1

A Robust Continuous-Time MPC of a DC-DC

Boost Converter Interfaced with a Grid-Connected
Photovoltaic System
Rachid Errouissi, Member, IEEE, Ahmed Al-Durra, Senior Member, IEEE, and S. M. Muyeen, Senior
Member, IEEE,

Abstract—The main function of the dc-dc converter in a grid- of the switching devices [4]. Among a large number of the
connected photovoltaic system, is to regulate the terminal voltage existing dc-dc converters, boost converter has become the most
of the PV arrays to ensure delivering the maximum power to commonly used for feeding a grid-tied inverter. The real merit
the grid. The purpose of this paper is to design and practically
implement a robust continuous-time model predictive control of using a boost converter is its relatively simple topology that
(CTMPC) for a dc-dc boost converter, feeding a three-phase is lacking in others dc-dc converters such as a quadrature boost
inverter of a grid-connected PV system to regulate the PV converter, and interleaved boost converter [5], [6]. However,
output voltage. In CTMPC, the system behavior is predicted because of its limited efficiency, the boost converter is usually
based on Taylor series expansion, raising concerns about the adopted for low power single-phase system, which presents
prediction accuracy in the presence of parametric uncertainty
and unknown external disturbances. To overcome this drawback, a concern about the DC-link voltage ripples. Such a concern
a disturbance observer is designed and combined with CTMPC reveals the need for designing an appropriate control of the
to enhance the steady-state performance in the presence of model grid-tied single-phase inverter system, whilst the main focus
uncertainty and unknown disturbance such as the PV current, of this paper is mainly concerned with the control of the boost
which varies nonlinearly with the operating point. An interesting converter independently of the inverter type. On the other side,
feature is that the composite controller reduces to a conventional
PI controller plus a predictive term that allow to further improve a three-phase inverter can guarantee lower DC-link ripples
the dynamic performance over the whole operating range. The with a classical Proportional-Integral (PI) controller, and it can
effectiveness of the proposed controller was tested numerically also be used for some specific low power PV applications such
and validated experimentally with the consideration of the grid- as the household PV installed system. This partially explains
connected PV inverter system and its controller. why a three-phase grid-tied inverter is chosen, in this research
Index Terms—Continuous-time model predictive control work, to test the performance of the proposed controller for
(CTMPC), dc-dc boost converter, disturbance observer, grid- a dc-dc boost converter. However, all the results presented in
connected inverter system, photovoltaic system, PI observer, this paper can be expanded to single-phase grid-connected PV
renewable energy.
systems.
In the dc-dc boost converter control, a conventional cas-
I. I NTRODUCTION caded scheme is widely adopted because of its relatively
Ower electronic converters are essential to ensure efficient simple structure, which can ease the controller design and
P and reliable use of the PV power generation in either
grid-connected or stand-alone applications. In a grid-connected
the practical implementation. The cascaded control scheme
consists of a fast inner-loop, whose reference value, i.e., the
application, which is the focus of this work, a single/three- current reference, is provided by a slower outer-loop. The later
phase inverter and a dc-dc converter are usually utilized to in- is designed to control the terminal voltage of the PV array, and
terconnect the PV unit to its host grid via a DC-link capacitor. its reference value is usually determined by a maximum power
In addition, an input capacitor is normally placed between the point tracking (MPPT) algorithm [7].
dc-dc converter and the PV array to form the PV generator [1]. In the outer-loop control, a PI controller is well suited to
The main role of the inverter is to regulate the power exchange initiate a stable and accurate control, especially when the input
between the grid and the PV system, so as to comply with the capacitor is very big [8]. However, the transient performance
grid code. In such a topology, the active power is controlled by may be heavily influenced by the changes in the operating
regulating the DC-link voltage [2], while the reactive power point if the input capacitor is very small due to the so-
is maintained at a specified level, which is mainly dictated by called dynamic resistance [9]. More recent works have been
the grid connection requirement [3]. The dc-dc converter is dedicated to the influence of the dynamic resistance on the
considered to enable extracting the maximum available power PV voltage regulation, including those reported in [10]–[13],
from the PV generator by exploring the control capabilities and revealed that a PI controller alone may not be an adequate
choice to accurately control the boost converter for the whole
R. Errouissi, A. Al-Durra, and S. M. Muyeen are with the Petroleum operating range. Therefore, a PI controller combined with the
Institute, P.O. Box 2533, Abu Dhabi, United Arab Emirates (e-mails: dynamic resistance estimation might be a judicious solution to
rerrouissi@pi.ac.ae, aaldurra@pi.ac.ae, and smmuyeen@pi.ac.ae).
This work was supported by "The Petroleum Institute Research Center guarantee a good dynamic performance independently of the
(PIRC)" Research Grant. input capacitor size. An early attempt to include the estimate
 2

of the dynamic resistance into the PI controller, was through real-time implementation.
the use of an adaptive PI controller [12]. The basis for such In the inner-loop control, several approaches have been
a controller relies on estimating the dynamic resistance to proposed to achieve a fast transient response, including the
continuously adjust the controller parameters, so as to cancel finite-control-set model predictive control (FCS-MPC) [21],
the undesired effect caused by the PV current variation. In that sliding mode control (SMC) [22], [23], PI controller [24],
work, the proposed controller was applied to a single-phase etc. In this paper, a PI controller is derived for the inner-
system, and the inherent current and voltage ripples were used loop control using the same design process as for the outer-
to determine the dynamic resistance. However, this approach loop. Finally, the effectiveness of the proposed controller was
breeds other problems such as the limited estimation accuracy validated with the consideration of a grid-connected three-
in the presence of the measurement noise. Another drawback phase inverter, so as to take into account the real dynamics
of that method is its limited applicability to a single-phase of the whole system. PI controllers are designed to control the
system. A recent attempt to preserve the nominal tracking three-phase grid-tied inverter.
performance over the entire operating range was through the
use of a disturbance observer-based control (DOBC) [13]. In II. S YSTEM M ODELING
that work, good performances were obtained, but only a stand-
alone application was considered, and no realistic scenarios Figure 1 shows the schematic of a dc-dc boost converter
were performed under grid-connected PV system. tied to the ac bus via a DC-link capacitor C, a three-phase
In light of the aforementioned problems related to the PI inverter, and a line filter composed of an inductance L and
controller, this paper proposes a design methodology to derive a resistance R. The DC-link voltage vdc is kept constant by
a predictive PI controller for a dc-dc boost converter feeding controlling the switching devices S{1,6} . The main focus of
a grid-tied inverter. The whole derivation is based on com- this work is to regulate the PV output voltage v0 through the
bining a Model Predictive Control (MPC) and a Disturbance switching actions of the semiconductor device Sb . Here, Cb
Observer (DO). The MPC is essentially an optimal control and Lb , represent the input capacitor and the boost inductor,
that minimizes a quadratic cost function consisting of the respectively. For the boost converter, the measurable variables
difference between the system output and the trajectory to are the inductor current iL and the PV voltage v0 . The
be tracked over a finite time horizon. The proposed MPC, voltage vdc is seen as a known disturbance, whereas the
known as Continuous-Time MPC (CTMPC), uses Taylor series current ip is considered as an unknown disturbance that needs
expansion to predict the system behavior by considering the to be estimated and compensated in the control law.
nominal model [14]–[16]. The disturbance observer is intro-
duced to estimate the uncertain part, not considered in the ip i L Lb i0
D S1 S2 S3
nominal model, to improve the prediction accuracy [17]–[19].
+ + ia L R ea
In this work, the major role of the disturbance observer is to va
v0 vdc ib eb
estimate the PV current instead of the dynamic resistance. It PV
Cb Sb C
vb
ic ec
turns out that the composite controller, consisting of CTMPC − − S4 S5 S6 vc
and DO, reduces to a PI controller plus a predictive term
that has the role of improving the tracking performance for a Boost converter Grid-connected three phase inverter
smooth reference signal. The design process and the additional
predictive part represent the essential difference between the Fig. 1. Schematic diagram of a PV three-phase conversion system.
proposed method and the existing adaptive PI controllers. An Assuming that the converter operates in a continuous con-
adequate choice of the reference signal permits to have a good duction mode, then, the dynamics of the boost converter
dynamic response by exploring the tracking performance ca- depicted in Fig. 1 can be described by the following set of
pabilities of the predictive controller, while the integral action differential equations
eliminates the steady-state error. Another advantage lies in the
di v v v

estimation of the PV current instead of the dynamic resistance.  L = 0 − dc + dc d

Such an estimation can be directly used for other algorithms, dt Lb Lb Lb (1)
such as MPPT technique, which avoids the need for additional dv0 iL ip

 =− +
filter for the current measurement. Similar design process is dt Cb Cb
recently adopted in [20] to develop an accurate control of a where d represent the duty-cycle control. In the cascaded
solar energy conversion system consisting of a dc-dc boost scheme, the inner current control loop provides the duty-
converter and a grid-tied inverter. However, that work focuses cycle d, which is realized by means of a fixed frequency PWM.
only on the control of a grid-tied inverter without treating the Making use of (1), the current controller can be designed based
control of the dc-dc boost converter, and only a PI controller on the linear model of the current equation given by
is used to regulate the PV output voltage. Another difference
is that the proposed disturbance observer reduces to a PI diL
= Ai iL + Bi ui + Fi bi , ui = (d − 1) . (2)
controller in this work, while the work presented in [20] dt
considers the integration of the system model to estimate where
the lumped disturbances due to the existence of unmatched vdc 1
disturbances, raising concern about the relative complexity of Ai = 0, Bi = , Fi = , bi = v0 + δi (3)
Lb Lb
 3

The term δi is added to the model to represent the lumped Hence, (10) can be simplified as
disturbances caused by model uncertainty. In the outer voltage
loop, the current iL is treated as a control input. Thus, the e (t + Tr ) = Π(Tr ) (H (y) − Gu − M b) (12)
current reference, for the inner-loop, can be determined from
the outer-loop voltage control based on the following linear where
model    T
Π= 1 Tr , H (y) = e (t) ẏr (t) − Ay (13)
dv0
= Av v0 + Bv uv + Fv bv , uv = iL . (4)
dt The column matrices G and M are given by
where    
1 1 0 0
Av = 0, Bv = − , Fv = , bv = i p + δ v (5) G= , M= (14)
Cb Cb B F
The term δv represents parameter variations and external Invoking (13)–(14), and replacing e(t + Tr ) in (8) by its
disturbances. In order to simplify the controller design, it is expression given by (12), the approximate cost function =
assumed that can be expressed as follows
lim δ̇i (t) = 0, lim δ̇v (t) = 0 (6) T
t→∞ t→∞ = = (H (y) − Gu − M b) Υ (Tr ) (H (y) − Gu − M b)
(15)
III. ROBUST C ONTINUOUS -T IME M ODEL P REDICTIVE where Υ (Tr ) is a 2 × 2 matrix and is determined as follows
C ONTROL (CTMPC)
A. Baseline Controller: Formulation of CTMPC Υ (Tr ) = ΠT (Tr ) Π (Tr ) (16)
Consider a mathematical model for a single-input-single-
output (SISO) disturbed linear system The derivative of the cost function with respect to the control
input is given by
ẏ = Ay + Bu + F b (7)
d=  
where u ∈ R, y ∈ R, and b ∈ R, are the input, the output, = −B Υ1 Υ2 H (y) + BΥ2 Bu + BΥ2 F b (17)
du
and the disturbance, respectively. The continuous-time MPC
is essentially an optimal control that results from minimizing where Υ1 = Υ2,1 , and Υ2 = Υ2,2 . Thus, considering the
a quadratic cost function defined by matrix Υ(Tr ), it can be shown that
2 2
= = [e (t + Tr )] = [yr (t + Tr ) − y (t + Tr )] (8) Υ1 = Tr , Υ2 = Tr2 (18)
where, yr represents the output reference, e(t) is the tracking Making use of d=
= 0, the optimal control is given by
du
error, and Tr is known as predictive time. In the continuous-
time MPC formulation, the control input is not usually in- u (t) =B −1 Υ−1

Υ1 Υ2 H (y) − Υ−1


2 2 Υ2 F b
cluded in the cost function to simplify the stability analysis. In (19)
=B −1
 

K 1 H (y) − F b
such conditions, the control effort can be restricted by tuning
the predictive time Tr or/and limiting the set-point changes. where the controller gain K is expressed as follows
The optimal control is derived based on the optimality condi-
tion given by 1
d= K= (20)
=0 (9) Tr
du
Following [18], the design methodology of a continuous- Substituting (19) in (11) gives the closed-loop system error
time MPC is based on approximating the future tracking equation as follows
error e(t + Tr ) with the use of Taylor series expansion up ė + Ke = 0 (21)
to (ρ + r)th order, with r denotes the control order and ρ is
the relative degree of the system. The main role of the control Therefore, since the predictive time is positive, it is clear
order is to ensure the stability of the closed-loop system for that the closed-loop system, under the continuous-time MPC,
systems having high relative degree [25]. However, for the is asymptotically stable. For real-time implementation, the
system under investigation, it is clear that the relative degree disturbance is not always available for measurement. For the
is equal to 1 for both loops. That is why, the control order r purpose of making the proposed controller more convenient
is set equal to be zero in this work. Hence, an approximate for practical implementation, the continuous-time MPC can
of e(t + Tr ) is given by be modified as
 
e (t + Tr ) = e (t) + Tr ė (t) u (t) = B −1

(10) K 1 H (y) − F b̂
  (22)
Making use of (7), from the definition of the relative degree ρ, = B −1 Ke (t) + ẏ r − Ay − F b̂
it follows that
ė (t) = ẏr (t) − ẏ = ẏr (t) − Ay − Bu − F b (11) where b̂ is an estimate of the real disturbance b.
 4

B. Composite Controller: CTMPC and a Disturbance Ob- methodology is based on Taylor series expansion. For power
server converter applications, the predictive time is mainly decided
Considering the measurement of the output y, an observer by the switching frequency [27], which dictates the nominal
can be derived to estimate the disturbance as follows [26], performance specification of the settling time defined by (21).
The observer gain is selected to correspond to the desired
˙
 
b̂ = µ ẏ − Ay − Bu − F b̂ (23) settling time under the composite controller by considering
the transfer function (28).
where µ is the observer gain. Combining (7) with (23), gives Remark 2: It it noticed that if the disturbance b is composed
the dynamics of the disturbance observer as follows of a measurable variable bm and an unknown component bu ,
ėb = −µF eb − ḃ (24) with b = bm + bu , the composite controller can be modified
as
where eb = b̂ − b is the disturbance estimation error. It is  
u (t) = B −1 Ke (t) + ẏ r − Ay − F bm − F b̂u (29)
evident that the disturbance observer can be made stable by
choosing the parameter µ so as to have µF > 0. This means
where b̂u is simply computed by the PI observer given by (26).
that the estimation error can be made bounded and its bound
Now, assuming that b̂u (0) = −µe (0), and substituting (26)
depends on ḃ. Hence, with the assumption that lim ḃ = 0, it
t→∞ into (29) gives the predictive PI controller as follows
is clear that the observer converges to the actual disturbance
as t → ∞. According to (24), a large observer gain µ Zt
results in a fast disturbance estimation, but it may magnify the u (t) = Pb e (t) + Ib e (τ ) dτ + Nb (t) (30)
measurement noises. Hence, attention should be given when 0
selecting the observer gain for practical implementation. The where Pb = B (K + F µ), and Ib = B −1 F Kµ are the
−1
major drawback of the observer (23) is that it includes the proportional and the integral gains of the PI controller, respec-
time derivative of the output. To tackle the need for ẏ, the tively. The predictive term Nb (t) = B −1 (ẏ r − Ay − F bm ) has
disturbance observer can be further simplified by substituting the role of predicting the error between the system output y
the control law (22) into (23). In doing so, we obtain and the trajectory to be tracked yr . Such a term is not usually
˙ considered in the classical PI controller, which makes the pro-
b̂ = −µKe (t) − µė (t) (25)
posed controller superior in terms of the tracking performance,
By integrating the above equation, one can simplify the particularly, when dealing with a smooth reference.
disturbance observer as follows
Zt C. Application to the Boost Converter
b̂ (t) = −µK e (τ ) dτ − µe (t) + µe (0) + b̂ (0) (26) The composite controller, consisting of a continuous-time
0 MPC and a disturbance observer, is applied to the dc-dc boost
converter by means of the conventional cascaded scheme.
Hence, as pointed out in [18], selecting b̂ (0) = −µe (0) For the inner-loop, the composite controller is applied to the
allows recovering approximately the nominal performance, current dynamics (2) to determine the duty-cycle d = (ui + 1)
defined by (21), in the absence of disturbances. For instance, minimizing the cost function (8), with y = iL , and yr = iLref ,
substituting (22) into (11) with the consideration of (26) gives where iLref is the current reference. Invoking (30), the inner-
the output tracking error dynamics as follows loop controller can be expressed as
Zt 
  Zt

ė (t) + (K + F µ) e (t) + F µK e (τ ) dτ = −F b + F ξ (0) L  µi µi Ki 1 
ui = Ki + ei (t) + ei (τ ) dτ − v0
vdc Lb Lb Lb
0 0
(27) (31)
where ξ (0) = b̂ (0) + µe (0). Therefore, by neglecting the where ei = iLref − iL is the current tracking error. Under a
disturbance variation, the reference-to-output transfer func- cascaded structure, the current reference iLref is generated by
tion P (s), for a constant set-point, can be expressed by the outer-loop control. This can be accomplished by applying
the predictive PI controller (30) to the equation (4) to find the
(K + F µ) s + F µK
P (s) = (28) optimal input uv = iLref minimizing the cost function (8),
s2 + (K + F µ) s + F µK with y = v0 and yr = v0ref , where v0ref is the desired
The poles associated with (28) are s1 = −K and s2 = −F µ. voltage reference. In doing so, we get
This implies that the closed-loop system is asymptotically 
  Zt

stable, since the predictive time Tr is positive and the observer µ v µ v K v
uv = −Cb  Kv + ev (t) + ev (τ ) dτ + v̇0ref 
gain µ satisfies µF > 0, as mentioned above. According Cb Cb
0
to (28), it is clear that the disturbance observer has also an (32)
impact on the dynamic performance. where ev = v0ref − v0 is the PV voltage tracking error.
Remark 1: To facilitate the design process, the predictive Remark 3: It should be noted that the time derivative of the
time Tr can be considered as the first design parameter, current is not included in the current loop control to avoid
which must be selected as small as possible, since the control magnification of the measurement noise, since the current
 5

reference iLref = uv is based on the voltage measurement. ratio and ωni,v represents the natural angular frequency. The
As a result, the composite controller for the inner-loop reduces typical value of ζ is equal to √12 , while ωni,v can be selected
to a PI controller with a feed-forward term that compensates 4
according to the desired settling time approximated by ζωni,v ,
the variation of both the DC-link and the PV voltages. Note with the consideration of the switching frequency [30]. It is
that the value of Vdc is updated in the inner-loop control. noticed that, in the cascaded scheme, the outer-loop should be
Remark 4: For the voltage regulation, the composite con- designed to have slower response than that of the inner-loop.
troller includes the time derivative of the reference, which To demonstrate the effectiveness of the proposed approach,
allows improving the tracking performance in comparison simulation tests are carried out using MATLAB/SIMULINK
to a conventional PI controller. Therefore, a filtered voltage with the use of a single diode PV panel model, developed
reference can be used, instead of a step input, to take advantage in [31]. Under standard conditions, the PV panel can operate
of the tracking capability of the proposed controlled, while at at a maximum power point (MPP) of 1 kW, and its I–V
the same time, to limit the inductor current during the transient. characteristic curve is plotted in Fig. 3. The switching
frequency for the dc-dc boost converter is selected to
v0ref iLref Comparator be fsc = 12.5 kHz, while that for the three-phase inverter is
ui + d +
v0 Eq : 32 iL Eq : 31 Sb set to be fsi = 6.25 kHz. The simulation tests were performed
+ - with a control period of 80 µsec. The time step for the
1 Carrier complete developed model is set equal to 1 µsec.
vdc
Signal
(a)
ed 10
Kpv + Kpi + + vd∗
vdcref + idref + +
PV output current ip (A)
K 8
- - -
Kiv 1/s + Kii 1/s + MPP
vdc Lωiq 6
ia,b,c id +
Kpi vq∗
iqref + + 4
ea,b,c abc → dq
θ
PLL i -
q Kii 1/s + +
2
Lωid
(b) 0
0 40 80 120 160
PV output voltage v0 (V)
Fig. 2. Control schemes for (a) dc-dc converter (b) grid-tied inverter,
2vdcref
with K = , and Kpi = 14.2419, Kii = 7.4570×103 ,
3ed Fig. 3. Characteristic I–V curve of the PV array under standard conditions,
Kpv = 0.1403, and Kiv = 7.0133. with IM P P = 7.75 A, VM P P = 129 V, and PM P P = 1 kW.

IV. C OMPUTER S IMULATIONS

The parameters of the proposed controller for the dc-dc
A. Control Loop Diagram boost converter can be determined based on the nominal
Figure 2 shows the control block diagram for both the specifications of the settling time tsc ≈ 4Tr , defined by (21),
dc-dc boost converter and the three-phase grid-connected by considering the minimum switching frequency fsc of the
inverter. The control scheme for a three-phase grid-connected semiconductor device. That is, the predictive time Tr , which is
inverter, reported in the literature, has traditionally a cascaded the first design consideration, for the inner-loop and the outer-
structure, consisting of PI controllers. The outer-loop regulates loop is set to be 0.2 ms and 2 ms, respectively, to achieve
the DC-link voltage by considering the d-axis current id a fast and stable control under cascaded structure. For the
as a control input, while the inner-loop uses the voltage observer gains, selecting µi = 0.1 modifies the settling time
components vd and vq to control the currents id and iq . The of the closed-loop current control to be equal to nine times
voltage commands vd∗ and vq∗ are then converted to three-phase the switching period Tsc = 1/fsc , which is fast enough [27].
voltage commands va∗ , vb∗ , and vc∗ , which can be realized The observer gain µv should be chosen as large as possible,
with PWM techniques. The q-axis current iq is generally while maintaining the response of the voltage control slower
maintained equal to zero to achieve unity power factor than that of the current control to ensure the stability of the
operation. The components in the synchronous rotational cascaded control scheme. Thereby, the observer gain µv is set
frame (d, q) are obtained via abc − dq transformation with to be 0.5, so that the settling time of the outer-loop becomes
the use of a phase-locked loop (PLL) algorithm to generate equal to five times that of the inner-loop. Moreover, the voltage
the reference angle so as to maintain eq = 0, where eq reference is realized by a first-order linear filter, with a time
denotes the q-axis grid voltage [28]. This means that the constant equal to Tr = 2 ms, to avoid overshoot that can
d-axis grid voltage ed will be aligned with the grid voltage be caused by the integral action in response to a step input.
vector. Following [29], the coefficients Kpi and Kii of Such a strategy allows limiting the inductor current during
the PI controller for the inner-loop can be designed with the transient, and eliminating the steady-state error without
2
Kpi = 2ζLωni −R, and Kii = Lωni , respectively, while those scarifying the nominal tracking performance. The parameter
for the outer-loop can be determined with Kpv = 2ζCωnv , values of the complete developed system are given in the
2
and Kiv = Cωnv , respectively, where ζ denotes the damping Appendix.
 6

B. Tracking Performance Under Maximal Power Point Figure. 4 shows that the PV output voltage tracked its
First test was concerned with the reference tracking per- reference with zero steady-state error in spite of unknown
formance evaluation with a step change in the PV voltage PV current ip , which varies nonlinearly with the operating
reference from 158 to 130 V. The PV voltage of 130 V has point. As can be seen in Fig. 5, the estimate bv followed
been selected to extract the maximum power from the PV the inductor current iL with an error which tends to zero as
panel. Such a test is equivalent to a change in the active time goes to infinity, indicating that the disturbance observer
power P delivered to the grid, with P : 0 → 1kW . is asymptotically stable, since the current iL is equal to
the unknown component ip in the steady-state regime. As
180
illustrated in Fig. 6, the reactive power Q is maintained equal
to zero, while the active power P follows its command of 1 kW
with a steady-state error resulting from the inverter losses.
PV output voltage (V)

160
PV voltage
Reference
140

C. Tracking Performance Under a Time-Varying Reference

120
This test was performed with step changes in the PV output
100 voltage to verify the effectiveness of the proposed controller
0.8 0.85 0.9 0.95 1
Time (s) throughout the whole operating range. That is, downward steps
of the PV voltage were realized from 158V, near to the open-
Fig. 4. PV output voltage response. circuit voltage, to 120 V, below the MPP voltage, as v0ref =
158 → 145 → 135 → 120 V. Also, upward steps were realized
10
as v0ref = 120 → 135 → 145 → 158 V.
Current iL (A) and esimate bv (A)

5
iL 180

bv
PV output voltage (V)

160
0
PV voltage
Reference
140
−5
0.8 0.85 0.9 0.95 1
Time (s) 120

(a) Inductor current and the disturbance estimation.

100
2 0.8 1.1 1.7 2.3 2.9
Time (s)

(a) PV output voltage response

Estimation error ebv

1
10
Current iL (A) and esimate bv (A)

0
5
iL
bv
−1
0.8 0.85 0.9 0.95 1 0
Time (s)

(b) Estimation error ebv = iL − bv

−5
Fig. 5. Inductor current response to a step change in v0 . 0.8 1.1 1.7 2.3 2.9
Time (s)

(b) Inductor current and the disturbance estimation bv

Fig. 7. Tracking performance with downward steps of v0 .

Active/Reactive Power (W)

900

600
According to Figs. 7(a) and 8(a), the proposed controller
P allows achieving good transient and steady-state performances
Q
300 independently of the operating points. More interestingly,
similar dynamic performance can be observed over the entire
0 operating range. The inductor current iL , shown in Figs. 7(b)
−200 and 8(b), exhibits a good dynamic performance without a
0.8 0.85 0.9 0.95 1
Time (s) significant overshoot as the PV voltage changed due to the
filtered PV voltage reference. Similarly to the previous test,
Fig. 6. Active and reactive powers delivered to the grid in response to a step the estimate bv closely followed the inductor current iL .
change in v0
 7

180 found that the PV output voltage regulation, with µv = 0.5,

is unstable. Such a difference between the simulation and
experimental test is possibly explained by the fact that µv = 0.5
PV output voltage (V)

160
causes the settling time of the voltage control to approach that
140 of the current control, which certainly affects the stability of
PV voltage the cascaded control scheme. To address such a concern, the
Reference
120 observer gain µv is set to be 0.1, so that the settling time of
the outer-loop is equal to twelve times that of the inner-loop.
100
1 1.2 1.8 2.4 3
Time (s)
B. Tracking Performance Under Maximal Power Point
(a) PV output voltage response
This experimental test was performed to validate the first
10
Current iL (A) and esimate bv (A)

simulation results, where a step change in the PV voltage

was applied to the dc-dc converter to ensure delivering the
5 maximum power to the grid. The experimental results are
presented in Figs. 10 and 11, showing that the PV voltage
regulation performance is almost similar to that obtained with
0
i
simulation test. However, it can be observed that the inductor
L
bv
current response is evidently slower than that obtained with
the simulation test. Such a difference between simulation and
−5
1 1.2 1.8 2.4 3 experimental results can be possibly explained by the inherent
Time (s)
dynamics of the PV emulator to a rapid change in PV voltage.
(b) Inductor current and the disturbance estimation bv

Fig. 8. Tracking performance with upward steps of v0 .

V. E XPERIMENTAL RESULTS
A. Experimental Setup
Experimental tests were conducted to validate the proposed
controller with the consideration of realistic scenarios by con-
necting the output of the dc-dc converter to a grid-connected
inverter as shown in Fig. 9.

Fig. 10. Boost converter’s response under a step input of PV voltage: v0 (20
V/div), iL (5 A/div), bv (5 A/div), and ebv (5 A/div).

Fig. 9. Laboratory setup for testing the proposed controller

The experimental setup consists of a PV emulator including
the input capacitor Cb , a dc-dc boost converter, a DC-link
capacitor, a three-phase inverter and an L type filter to smooth
the current injected into the grid. A XR160-12 power supply
module, manufactured by Magna-Power Electronics, was used
to produce the PV power according to the I-V characteristic
shown in Fig 3. Semiteach power electronics module (AN-
8005), manufactured by Semikron, was utilized to construct
a six-pulse IGBT inverter and a dc-dc converter. The PI con-
troller for the grid-side inverter and the predictive controller for
the dc-dc converter were realized using the dSPACE ds1103
DSP board. The hardware realization employs the previously
given values of the controller parameters, the switching fre-
quencies, and the control period. However, in practice, it has Fig. 11. DC-link voltage and active/reactive power delivered to the grid under
a step input of PV voltage: vdc (40 V/div), P (250 W/div), and Q(250 W/div).
 8

Also, it can be seen that the estimate bv tracked accurately test, the inductor current response is slow in comparison with
the inductor current iL and the estimation error ebv converges the simulation test because of the limited dynamics of the PV
to zero at the steady-state regime. For the grid-tied inverter emulator.
control, it is clear that that the DC-link voltage and the
reactive power Q are well controlled. Here, P represents
the active transferred to the grid. The unity power factor
operation is guaranteed as shown in Fig. 12. It is noted that
the modulating signals m∗a , m∗b , and m∗c are generated using
third harmonic injection technique, so as to to prevent over-
modulation problem.

Fig. 14. Performance evaluation under upward steps of v0 : v0 (20 V/div),

iL (5 A/div), and bv (5 A/div).

D. Comparison Between CTMPC and the PI Controller

This experiment is concerned with the performance compar-
ison of the proposed approach and the classical PI controller,
Fig. 12. Grid current ia , grid voltage ea , the voltage vab at the output of for the PV output voltage regulation, while maintaining the
the inverter, and the reference of the modulating signal m∗a using the third same inner-loop control to have a fair comparison. Follow-
harmonic injection technique: ia (10 A/div), ea (40 V/div), vab (120 V/div),
and m∗a (0.5 V/div). ing [23], the coefficients Pb and Ib of the PI controller
2
can be designed as Pb = 2ζCb ωnb , and Ib = Cb ωnb ,
with ζ = 0.7. The natural angular frequency ωnb is set to
be equal to 661 rad/s, so that the resulting settling time is
the same as that obtained with CTMPC, i.e., equal to twelve
times that of the inner-loop. Such a design process enables us
to perform a fair comparison.

Fig. 13. Performance evaluation under downward steps of v0 : v0 (20 V/div),

iL (5 A/div), and bv (5 A/div).

C. Tracking Performance Under Different Operating Points

This experiment was performed to experimentally test the
tracking performance with the continuous-time MPC in re-
Fig. 15. Voltage and current waveforms in response to upward steps of v0
sponse to downward and upward steps of the PV voltage under PI controller, and with nominal value of Cb : v0 (20 V/div), iL (5 A/div).
reference over the entire operating range. Experimental results
shown in Figs. 13 and 14 , confirmed the simulation results for Fig. 15 gives the voltage and current waveforms in response
good transient and steady-state performances throughout the to downward steps of the PV output voltage under the PI
whole operating range. Also, it can be seen that the distur- controller, designed with the nominal value of Cb . Figs. 16
bance observer responds well, with the estimate bv following and 17 compare the performance of the proposed CTMPC with
accurately the inductor current iL . Similarly to the previous the PI controller under model uncertainty. More specifically,
 9

the value of Cb is set incorrectly in the controllers, and it existing semiconductor devices are now capable of operating
is chosen to be equal to 25% of its nominal value. As seen, at a high switching frequency without significant switching
under nominal value of Cb , the PI controller allows achieving losses, especially for low power applications. The efficiency
a good transient response of the boost converter over the of the dc-dc boost converter can be further improved if the
whole operating range, but a degraded performance is observed power converter module, consisting of the inverter and the
under model uncertainty as shown in Fig. 16. As illustrated in boost converter, operates at its nominal power of 20 kVA. This
Fig. 17, the proposed design process is proved to be effective is because the power converter is usually designed to function
regarding model uncertainty in comparison with the classical at high efficiency when it operates at its nominal power, while
PI controller design. the maximum power of the PV emulator is limited to 1.6 kW.

VI. C ONCLUSION

In this paper, a robust continuous-time model predictive

control has been proposed with the aim of controlling a
dc-dc boost converter feeding a grid-connected three-phase
inverter by using a cascaded control scheme, as found in
many PV applications. The use of continuous-time MPC
allows achieving a good tracking performance in response to
a smooth reference, whereas the disturbance observer permits
to eliminate the steady-state error caused by parametric uncer-
tainty and unknown PV current. Therefore, the composite con-
troller offers excellent transient and steady-state performances
throughout the whole operating range. Moreover, the design
process was comprehensively described. A PI controller was
also designed to control the grid-tied inverter for performance
Fig. 16. Voltage and current waveforms in response to upward steps of v0
under PI controller, and with incorrect value of Cb : v0 (20 V/div), iL (5 A/div). testing of the proposed controller with the consideration of the
real dynamics.
Both simulation and experimental results demonstrated the
effective control of the dc-dc boost converter to initiate a
stable and accurate steady-state regime, while providing a
good dynamic performance over the entire operating range.

A PPENDIX A
PARAMETERS OF THE C OMPLETE S OLAR E NERGY
C ONVERSION S YSTEM U NDER S TUDY

The parameters of the grid-tied inverter and the boost

converter are vdc = 165 V, L = 6.8 mH, R = 0.1 Ω,
C = 1.052 mF, ω = 314.15 rad/s, Cb = 0.16 mF, and
Lb = 5 mH. The line-to-line grid voltage is equal to 70 V.

A PPENDIX B
Fig. 17. Voltage and current waveforms in response to upward steps of v0
under CTMPC, and with incorrect value of Cb : v0 (20 V/div), iL (5 A/div), M ODELLING AND PARAMETERS OF THE P HOTOVOLTAIC
and bv (5 A/div). A RRAY

E. Evaluation of the Efficiency of the Developed System An equivalent circuit of the PV cell, known as a single-diode
model, including the series and parallel resistances, is shown
This test was conducted to investigate the switching fre-
in Fig. 18. But, the practical equation that usually adopted
quency effect on the efficiency η of the dc-dc boost converter.
to describe the I–V characteristic of a PV array is given as
Such an objective is achieved by evaluating the efficiency η,
follows
around the maximum power point, for different values of the
switching frequency fsc . From the results, it was found that, in
   
V + Rs I V + Rs I
the frequency range fsc : 3.125 → 12.5 kHz, the efficiency η I = Ipv − I0 exp −1 − (33)
Vt a Rp
remains almost the same and is about 94 %, indicating that the
change in the switching frequency does not have a prominent where Ipv , I0 and Vt are the PV current, the saturation
effect on the efficiency. The possible reason is that the dc- current, and the thermal voltage, respectively, of the array.
dc boost converter uses only one switching device and the Such variables are expressed as functions of the operating
 10

conditions as follows [4] P. Sharma and V. Agarwal, “Maximum power extraction from a partially
shaded PV array using shunt-series compensation,” IEEE J. Photovolt.,
G


 Ipv = (Ipv,n + KI (T − Tn )) vol. 4, no. 4, pp. 1128–1137, July 2014.

 Gn [5] Z. Chen, P. Yang, G. Zhou, J. Xu, and Z. Chen, “Variable duty cycle
(Isc,n + KI (T − Tn ))

control for quadratic boost PFC converter,” IEEE Trans. Ind Electron.,

I0 = (34) vol. PP, no. 99, pp. 1–1, 2016.

 exp ((Voc,n + Kv (T − Tn ))/aVt ) − 1 [6] O. Lopez-Santos, L. Martinez-Salamero, G. Garcia, H. Valderrama-
 N KT
 Vt = s Blavi, and T. Sierra-Polanco, “Robust sliding-mode control design for


q a voltage regulated quadratic boost converter,” IEEE Trans. Power
Electron., vol. 30, no. 4, pp. 2313–2327, April 2015.
where T and G are the actual temperature and irradiation. [7] D. Sera, L. Mathe, T. Kerekes, S. V. Spataru, and R. Teodorescu, “On
Here, the temperature measurement is expressed in Kelvin. the perturb-and-observe and incremental conductance MPPT methods
for PV systems,” IEEE J. Photovolt., vol. 3, no. 3, pp. 1070–1078, July
The rest of the parameters are given in the Table I. 2013.
[8] M. Villalva, T. de Siqueira, and E. Ruppert, “Voltage regulation of
Ideal PV Cell I photovoltaic arrays: small-signal analysis and control design,” IET Power
Electron., vol. 3, no. 6, pp. 869–880, 2010.
Rs [9] J. Thongpron, K. Kirtikara, and C. Jivacate, “A method for the deter-
Ipv Id Rp V mination of dynamic resistance of photovoltaic modules under illumi-
nation,” Sol. Energy Mater. Sol. Cells, vol. 90, no. 18, pp. 3078–3084,
2006.
[10] L. Nousiainen, J. Puukko, A. Mäki, T. Messo, J. Huusari, J. Jokipii,
J. Viinamäki, D. Lobera, S. Valkealahti, and T. Suntio, “Photovoltaic
Fig. 18. Equivalent circuit of a single-diode model of a PV cell generator as an input source for power electronic converters,” IEEE
Trans. Power Electron., vol. 28, no. 6, pp. 3028–3038, 2013.
[11] A. Urtasun and D. Lu, “Control of a single-switch two-input buck
TABLE I converter for MPPT of two PV strings,” IEEE Trans. Ind. Electron.,
PARAMETERS OF THE SIMULATED PV ARRAY AT Tn = 25 ◦ C AND vol. 62, no. 11, pp. 7051–7060, 2015.
Gn = 1000 W/m2 [12] A. Urtasun, P. Sanchis, and L. Marroyo, “Adaptive voltage control of
the dc/dc boost stage in PV converters with small input capacitor,” IEEE
Series Resistance Rs 0.221 Ω Trans. Power Electron., vol. 28, no. 11, pp. 5038–5048, 2013.
Parallel Resistance Rp 415.405 Ω [13] M. Sitbon, S. Schacham, and A. Kuperman, “Disturbance observer-based
Nominal light-generated current Ipv,n 8.214 A voltage regulation of current-mode-boost-converter-interfaced photo-
Nominal short-circuit current Isc,n 8.21 A voltaic generator,” IEEE Trans. Ind. Electron., vol. 62, no. 9, pp. 5776–
Short-circuit current/temperature coefficient KI 0.0032 A/K 5785, 2015.
Open-circuit voltage/temperature coefficient Kv -0.1230 V/K [14] M. Soroush and C. Kravaris, “A continuous-time formulation of nonlin-
Boltzman constant K 1.3806503×10−23 J/K ear model predictive control,” Int. J. Control, vol. 63, no. 1, pp. 121–146,
Nominal temperature in Kelvin Tn 25 ◦ C 1996.
Nominal irradiation Gn 1000 W/m2 [15] P. J. Gawthrop, H. Demircioglu, and I. I. Siller-Alcala, “Multivariable
Electron charge q 1.60217646×10−19 C continuous-time generalised predictive control: A state-space approach
Diode ideality constant a 1.3 to linear and nonlinear systems,” in IEE Proc. Control Theory Appl.,
Nominal open-circuit voltage Voc,n 32.9 V vol. 145, no. 3. IET, 1998, pp. 241–250.
Number of cells connected in series Ns 54
[16] W.-H. Chen, D. J. Ballance, P. J. Gawthrop, J. J. Gribble, and J. O’Reilly,
“Nonlinear PID predictive controller,” IEE Proc. Control Theory Appl.,
In practical applications, the PV system consists of connect- vol. 146, no. 6, pp. 603–611, 1999.
[17] J. Yang and W. X. Zheng, “Offset-free nonlinear MPC for mismatched
ing Nm PV arrays in series to increase the voltage, and Np PV disturbance attenuation with application to a static var compensator,”
arrays in parallel to increase the current. In such conditions, IEEE Trans. Circuits Sys. II: Express Briefs, vol. 61, no. 1, pp. 49–53,
the practical equation (33) becomes Jan 2014.
    [18] J. Yang, W. Zheng, S. Li, B. Wu, and M. Cheng, “Design of a prediction
V + Rseq I V + Rseq I accuracy enhanced continuous-time MPC for disturbed systems via a
I = Np Ipv − Np I0 exp −1 − disturbance observer,” IEEE Trans. Ind. Electron., vol. 62, no. 9, pp.
Nm Vt a Rpeq 5807–5816, 2015.
(35) [19] R. Errouissi, M. Ouhrouche, W.-H. Chen, and A. M. Trzynadlowski,
where Rseq , and Rpeq are the equivalent resistances, and are “Robust cascaded nonlinear predictive control of a permanent magnet
determined as follows synchronous motor with antiwindup compensator,” IEEE Trans. Ind.
Electron., vol. 59, no. 8, pp. 3078–3088, 2012.
Nm Nm [20] R. Errouissi, S. M. Muyeen, A. Al-Durra, and S. Leng, “Experimental
Rseq = Rs ; Rpeq = Rp (36) validation of robust continuous nonlinear model predictive control based
Np Np
grid-interlinked photovoltaic inverter,” IEEE Trans. Ind. Electron., DOI:
For the purpose of generating the I–V characteristic plotted in 10.1109/TIE.2015.2508920, 2015.
[21] P. E. Kakosimos, A. G. Kladas, and S. N. Manias, “Fast photovoltaic-
Fig. 3, it was found that Nm = 4.9 and Np = 1.02. system voltage-or current-oriented mppt employing a predictive digital
current-controlled converter,” IEEE Trans. Ind. Electron., vol. 60, no. 12,
R EFERENCES pp. 5673–5685, 2013.
[22] E. Mamarelis, G. Petrone, and G. Spagnuolo, “An hybrid digital-analog
[1] J. Viinamäki, J. Jokipii, T. Messo, T. Suntio, M. Sitbon, and A. Ku- sliding mode controller for photovoltaic applications,” IEEE Trans. Ind.
perman, “Comprehensive dynamic analysis of photovoltaic generator Info., vol. 9, no. 2, pp. 1094–1103, May 2013.
interfacing dc–dc boost power stage,” IET Renew. Power Gen., vol. 9, [23] E. Bianconi, J. Calvente, R. Giral, E. Mamarelis, G. Petrone, C. Ramos-
no. 4, pp. 306–314, 2015. Paja, G. Spagnuolo, and M. Vitelli, “A fast current-based MPPT tech-
[2] M. Mahmud, H. Pota, and M. Hossain, “Dynamic stability of three- nique employing sliding mode control,” IEEE Trans. Ind. Electron.,
phase grid-connected photovoltaic system using zero dynamic design vol. 60, no. 3, pp. 1168–1178, 2013.
approach,” IEEE J. Photovolt., vol. 2, no. 4, pp. 564–571, Oct 2012. [24] M. Goncalves Wanzeller, R. Alves, J. da Fonseca Neto, and W. Fonseca,
[3] C. Y. Tang, Y. T. Chen, and Y. M. Chen, “PV power system with multi- “Current control loop for tracking of maximum power point supplied for
mode operation and low-voltage ride-through capability,” IEEE Trans. photovoltaic array,” IEEE Trans. Instrum. Measurement, vol. 53, no. 4,
Ind Electron., vol. 62, no. 12, pp. 7524–7533, Dec 2015. pp. 1304–1310, 2004.
 11

[25] W.-H. Chen, D. J. Ballance, and P. J. Gawthrop, “Optimal control of S. M. Muyeen (S’03-M’08-SM’12) received his
nonlinear systems: a predictive control approach,” Automatica, vol. 39, B.Sc. Eng. Degree from Rajshahi University of
no. 4, pp. 633–641, 2003. Engineering and Technology (RUET), Bangladesh
[26] Y. I. Son, I. H. Kim, D. S. Choi, and H. Shim, “Robust cascade control formerly known as Rajshahi Institute of Technology,
of electric motor drives using dual reduced-order PI observer,” IEEE in 2000 and M. Sc. Eng. and Dr. Eng. Degrees
Trans. Ind. Electron., vol. 62, no. 6, pp. 3672–3682, 2015. from Kitami Institute of Technology, Japan, in 2005
[27] D. R. Espinoza-Trejo, E. Barcenas-Barcenas, D. U. Campos-Delgado, and 2008, respectively, all in Electrical and Elec-
and C. H. De Angelo, “Voltage-oriented input–output linearization tronic Engineering. His PhD research work focused
controller as maximum power point tracking technique for photovoltaic on wind farm stabilization from the viewpoint of
systems,” IEEE Trans. Ind. Electron., vol. 62, no. 6, pp. 3499–3507, LVRT and frequency fluctuation. After completing
2015. his Ph.D. program he worked as a Postdoctoral
[28] S.-K. Chung, “A phase tracking system for three phase utility interface Research Fellow under the versatile banner of Japan Society for the Promotion
inverters,” IEEE Trans. Power Electron., vol. 15, no. 3, pp. 431–438, of Science (JSPS) from 2008-2010 at the Kitami Institute of Technology,
2000. Japan. At the present, he is working as Associate Professor in Electrical
[29] R. Kadri, J.-P. Gaubert, and G. Champenois, “An improved maximum Engineering Department at the Petroleum Institute, Abu Dhabi. His research
power point tracking for photovoltaic grid-connected inverter based on interests are power system stability and control, electrical machine, FACTS,
voltage-oriented control,” IEEE Trans. Ind. Electron., vol. 58, no. 1, pp. energy storage system (ESS), Renewable Energy, and HVDC system. He
66–75, 2011. has been a Keynote Speaker and an Invited Speaker at many international
[30] P. Verdelho and G. Marques, “DC voltage control and stability analysis conferences, workshops, and universities. He has published over 150 articles
of PWM-voltage-type reversible rectifiers,” IEEE Trans. Ind. Electron., in different journals and international conferences. He has published five books
vol. 45, no. 2, pp. 263–273, 1998. as an author or editor. Dr. Muyeen is the senior member of IEEE.
[31] M. Villalva, J. Gazoli, and E. Filho, “Comprehensive approach to
modeling and simulation of photovoltaic arrays,” IEEE Trans. Power
Electron., vol. 24, no. 5, pp. 1198–1208, 2009.

Rachid Errouissi (M’15) received the B.Sc. degree

in electronics from the Faculty of Sciences and
Technology of Mohammedia, Morocco, in 1998; the
M.Sc. degree in power electronics from École Mo-
hammadia d’ingénieurs in Rabat, Morocco, in 2001;
the double M.Sc. degrees in electrical engineering
and in automation and system engineering from Uni-
versity Claude Bernard, Lyon, France, in 2002 and
2004; and the Ph.D. degree in electrical engineering
from the University of Quebec, Chicoutimi, QC,
Canada, in 2010. From 2011 to 2014, he worked
as a Postdoctoral Researcher with the Department of Electrical and Computer
Engineering, University of New Brunswick. Since 2014, he has been with
the Petroleum Institute, Abu Dhabi, UAE, where he is currently conducting
research works in renewable energy and advanced control systems. His area
of interests include advanced control, nonlinear control, electric machines
and drives, and renewable energy conversion systems. He is a registered
Professional Engineer in the province of New Brunswick, Canada.

Ahmed Al-Durra (S’07-M’10-SM14) received the

B.S., M.S., and PhD in Electrical and Computer
Engineering from the Ohio State University in 2005,
2007, and 2010, respectively. For his M. Sc. degree,
he investigated the application of several nonlinear
control techniques on automotive traction PEM fuel
cell systems. He conducted his PhD research at the
Center for Automotive Research in the Ohio State
University on the applications of modern estimation
and control theories to automotive propulsion sys-
tems. At the present, he is an Associate Professor in
the Electrical Engineering Department at the Petroleum Institute, Abu Dhabi,
UAE. His research interests are application of estimation and control theory
in power system stability, Micro and Smart Grids, renewable energy, and
process control. He has published over 70 scientific articles in Journals and
International Conferences. Dr. Ahmed has successfully accomplished several
research projects at international and national levels. He is the co-founder of
Renewable Energy Laboratory at the Petroleum Institute, and Senior Member
in IEEE.