function. The references
However, in some cases, under very fast irradiance change, provided to FCS-MPC-MPPT are calculated by using P&O
the same action should be set successively in order to track the [25] or INC [26]-[29]. In DO-MPC-MPPT, a digital observer
MPP. In [13], an adaptive algorithm that tunes continuously is adopted for the prediction of a PV currents conformable to
the size of the increment in P&O has been proposed. Although an assumed PV voltages, where the PV voltages are shifted by
this technique alleviates the drift, large oscillations are a predicted step size. In [32], the efficiency of Discrete-MPC-
produced during the change in irradiance. Moreover, based MPPT has been deeply studied, considering different
according to realistic weather changes, if the MPP controller is weather conditions as well as various power converter
sufficiently sophisticated to provide right decisions during topologies, and, as it has been reported, when using FCS-
varying irradiance/temperature, a fixed step size would be MPC-based MPPT; the resulted MPP tracker will have the
adequate even if the atmospheric change is fast [41]-[42]. The same shortcomings as the used reference (in that paper P&O is
combination of FSCC and P&O has been proposed in [14] to considered). Regarding DO-MPC-MPPT, a better performance
detect the change in irradiance. The method starts first with during a changing environmental conditions compared to
FSCC by estimating the current of the MPP based on the FCS-MPC-MPPT can be obtained, but tracking the MPP
measured short circuit current. Afterwards, the algorithm starts under fast environmental condition changes is still a challenge
working by using P&O. At each iteration, the PV current is for DO-MPC as well.
compared with one calculated first with FSCC, if the differ- Based on the existing body of literature, the drift issue in
ence exceeds a certain limit (in case of irradiance change), the MPC-MPPTs is still unsolved, which retained the research in
algorithm start over. The convergence time is short at the start- this area ongoing. In this paper, a method using model
up, and the drift issue is mitigated, however, may still some predictive control in both sides, PV and converter is proposed,
losses since the operation principle is approximation based. where the main objective is drift avoidance during fast
Recently, intelligent controllers such as fuzzy logic irradiance change by using MPC.
controller [15], neural network controller [16], sliding mode
[17], and model predictive controller [25], have been used for
tracking the MPP to overcome the drawbacks of the classical II. OPERATION PRINCIPLE OF FINITE-CONTROL-SET MPC-
ones. Both fuzzy logic and neural network controllers are BASED MPPT
appropriate for applications where the mathematical model of
the system or some of its parameters are undefined. Sliding The dc-dc buck topology used in this paper is illustrated in
mode offers robustness and takes into consideration the Fig. 1. Since only one switch is used in the selected topology,
switching nature of the power converter [21]. The main the control operation is simpler than other topologies, such as,
feature of MPC is its estimation of the future conduct of the series capacitor buck converter [33]. A one step ahead is the
controlled variable. horizon length used in this paper. The first step of FCS-MPC-
The computational cost of MPC, which was important in the MPPT implementation procedure is defining the system
past years, has now become a minor issue, since powerful equations. By applying Kirchhoff´s voltage and current laws
digital microprocessors and FPGAs that can execute complex on the electrical circuit in Fig. 1, the model in continuous-time
calculations in a short time were developed. This fact has led domain of the buck converter for the two states can be found
to a significant attention to the implementation of MPC in as follows
power electronics applications such as dc-dc converters, Switch ON
electric drives, multilevel inverters, and matrix converters  di L
[18]-[20]. MPC in power electronics is subdivided into two  L dt = v pv  v C 2  rL i L (1)
main categories [21], [22]: continuous control set MPC (CCS- 
C dv
MPC) and finite control set MPC (FCS-MPC). In the first
C 2
= iL  iR
 2 dt
class, the gate drive signals are generated from a modulator,
where its input is a continuous predicted variable. The second Switch OFF
class exploits the finite number of the switching states of the  diL
converter to restrain the error between the controlled variable  L dt =  d aux vC2  d aux rL iL (2)

and its given reference [22]. C dv
As reported in the literature, MPC in MPPTs is subdivided
C2
= d aux iL  iR
 2 dt
into two major classes, CCS-MPC-MPPT [23]-[24] and such as the four state variables vpv, iL, vC2 and, iR are the PV
Discrete-MPC-MPPT, the later itself is subdivided into FCS-
 3

voltage, the current through the inductor L, the output voltage Converter model
Converter
of the converter, and the current going through the load R, FCS‐MPC Gating
signal
Actual samlpe
respectively. daux is equal to “1” during the continuous current Does the
predicted power at No Apply an opposite action to
mode (CCM), whereas during the discontinuous current mode
*
the previous sampling time the on appiled in the IPV (k+1)
equal to the actual previous sampling time
(DCM) and after the switch opens and the current in the power?
inductor gets nulled, it takes the value of “0” [34]. In what Yes

follows, it is assumed that the inductor stray resistance rL is Part of the PV curve
Two previous samples extraploation Predicted PV current/voltage
equal to zero, and the converter is operating in CCM. (Lagrange)
Usually, the discrete-time model of the system is obtained
by using Euler´s forward-difference law, which can be Fig. 3. Overview of the proposed approach.
expressed as
dx x ( k +1) ‐ x ( k ) (3)
 where, λI and λv are the current and voltage weighting factors,
dt Ts
where Ts is the sampling time. The substitution of (3) into respectively. In the cost function, each term is weighted
buck converter’s Switch ON equations yields to through these weighting factors in order to reach the desired
balance between the priorities among the control targets and
 Ts

 iL ( k + 1)  L v pv ( k )  vC2 ( k )  + iL ( k ) (4) constraints. Definitely, the larger the weighting factor, the
 larger priority assigned to the corresponding term. Different
 iL ( k + 1)  iR ( k )   vC2 ( k )
Ts
 vC2 ( k + 1) = approaches are usually used to determine the weighting
 C2
factors, the most adopted one is based on empirical methods
The model of buck converter’s Switch OFF state in discrete- [22]. Despite the fact that in [35] some guidelines for the
time domain was found similarly as design of the weighting factors are given, there are still no
 Ts analytical or mathematical methods to ultimately overcome
iL ( k +1) =  L vC2 ( k ) + iL ( k ) (5)
 this issue. The weighting factors design could be complex
vC2 ( k +1) =
Ts
 iL (k +1)  iR (k )  + vC2 (k ) since in some systems the design done for a specific operating
 C2 region, is not valid for another one. On that account,
The average value of the current going through the capacitor intelligent controllers, such as, Artificial Neural Network [36]
C2 is zero. Hence, the average current going through the and Fuzzy Logic [36][37] are being employed to address this
inductor L equals to the average value of the output current. issue, where the optimization process is performed online.
The relationship between the inductor current and input After the cost function optimization, the controller has to
current can be then expressed as follows wait until tk reaches Ts. Where tk is the time from the last
i pv ( k +1)  D.iL ( k +1) (6) application of the gate signals. Thereafter, the switching state
corresponding to the evaluated cost function can be applied
where D is the duty cycle of the gating signal.
directly to the converter.
The predicted PV current can be calculated by substituting
(6) into (4) and (5). Generally, the cost function is calculated
by using the predicted variables and their references, where
the references are calculated based on P&O or INC algorithm, III. PROPOSED MODEL PREDICTIVE CONTROL-BASED MPPT
g  I  i pv ( k + 1)  i * + v  v pv ( k + 1)  v * (7) In the literature, P&O and its alternate implementation, the
0,1 0,1 0,1
INC are the ones used for providing the references to FCS-
MPC [25]-[29]. But, P&O and INC methods have a poor
dynamic performance under rapidly varying environmental
conditions, which influence on the MPPT efficiency
iPV iL rL L iR negatively. Also, the generated power by using these two
methods fluctuates in the steady state, causing some losses.
 S  The application of FCS with the inclusion of these two main
PV vPV C1
D C2 vC2 R drawbacks of P&O/INC method, will result to an MPPT
  hampered by them. For this purpose, an improved predictive
control algorithm has been designed in this paper, its
Fig. 1. A simplified configuration of PV system interfaced by a dc-dc buck
flowchart is sketched in Fig. 5. The novelty of this work
converter. consists of integrating two predictions into a single MPP
tracker as depicted in Fig. 3, where:
iPV iL rL L iR iPV rL L iR  The first prediction is based on the estimation of the
predicted PV voltage/current on the extrapolated PV curve,
vPV
 
vC2 vPV
 
vC2
which will then serve as a reference (blue color in Fig. 3).

C1 C2

R

C1 C2

R
The prominent role of these predictions is during dynamic
weather conditions as will be explained next.
(a) (b)  In order to increase the dynamic reference tracking
performance during the start-up of the system, and also
Fig. 2. (a) equivalent dc-dc buck converter during Switch ON state, (b)
equivalent dc-dc buck converter during Switch OFF state.
during load variation, the dynamic behavior of the
converter is going to be predicted by introducing FCS-
 4

MPC as a second prediction part in the proposed algorithm, Start

which also allows the elimination of the PI controller from
the voltage/current regulation loop as well as the
Updates
modulation stage (red color in Fig. 3).
v pv ( k ) , i pv ( k ) , vC2 ( k ) , tk = 0 ;

Reference generation
1) Static weather conditions Yes
A. Reference Generation Pexp - Ppv (k)  

In MPC techniques, the discretized equations of the system No

are used for the estimation of the future action of the Predicted voltages
controlled variable. Concerning the PV array, a high accuracy
i pv ( k +1)1,2 = i pv ( k )  i ;
model of the system is extremely difficult and unpractical to
build, because a lot of factors are continuously changing such Extrapolation of the predicted
as the solar irradiance, temperature, and the degradation of the voltages by using (18), (19),
(20), and (15) Apply a decision opposite to
PV modules. For this reason, an algorithm that identifies the the one taken in the last
model of part of the PV curve at each sampling period by First cost function
sampling time
i pv ( k + 1) * = i pv ( k ) * +
interpolating it based on Lagrange polynomial has been g1 1 ,2  Ppv ( k + 1)1,2 - Ppv ( k ) i pv ( k - 1) * - i pv ( k - 2 ) *
developed in this work (please see Fig. 4). Lagrange
polynomial (POl) is interpolated using a data points as follows First cost function evaluation
Pol ( xi )  y i ........ for . .i  0,1, . .., n 
ipv (k +1)*  f max g1i i 1 ,2 
And Lagrange polynomial will have the following form 
Pexp  f max  g1 i i 1 ,2 
Pol ( x )  an x n  an 1 x n 1  ...  a2 x 2  a1 x  a0 (8)
Predicted currents estimation
n 1 n2
 x ... x ... x ...... x 0 ... 1  a n   y 0 
n
0 0 0  Ts 
 n n 1 n  2     i pv ,s=0 ( k +1) =  
 L
vC2 ( k ) + iL (k )   D

 x1 ... x1 ... x1 .. .... x1 ... 1  a n 1    y1  (9)  Ts 
i pv ,s=1 ( k + 1) =   v pv ( k )  vC2 ( k )  + iL ( k )   D
................ ...      L 
    
 x n ... x n ... x n .. .... x n .. .1   a 0   y n 
n n 1 n2
Second cost function
2
g2 0,1  i pv 0,1 (k +1) - i pv (k +1) *
     
By substituting (8) in (9), we get a system of set of equations
in am coefficients. The matrix on the left is usually referred to Second cost function evaluation

as Vandermonde Matrix. There are various proposed 

S ( k + 1)  f min g 2 i i 0 ,1 
algorithms that exploit the scheme of Vandermonde Matrix to
compute stable solutions by Gaussian elimination [38], [39]. No Yes Apply the switching
The interpolated polynomial can be written in terms of the t k  Ts state

Lagrange polynomials as follows

( x  x1 )( x  x2 )   ( x  xn ) Fig. 5. Flowchart of the proposed MPC-MPPT.
Pol ( x )  y0 
( x0  x1 )( x0  x2 )   ( x0  xn )
( x  x0 )( x  x2 )   ( x  xn )
:::::::::::::: y1  (10) Lagrange polynomials are generally expressed in
( x1  x0 )( x1  x2 )   ( x1  xn ) Sylvester´s Formula, as the following way
:::::::::::::::::::::::::::::::::   
x  xj
Pol ( x )    
n
( x  x0 )( x  x1 )   ( x  xn 1 )  y (11)
:::::::::::    i
i  0  0  j  n xi  x j
yn
( xn  x0 )( xn  x1 )   ( xn  xn 1 ) 
 :: j  i 
To interpolate a part of the PV curve that is in the
neighborhood of the operating point, a data points constituted
of ipv(k-2){vpv(k-2)}, ipv(k-1){vpv(k-1)}, and ipv(k){vpv(k)} are
vPV (k  6), vPV (k  5), vPV (k  4),  vPV (k  3)
PPV vPV (k  5), vPV (k  4), vPV (k  3),  vPV ( k  2) used. Where k-2 denotes to the sampling time before the last
vPV (k  4), vPV (k  3), vPV (k  2),  vPV (k 1) one, k-1 denotes to the previous sampling time, and k denotes
vPV (k  3), vPV (k  2), vPV (k 1),  vPV (k )  vMPP
to the present sampling time. Hence, the following Lagrange
PMPP polynomial for the PV curve is proposed
v pv ( k )  a 2 i p v ( k ) 2  a1i pv ( k )  a 0 (12)
Vandermonde Matrix can be then written as follows

i pv 2 ( k - 2 ) i pv ( k - 2 ) 1  a   v pv ( k - 2 ) 
vPV( 5)
v PV 6)
v PV

vPV 3)
vPV(k 4)

v PV 2)
v PV 1)

 vMPP  2  2   (13)
(k
(k

(k
(k
(k)
k

vMPP  pv
i ( k - 1) i ( k - 1) 1   a1    v pv ( k - 1) 
vPV  2
pv
  
Fig. 4. Extrapolation of the predicted current based on the predicted PV  i pv ( k ) i pv ( k ) 1   a0   v pv ( k ) 
voltage and the interpolated PV curve.
 5

By substituting (13) into Sylvester´s Formula, the factors a0, function. In a dc-dc stage of MPPT application, the objective
a1, and a2 can be found as in (18), (19), and (20), respectively. is the PV voltage and PV current. Since the desired current
These factors are updated in each sampling time in order to and voltage correspond to the same operating point on the PV
allow an accurate prediction for all regions of the PV curve. curve, and also to minimize the computational burden, only
They are also constantly updated since the whole PV curve the PV current is considered in the second cost function,
changes with the weather conditions. Another essential role of g 2   i pv ( k + 1) - i pv ( k + 1) * 
2
(17)
updating these factors will be revealed in the Dynamic 0 ,1   0 ,1 
weather conditions sub-section. where ipv(k+1) is estimated by using (4) and (5), and ipv*(k+1)
The predicted PV currents can be calculated for two states is provided by the first cost function (16). Due to the inclusion
using the following expression of only term in the cost function of the proposed method, no
i pv ( k + 1)1,2  i pv ( k )   i (14) weighting factors design is needed. The second cost function
is calculated for the two converter states, and the state that
Once the predicted PV currents are estimated for the two corresponds to the minimum cost function will be applied
states, the interpolated equation (12) in the next time horizon during the next sampling cycle.
can be used for the extrapolation of the PV currents for the
two states corresponding to these predicted voltages 2) Dynamic weather conditions
v pv ( k + 1)  a 2 i pv ( k + 1) 2  a1i pv ( k + 1)  a 0 (15) As it can be seen from Fig. 6, during fast solar irradiance
g1 1,2  ipv (k +1)1,2  vpv (k +1)1,2 :  : ipv (k )  vpv (k ) (16) change, each update instant could be from a different PV
 
curve. In this case, the interpolation does not emulate the
Equation (16) is used for the evaluation of the first cost model of a part of the PV curve. In fact, the interpolation
function, and the predicted PV current/voltage matching the reflects the path of the operating point movement from one PV
evaluated cost function will be chosen to be the PV curve to another. The resulted extrapolated line is shown in
current/voltage that needs to be applied in the next sampling blue dashed line, the solid blue line represents the path of the
instant. The selected PV current/voltage will be used for the operating point (the blue dashed line may not be seen in some
evaluation of the second cost function as explained in the next areas where it coincides with the solid one).
sub-section. If it is assumed that, the voltage reference is increasing and
the operating point is on the right side of the MPP point (point
B. Switching states generation A in Fig. 6), the predicted operating point would be B.
In this method, the switching states are generated by However, since the prediction on the new PV curve is out of
involving another model predictive control algorithm (FCS- the arced area, the measured power during the next sampling
MPC), which can be implemented without the need of a PI time would be much less than what was expected (point C).
controller or a modulator. Hence, no PI gains tuning effort is Hence, the predicted power is stored in the controller and
needed. Also, the steady-state operation is reached in labeled as the expected power (Pexp). During the next sampling
relatively a long time by using a PI controller. And decreasing time, the expected power Pexp is compared to the measured one
the response time between two successive references impairs PPV. If the difference between the measured and expected
the operation of the system under dynamic conditions [40]. On power exceeds a define threshold (ε), it implies that the
the contrary, the employment of FCS in the control of power operating point has just left the arced area of the PV curve, as
converters provides an excellent dynamic response [22], shown in Fig. 6. In this case, an opposite action to the one
which makes it advantageous for PV systems operating under applied during the previous sampling time should applied i.e.
rapidly changing atmospheric conditions. In the previous sub- if in the last sampling time a voltage/current increment has
section, the predictions were carried out by taking into been added, then, in the current sampling period it should be
consideration the PV characteristic or the path in which the subtracted, and vice-versa “iPV(k+1)*= iPV(k)*+iPV(k-1)*-
PV voltage and current are varying. But in FCS-MPC, the iPV(k-2)*”, as illustrated in Fig. 5. By adding this loop to the
predictions are performed by taking into account the model of algorithm, a tracking in the right direction is always fulfilled
the converter and the model of any object connected to that whether under increasing or decreasing irradiance.
converter, such as filter, grid, synchronous machine…etc. In The threshold ε must be greater than the oscillation of the
FCS, all the targeted objectives such as currents, flux, torque input power of the converter under the same duty ratio and the
and active and reactive power, are included in the cost estimated error of Lagrange polynomial extrapolation added

1
a0   v ( k - 2)  i pv 2 ( k - 1)i pv ( k )  i pv 2 ( k )i pv ( k - 1)   v pv ( k - 1)  i pv ( k - 2 )i pv 2 ( k )  i pv 2 ( k - 2)i pv ( k )  
  pv (18)
v pv ( k )  i pv 2 ( k - 2)i pv ( k - 1) - i pv ( k - 2 )i pv 2 ( k - 1)  
1
a1   v (k - 2)  i pv 2 (k )  i pv 2 (k - 1)   v pv ( k - 1)  i pv 2 ( k - 2)  i pv 2 ( k )   v pv ( k )  i pv 2 (k - 1)  i pv 2 (k - 2)   (19)
  pv 
1
a2   v ( k - 2 )  i pv ( k - 1)  i pv ( k )   v pv ( k - 1)  i pv ( k )  i pv ( k - 2 )   v pv ( k )  i pv ( k - 2 )  i pv ( k - 1)   (20)
where   pv 

  i pv 2 ( k - 2 )  i pv ( k - 1) - i pv ( k )   i pv ( k - 2 )  i pv 2 ( k ) - i pv 2 ( k - 1)   i pv 2 ( k - 1)i pv ( k )  i pv ( k - 1)i pv 2 ( k ) (21)

 6

PPV The predicted operating point

and
The operating point  vC 2  iR  vC 2   iR   con  i pv   v pv (25)
 i pv 
The MPP point  con  v pv
where conis the converter efficiency, and fsw is the switching
v
frequency. The ripple in the output current can be calculated
based on the following expression
Solar v 1 D (26)


D
Irradiance i  i  C 2 
R L
B f sw L
Pexp  PPV  
A C
The estimated error of the theorem of Lagrange
Pexp  PPV   extrapolation can be written as follows
R   x(k )  x(k  1)    x(k )  x( k  2)  ....  x( k )  x( k  n)  
(27)
PMPP ,21% f ( n 1) ( )
,    k  1, k  n 
PMPP ,20 % n!
The substitution of the PV data points used in this paper into
(27) yields to
vPV R   i pv ( k )  i pv ( k  1)    i pv ( k )  i pv ( k  2)  
(28)
Fig. 6. Extrapolation of the predicted PV current based on the predicted
v pv ( )
voltage and the interpolated path in case of solar irradiation change. The blue ,    k  1, k  2 
dash line represents the predicted PV power at the previous sampling time. 2

both together, which can be written in the following form IV. SIMULATION RESULTS
   P  R  v M PP (22)
A simulation analysis according to the schematic shown in
where Fig. 1 has been performed, were both the classical P&O and
 Ppv   v pv  i pv  v pv   i pv (23) the proposed control algorithm have been tested. A PI control-
The ripple in the input voltage and current can be calculated ler is adopted in this paper to minimize the error between the
by using the same equations used for the design of the provided reference by P&O and the PV current. The PV array
converter. In case of buck converter, the ripple in voltage and and MPPT parameters are shown in Table I. The main distur-
current can be calculated by using the following equations bance in the simulation test is a changing load, the system
starts first feeding an 8Ω resistive load, and then a sudden load
v pv 
iR
 ( D - D2 ) (24)
con  f sw  C1 change takes place, where the load increases to 4Ω.

Fig. 7. Simulation results when the conventional P&O is used showing the Fig. 8. Simulation results when the proposed controller is used showing the
case of change in the load “from 8Ω to 4Ω”, (a) the drown power from the PV case of change in the load “from 8Ω to 4Ω”, (a) the drown power from the PV
array, (b) the PV current and its reference and, (c) the duty cycle. array, (b) the PV current and its reference and, (c) the duty cycle.
 7

G.W/m
2
Sequence 1 Sequence 2 Sequence m where 05% refers to the efficiency of MPPT for a PV array
(n1 repetition) (n2 repetition) (nm repetition)
tTOP working under 5% of the solar irradiation in standard test
100%
50% conditions. At each irradiance level, the efficiency is
calculated based on the following expression
1 n

 (31)

 100 ...... m 2
  Ppv   T

2
 50..... m
..... m 2
.. m 2

H2
L2
Pav  TM i 1

sl op e
slope
 10..

L1  0.5

slop e Hm
slope Lm
where Ppv is the power drawn from the PV string, Pav is the
H1
slope
slope

30% available power in the PV array, TM is the total measurement

10%
Settling t Down
time, n is the number of periods, and ΔT is the sampling rate.
time The dynamic efficiency is calculated based on two successive
t s
series of a trapezoidal solar irradiance profiles. In the first
Fig. 9. The irradiance profiles used to assess the dynamic efficiency of the
MPPT, according to the standard EN50530. The blue and red colors indicate
series, the minimum and maximum of the trapezoidal profiles
to the insolation ranges and slopes of low to high solar irradiation test and are 100W/m2 and 500W/m2, respectively. And the ramps are
very low to medium irradiation test, respectively. varying from 0.5 W/m2/s in the first sequence (slopeL1) up to
50 W/m2/s in the last sequence (slopeLm) as shown in Fig. 9.
Whereas in the second series, the minimum and maximum of
The results shown in Fig. 7 and Fig. 8 correspond to P&O the trapezoidal profiles are 300W/m2 and 1000W/m2,
and the proposed MPC-MPPT controller, respectively. As it respectively. And the ramps are varying from 10 W/m2/s in the
can be seen from Fig. 7, at the instant 1.75s of the test, where first sequence (slopeH1) up to 100 W/m2/s in the last sequence
the load suddenly changes, the PI controller takes relatively a (slopeHm). In each repetition, the efficiency is calculated based
long time to adjust the new duty cycle. In this case, the PV on the following product
array was drifted to operate near the short circuit current (iSC) 1 (32)
n
point, which corresponds to approximately 24W. Moreover,  
D yn , i n
.. P  T  pv
the long response time caused by the PI controller, has led the P av  T j 1

P&O to make a wrong tracking direction, on account of the j 1

operating point is not in the neighborhood of the provided The dynamic efficiency corresponding to EN50530 standards
reference. In contrast, since the proposed controller is faster in is the average efficiency of all these repetitions
adjusting the duty cycle, the operating point barely moved 1 nm
from the MPP point during the step load change. One should  Dyn 
nm
 Dyn ,i
(33)
note, that the oscillation around the provided reference in the i 1
classical P&O has been significantly reduced in the proposed where nm is the total number of repetitions.
MPC-MPPT controller.
B. Experimental test bench
V. EXPERIMENTAL RESULTS
In order to verify the theoretical analysis, experimental tests
A. Test conditions have been carried out. Fig. 10 shows the experimental test
Different test types have been suggested in the literature for bench used for testing the proposed MPC-MPPT. The control
the evaluation of MPPT performances. The well-known test programs have been implemented in Matlab/Simulink, and by
composed of step irradiance changes. But this test does not using dSPACE real-time interface, they have been compiled
reflect all the possible weather conditions. Another test and uploaded to dSpace1103 controller board. The converter
consists of a random ramp profile, which emulates a moving used here is a 250-W, 35-V prototype buck converter, which
clouds also has been suggested. In 2006 a German has been designed to be installed on the back of a real PV
international working group suggested a standardized MPPT panel for withdrawing the local maximum power. The load
performance test. This test has been approved as a standard in was a resistive one (Rload), Rload has been computed in such a
the European Union and published as the Standard EN50530 way to guarantee a total dissipation greater than the largest
MPPT performance characterization by the end of 2009 [41]. PMPP to be evaluated. In this case, Rload has been selected to be
According to EN50530 standard, the performance of the 8Ω. Notice that, any converter topology that FCS has been
MPPT is assessed under both static and dynamic conditions. applied to in the literature, can be used here. Furthermore, this
The static test can be performed by running the system under system can be connected directly to a dc micro-grid, or to an
seven defined solar irradiance levels, for a duration of 10 min ac system through an inverter. Since the solar irradiance
in each level. The static efficiency can be calculated as
function of the European weighting factors by using the
following formula TABLE I
 EU  0.03  05%  0.06 10%  0.13  20% (29) SIMULATION AND EXPERIMENTAL IMPLEMENTATION PARAMETERS
PV parameters Value Other Parameters Value
.........  0.10  30%  0.48  50%  0.2 0 100%
Maximum power, PMPP 122W MPPT Frequency, fMPPT 10Hz
As well as by using California Energy Commission’s (CEC)
weighting factors Voltage at MPP, vMPP 24.8V Current increment, Δi 0.08A
 CEC  0.04 10%  0.05  20%  0.12  30% (30)
Open circuit voltage, vOC 31V Switching Frequency, fsw 30kHz
Short circuit current, iSC 5.1A Sampling time, Ts 30µs
...........  0.21  50%  0.53  75%  0.05 100%
 8

proposed scheme and P&O MPPT can be considered equal.

The advantage of the proposed method becomes prominent
during dynamic conditions, as shown in the next sub-section.

2) Dynamic tracking efficiency according to EN50530

standard: Fig. 12 and Fig. 14 show the response of both the
conventional P&O and the proposed scheme in term of PV
power and voltage during the complete EN50530 standard
test, respectively. The red represents the ideal variables,
“PMPP” and “vMPP”, whereas the black shows the measured
ones, “PPV” and “vPV”. It can be seen from these figures that
the voltage of the conventional P&O is close to the ideal
voltage vMPP under slow irradiance change, in both very low to
medium and low to high irradiance ranges. As a consequence,
the extracted PV power is close to its maximum. However, as
Fig. 10. The Experimental test setup used for testing the proposed MPC- the change in irradiance gets faster, P&O shows a
MPPT.
considerable drift issues, where the voltage goes much higher
and much lower than the ideal voltage, which implies that the
harvested power is less than the ideal available one. In contrast
profiles are trapezoidal, and with different slopes, a PV
the proposed method does not present any drift issue, and the
simulator was required. The used PV simulator was an Agilent
PV voltage is continuously close to vMPP along the entire
E4360A with two channels of up to 600-W (120-V, 5.1-A)
EN50530 standard test, which guarantees that the gathered
each. The PV simulator emulates the uploaded I-V curve of a
power is close to all the available in the PV array.
PV string with the specification under the STC shown in Table
The dynamic efficiency of the conventional P&O was
I. The PV curve has been uploaded to the PV simulator and
measured as 98.04%, which is close to the one reported in [7].
updated in case of irradiance changes by using Keysight
commands through Matlab. The dc-source in Fig. 10 is for Whereas the dynamic efficiency reached by the proposed
supplying the switching device gate driver of the converter. method was 99.01%. The dynamic efficiencies of both P&O
The MPPT parameters of both tested methods were the same and the proposed method in all sequences were varied from
for a fair comparison, they are shown in Table I. Their 99.10s to 98.80s, except in the last three ones in both ranges
optimization was according to the recommendations in [6]. (10%-50% and 30%-100%) as shown in Fig. 11. The average
Since the performance of FCS-MPC with respect to P&O dynamic efficiencies of P&O in these sequences in both
has been already deeply investigated in [32], and it has been ranges is 97.53%, 94.67%, and 88.45%, respectively. Whereas
shown that these two methods have equivalent performance, the proposed method reaches almost the same dynamic
the proposed MPPT is compared to P&O only. Furthermore, efficiency in these sequences as well, with an average of
P&O is the benchmark algorithm for MPPTs since it is the 98.86%. In this regard, two repetitions from this test were
most classical and adopted in industrial applications. selected to be shown in this paper in enlarged form.
Fig. 13 shows the response of P&O under the irradiance
C. Experimental Results profile corresponding to a repetition from the sequence before
the last one from low to high test (30%-100%), where the rate
1) Static tracking efficiency according to EN50530
of irradiance change was 50W/m2/s. It can be seen from this
standard: Table II shows the static efficiencies calculated
according to the European and California’s formulas, where
both the conventional P&O and the proposed MPC-MPPT are 100.0
100.0
considered. Normally, the efficiency of MPPT is calculated
with a resolution of two decimals [7]. This table shows that 98.0 98.0
the proposed scheme has an improvement in static efficiency
MPPT efficiency [%]
MPPT efficiency [%]

over P&O of 0.02% and 0.04% according to Euro and CEC, 96.0 96.0

respectively. These are imperceptible differences and are 94.0 94.0

within the measurement uncertainties–these results suggest
that in static conditions the tracking efficiencies of the 92.0 92.0

90.0 90.0
TABLE II
88.0 88.0
EXPERIMENTAL RESULTS OF THE STATIC MPPT EFFICIENCIES
0.1 1 10 100 1 10 100
UNDER EN50530 STANDARDS CONDITIONS (%) Ramp steepness [W/m2/s] Ramp steepness [W/m2/s]
 P&O Proposed MPC-MPPT (a) (b)
Euro 99.74 99.76 Fig. 11. Experimental results of the tracking efficiencies as function of
CEC 99.84 99.87 EN50530 standard’s ramps: (a) very low to medium irradiance range “10%-
50%”, (b) low to high irradiance range “30%-100%”. Where the blue
05% 98.75 98.77
represents P&O, and the red represents the proposed MPC-MPPT.
 9

P&O Proposed MPC-MPPT

Power [W]

Power [W]
(a) (a)
Voltage [V]

Voltage [V]
Time [s] Time [s]
(b) (b)

Fig. 12. (a) The harvested PV power of the conventional P&O during the Fig. 14. (a) The harvested PV power of the proposed method during the
complete EN50530 standard test, (b) the PV voltage variation of P&O MPPT complete EN50530 standard test, (b) the PV voltage variation of the proposed
during the complete EN 50530 standard test. method during the complete EN 50530 standard test.

figure that P&O diverged from the MPP several times. In this of 98.88%, which is improved compared to the conventional
situation, P&O diverges until the operating point gets out of P&O by 3.67% in this test.
the curved area of the P(v) characteristic, where the change in The second chosen repetition is corresponding to the last
power induced by the perturbation gets larger. The recorded sequence from very low to medium test (10%-50%). The
efficiency of P&O in this test was 95.21%. Fig. 15 shows the speed of the irradiance change in this sequence is 100W/m2/s
tracking performance when the proposed MPC-MPPT is (Fig. 16 and Fig. 18). It can be observed from Fig. 16 (a), that
applied. As expected, this method has the ability to provide a the conventional P&O is confused due to the fast increase in
current reference which continuously matches the operating irradiance. The ideal current iMPP has increased at the fourth
voltage with vMPP. The proposed scheme shows an efficiency second of this test, as a result P&O provided a decreasing

P&O Proposed MPC-MPPT

Iref [A]
Iref [A]

Time [s] Time [s]

(a) (a)
Voltage [V]
Voltage [V]

Power [W]
Power [W]

Time [s] Time [s] (c) Time [s] Time [s] (c)
(b) (b)

Fig. 13. A repetition from the sequence before the last one from low to high Fig. 15. A repetition from the sequence before the last one from low to high
(30%-100%) EN50530 standard test (50W/m2/s), with 5 seconds on top, when (30%-100%) EN50530 standard test (50W/m2/s), with 5 seconds on top, when
P&O is applied. (a) The current used as a reference, (b) the extracted PV the proposed control strategy is applied. (a) The current used as a reference,
power, (c) the PV voltage. (b) the extracted PV power, (c) the PV voltage.
 10

P&O Proposed MPC-MPPT

Iref [A]
Iref [A]

Time [s] Time [s]

(a) (a)
Voltage [V]

Voltage [V]
Power [W]

Power [W]
Time [s] Time [s] (c) Time [s] Time [s] (c)
(b) (b)
Fig. 16. A repetition from the last sequence from very low to medium (10%- Fig. 18. A repetition from the last sequence from very low to medium (10%-
50%) EN50530 standard test (100W/m2/s), with 5 seconds on top, when P&O 50%) EN50530 standard test (100W/m2/s), with 5 seconds on top, when the
is applied. (a) the current used as a reference, (b) the extracted PV power, (c) proposed MPC-MPPT is applied. (a) the current used as a reference, (b) the
the PV voltage. extracted PV power, (c) the PV voltage.

reference, which caused to a much higher PV voltage than Hence, the proposed MPC-MPPT has been also tested with
vMPP. Also, during the decrease of the irradiance, the P&O mismatched model parameters, where the range of modeling
reference has been confused, and stayed on the top of the errors was ±30%. Note that the system was operating under
ramp, while the ideal iMPP has started to decrease. Due to the the STC in this test. It can be seen from the results displayed
voltage drift during both the fast increase and fast decrease of in Fig. 17, that the effect of the underestimation of the load
the irradiance in this sequence, the efficiency of P&O barely resistor by 30% drops the efficiency to 99.01%, while the
reaches 88.33%. From Fig. 18, it can be seen that the proposed same mismatch of the inductor value worsens the efficiency to
approach is still robust even under such a fast irradiance 98.40%. One should note that, the effect of mismatched load
increase, in fact, it provides a non-confused reference, resistor is less than the effect of mismatched inductor.
conforming with the ideal one, and the harvested PV power From Fig. 17, it can be observed that the effect of -30%
was at its maximum during the whole profile. The efficiency mismatched inductor leads to a drop of the efficiency to
of the proposed MPC-MPPT in this test is 98.85%, which is 97.89%, whereas +30% mismatch exhibits a drop to 98.45%. It
improved over P&O by 10.52%. can be noted that, the mismatch of the inductor value, is
asymmetrical, i.e., the underestimation of this parameter has
3) Model parameter mismatch: more influence than its overestimation on the MPPT
One of the drawbacks of MPC schemes is the effect of model efficiency. And it is the case with the resistor as well.
parameters misestimation on the controller performance.

VI. CONCLUSION
Rload L Rload & L
100.0 An MPC-based MPPT for rapidly changing meteorological
conditions has been presented in this paper. The method
MPPT efficiency [%]

99.6
estimates the PV current/voltage that should be applied in
99.2
order to make the operating point converge to the MPP.
98.8
Moreover, it has the ability to detect whether the operating
98.4 point is still at PMPP, or it has been deviated e.g. due to a fast
98.0 change in the environmental conditions. The estimated PV
97.6 current/voltage serves as a reference to finite control set MPC,
-30 -20 -10 0 10 20 30 and the switching state that minimizes the difference between
Parameter mismatch [%]
this reference and the predicted variable is applied directly to
Fig. 17. Experimental results of the proposed approach under the STC in case the converter. The proposed method has been implemented
of model parameter mismatch, “Rload, and L”.
and compared to the conventional P&O according to EN50530
 11

in both static and dynamic conditions. The experimental Mode Control," IEEE Trans. Ind. Electron., vol. 60, no. 3, pp. 1168-
1178, March 2013.
results show that the proposed scheme offers an excellent [18] S. Vazquez, J. I. Leon, L. G. Franquelo, J. Rodriguez, H. A. Young, A.
dynamic performance with respect to P&O algorithm, Marquez, and P. Zanchetta, “Model predictive control. A review of its
providing a reference that matches the MPP locus even under applications in power electronics,” IEEE Ind. Electron. Mag., vol. 8, no.
1, pp. 16–31, Mar. 2014.
very fast environmental condition changes. [19] A. Linder, R. Kanchan, R. Kennel, and P. Stolze, "Model Based
Predictive Control of ElectricDrives," Germany: Cuvillier Verlag
VII. ACKNOWLEDGMENT Göttingen, 2010.
[20] W. Song, Y. Zhang, X. Sun, Q. Zhang, and W. Wang, “A study of
The authors acknowledge the support of the Danish Energy Zsource dual-bridge matrix converter immune to abnormal input voltage
Technology Development and Demonstration Program disturbance and with high voltage transfer ratio,” IEEE Trans. Ind. Info.,
(EUDP) through the project PVST – PV+STorage Operation vol. 9, no. 2, pp. 828–838, May 2013.
[21] P. Cortes, M. P. Kazmierkowski, R. M. Kennel, D. E. Quevedo, and J.
and Economics in distribution systems, project nr. 12,551. Rodriguez, “Predictive control in power electronics and drives,” IEEE
Trans. Ind. Electron., vol. 55, no. 12, pp. 4312–4324, Dec. 2008.
VIII. REFERENCES [22] J. Rodriguez, M. P. Kazmierkowski, J. R. Espinoza, P. Zanchetta, H.
Abu-Rub, H. A. Young, et al., "State of the Art of Finite Control Set
[1] Fernandes D, Almeida R, Guedes T, Sguarezi Filho AJ, Costa FF. "State Model Predictive Control in Power Electronics," IEEE Trans. Ind. Info,
feedback control for DC-photovoltaic systems," Electric Power Systems vol. 9, pp. 1003-1016, 2013.
Research. vol. 143, pp. 794-801. Feb. 2017. [23] R. Errouissi, S. M. Muyeen, A. Al-Durra, and Siyu Leng. "Experimental
[2] T. Esram and P. L. Chapman, "Comparison of photovoltaic array Validation of a Robust Continuous Nonlinear Model Predictive Control
maximum power point tracking techniques," IEEE Trans. Energy Based Grid-Interlinked Photovoltaic Inverter." IEEE Trans. Ind.
Convers., vol. 22, pp. 439-449, June 2007. Electron., vol, 63.no 7: pp. 4495-4505. July 2016.
[3] A. Lashab, A. Bouzid and H. Snani, "Comparative study of three MPPT [24] R. Erase, S. M. Muyeen, A. Al-Durra, and S. Leng. "A Robust
algorithms for a photovoltaic system control," 2015 World Congr. Info. Continuous-Time MPC of a DC–DC Boost Converter Interfaced With a
Technol. Comput. Appl. WCITCA 2015, no. 3, 2015. Grid-Connected Photovoltaic System." IEEE J. Photovolt., vol. 6, no. 6:
[4] J. Kivimäki, S. Kolesnik, M. Sitbon, T. Suntio and A. Kuperman, pp. 1619-1629, 2016.
"Revisited Perturbation Frequency Design Guideline for Direct Fixed- [25] M. B. Shadmand, X. Li, R. S. Balog and H. A. Rub, "Model predictive
Step Maximum Power Point Tracking Algorithms," IEEE Trans. Ind. control of grid-tied photovoltaic systems: Maximum power point
Electron, vol. 64, no. 6, pp. 4601-4609, June 2017. tracking and decoupled power control," First Workshop on Smart Grid
[5] F. Paz and M. Ordonez, "High-Performance Solar MPPT Using and Renewable Energy (SGRE), Doha, 2015, pp. 1-6.
Switching Ripple Identification Based on a Lock-In Amplifier," IEEE [26] M. Metry; M. B. Shadmand; R. S. Balog; H. A. Rub, "MPPT of
Trans. Ind. Electron., vol. 63, no. 6, pp. 3595-3604, June 2016. Photovoltaic Systems Using Sensorless Current-Based Model Predictive
[6] N. Femia, G. Petrone, G. Spagnuolo, M. Vitelli, "Optimization of Control,"IEEE Trans. Ind. App., vol. 53, no. 2, pp. 1157-1167, Ap 2017.
perturb and observe maximum power point tracking method," IEEE [27] O. Abdel-Rahim and H. Funato, "Model Predictive Control based
Trans. Power Electron. vol. 20, no. 4, pp. 963-973, July 2005. Maximum Power Point Tracking technique applied to Ultra Step-Up
[7] D. Sera, L. Mathe, T. Kerekes, S. V. Spataru, and R. Teodorescu, "On Boost Converter for PV applications," IEEE Innovative Smart Grid
the Perturb-and-Observe and Incremental Conductance MPPT Methods Technol., Asia (ISGT ASIA), Kuala Lumpur, 2014, pp. 138-142.
for PV Systems," IEEE J. Photovolt., vol. 3, no. 3, pp. 1070–1078, 2013. [28] M. B. Shadmand, M. Mosa, R. S. Balog and H. A. Rub, "An improved
[8] J. Ahmad, "A fractional open circuit voltage based maximum power MPPT technique for high gain DC-DC converter using model predictive
point tracker for photovoltaic arrays," 2nd Int.l Conf. Software Technol. control for photovoltaic applications," IEEE Applied Power Electro.
Engineering, San Juan, PR, 2010, pp. V1-247-V1-250. Conf. and Expo., APEC 2014, Fort Worth, pp. 2993-2999.
[9] M. Killi and S. Samanta, "Modified Perturb and Observe MPPT [29] M. Mosa, M. B. Shadmand, R. S. Balog and H. Abu Rub, "Efficient
Algorithm for Drift Avoidance in Photovoltaic Systems," IEEE Trans. maximum power point tracking using model predictive control for
Ind. Electron., vol. 62, no. 9, pp. 5549-5559, Sept. 2015. photovoltaic systems under dynamic weather condition," IET Renewable
[10] D. Sera, R. Teodorescu, J. Hantschel and M. Knoll, "Optimized Power Generation, vol. 11, no. 11, pp. 1401-1409, 9 13 2017.
Maximum Power Point Tracker for Fast-Changing Environmental [30] S. Sajadian, R. Ahmadi, "Model Predictive-Based Maximum Power
Conditions," IEEE Trans. Ind. Electron., vol. 55, no. 7, pp. 2629-2637, Point Tracking for Grid-Tied Photovoltaic Applications Using a Z-
July 2008. Source Inverter," IEEE Trans. Power Electron., vol. 31, no. 11, pp.
[11] J. Ahmed and Z. Salam, "A Modified P&O Maximum Power Point 7611–7619, Nov. 2016.
Tracking Method With Reduced Steady-State Oscillation and Improved [31] A. A. Abushaiba, S. M. M. Eshtaiwi and R. Ahmadi, "A new model
Tracking Efficiency," IEEE Trans. Sust. Energy, vol. 7, no. 4, pp. 1506- predictive based Maximum Power Point Tracking method for
1515, 2016. photovoltaic applications," 2016 IEEE Int. Conf. Electro Info. Technol.
[12] G. Escobar, S. Pettersson, C. N. M. Ho and R. Rico-Camacho, (EIT), Grand Forks, ND, pp. 0571-0575.
"Multisampling Maximum Power Point Tracker (MS-MPPT) to [32] A. Lashab, D. Sera, J. M. Guerrero, L. Mathe and A. Bouzid, "Discrete
Compensate Irradiance and Temperature Changes," IEEE Trans. Sust. Model-Predictive-Control-Based Maximum Power Point Tracking for
Energy, vol. 8, no. 3, pp. 1096-1105, July 2017. PV Systems: Overview and Evaluation," IEEE Trans. Power Electron.,
[13] S. K. Kollimalla and M. K. Mishra, “Variable perturbation size adaptive vol. 33, no. 8, pp. 7273-7287, Aug. 2018.
P&O MPPT algorithm for sudden changes in irradiance,” IEEE Trans. [33] P. S. Shenoy, M. Amaro, J. Morroni and D. Freeman, "Comparison of a
Sust. Energy, vol. 5, no. 3, pp. 718–728, Jul. 2014. Buck Converter and a Series Capacitor Buck Converter for High-
[14] H. A. Sher, A. F. Murtaza, A. Noman, K. E. Addoweesh, K. Al-Haddad Frequency, High-Conversion-Ratio Voltage Regulators," IEEE Trans.
and M. Chiaberge, “A New Sensorless Hybrid MPPT Algorithm Based Power Electron., vol. 31, no. 10, pp. 7006-7015, Oct. 2016.
on Fractional Short-Circuit Current Measurement and P&O MPPT,” [34] P. Karamanakos, T. Geyer and S. Manias, "Direct Voltage Control of
IEEE Trans. Sust. Energy, vol. 6(4), pp. 1426-1434, October 2015. DC–DC Boost Converters Using Enumeration-Based Model Predictive
[15] B. N. Alajmi, K. H. Ahmed, S. J. Finney and B. W. Williams, "A Control," IEEE Trans. Power Electron., vol. 29, no. 2, pp. 968-978, Feb.
Maximum Power Point Tracking Technique for Partially Shaded 2014.
Photovoltaic Systems in Microgrids," IEEE Trans. Ind. Electron., vol. [35] P. Cortes et al., “Guidelines for weighting factors design in model
60, no. 4, pp. 1596-1606, April 2013. predictive control of power converters and drives,” in Proc. IEEE Int.
[16] S. Messalti, A. G. Harrag and A. E. Loukriz, "A new neural networks Conf. Ind. Technol., Feb. 2009, pp. 1–7.
MPPT controller for PV systems," The Sixth Int. Renewable Energy [36] T. Dragicevic and M. Novak, "Weighting Factor Design in Model
Congrs (IREC), Sousse, 2015, pp. 1-6. Predictive Control of Power Electronic Converters: An Artificial Neural
[17] E. Bianconi, J. Calvente, R. Giral, E. Mamarelis, G. Petrone, C. A. Network Approach," IEEE Trans. Ind. Electron.. Accepted, doi:
Ramos-Paja, Giovanni Spagnuolo, Senior Member, IEEE, and Massimo 10.1109/TIE.2018.2875660
Vitelli "A Fast Current-Based MPPT Technique Employing Sliding
 12

[37] Z. Zhang, W. Tian, W. Xiong, and R. Kennel, “Predictive torque control

of induction machines fed by 3L-NPC converters with online weighting
factor adjustment using Fuzzy Logic,” in Proc. IEEE Transport.
Electrification Conf. Expo, Jun. 2017, pp. 84–89.
[38] Higham, N. J. (1988). "Fast Solution of Vandermonde-Like Systems
Involving Orthogonal Polynomials". IMA Journal of Numerical
Analysis. 8 (4): 473–486.
[39] Calvetti, D & Reichel, L (1993). "Fast Inversion of Vanderomnde-Like
Matrices Involving Orthogonal Polynomials". BIT. 33 (33): 473–
484. doi:10.1007/BF01990529.
[40] M. E. Ropp, S Gonzalez. “Development of a MATLAB/Simulink
modelof a single phase grid-connected photovoltaic system.” IEEE
Trans. Energy Convers., vol. 24, no. 1: pp. 195-202, 2009.
[41] H. Häberlin, P. Schärf, “New test procedure for Measuring Dynamic
MPP Tracking Efficiency at Grid connected PV inverters”, 24th
European Photovoltaic Solar Energy Conf., Germany, Sep 2009, pp.
3631-3637.
[42] H. Schmidt, B. Burger, U. Bussemas, & S. Elies, “How fast does an
MPP tracker really need to be?” 24th European Photovoltaic Solar
Energy Conference, Hamburg, Germany, Jan 2009, pp. 3273-3276.

Abderezak Lashab (S’13) received the bachelor’s

and master’s degrees in electrical engineering in
2010 and 2012, respectively, from Universié des
Frères Mentouri Constantine, Constantine, Algeria.
During the year 2013, he served as an engineer in
High Tech Systems (HTS).
He is currently working toward the Ph.D. degree
with the Department of Energy technology, Aalborg
University, Denmark. His current research interests
include control, modeling, and diagnostics of
photovoltaic power systems, and power electronics.

Dezso Sera (S’05–M’08–SM’15) received the B.Sc.

and M.Sc. degrees in electrical engineering from the
Technical University of Cluj, Cluj-Napoca,
Romania, in 2001 and 2002, respectively, the M.Sc.
degree in power electronics and the Ph.D. degree in
PV systems from the Department of Energy
Technology, Aalborg University, Aalborg,
Denmark, where he is currently an Associate
Professor. Since 2009, he has been Programme
Leader of the Photovoltaic Systems Research
Programme (www.pv-systems.et.aau.dk) at the same department.
His research interests include modeling, characterization, diagnostics and
maximum power point tracking (MPPT) of PV arrays, as well as power
electronics, and grid integration for PV systems.

Josep M. Guerrero (S’01–M’04–SM’08–F’15)

received the B.S. degree in telecommunications
engineering, the M.S. degree in electronics
engineering, and the Ph.D. degree in power
electronics from the Technical University of
Catalonia, Barcelona, Spain, in 1997, 2000, and
2003, respectively. Since 2011, he has been a Full
Professor with the Department of Energy
Technology, Aalborg University, Aalborg,
Denmark, where he is responsible for the
Microgrid Research Program. In 2012, he was a Guest Professor with the
Chinese Academy of Science and the Nanjing University of Aeronautics and
Astronautics; and in 2014, he was the Chair Professor with Shandong
University.
His research interests include different microgrid aspects, including power
electronics, distributed energy-storage systems, hierarchical and cooperative
control, energy management systems, and optimization of microgrids and
islanded minigrids.
Dr. Guerrero was awarded by Thomson Reuters as an ISI Highly Cited
Researcher