This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

Cascade-Free Model Predictive Control for

Single-Phase Grid-Connected Power Converters
Pablo Acuna, Member, IEEE, Ricardo P. Aguilera,Member, IEEE, Amer M.Y.M. Ghias, Member, IEEE,
Marco Rivera, Member, IEEE, Carlos R. Baier, Member, IEEE, and Vassilios G. Agelidis, Fellow, IEEE

Abstract—In a conventional FCS-MPC formulation, ac- width modulation (PWM) control methods, such as voltage-
tive and reactive power control loops rely on the predictive oriented control (VOC), the design procedure is 100% related
controller while the dc-bus voltage is usually governed by to tuning proportional-integral (PI) controllers in both dc-
a PI-based control loop. This comes from fact that the
dynamic equations for describing the predictions of these bus voltage and grid-current loops [4]–[7]. The other widely
variables are heavily coupled. In this paper, a cascade- accepted method is direct power control (DPC), which uses
free finite control set model predictive control (FCS-MPC) hysteresis comparators and look-up tables. Recently, a model
for single-phase grid-connected power converters is pre- based adaptive DPC [8] for three-phase grid-connected power
sented. The proposed control algorithm is formulated in converters has been proposed that avoids system parameter
terms of established dynamic references design, which was
originally proposed to directly govern active and reactive uncertainties, while improving the overall behavior of the
power, and dc-voltage in three-phase power converters. system compared with both standard PI and hysteresis-based
In this work, the dynamic reference design concept is controllers.
extended to control single-phase grid-connected power For the purpose of explicitly considering the converter
converters. The proposed control algorithm does not use switching effects, the so-called Finite-Control-Set Model Pre-
instantaneous ac-power calculations; instead it directly for-
mulates the optimal control problem on the grid-current in dictive Control (FCS-MPC) has been introduced to govern
the original stationary reference frame. The experimental power converters [9]. The main advantage of FCS-MPC is
results obtained with a single-phase grid-connected Neu- that offers the possibility to manage various control objec-
tral Point Clamped (NPC) converter confirm a successful tives in only one cost function and, at the same time, can
design, where system constraints, e.g. maximum power handle system constraints with ease. Although more than
and weighted switching frequency, are easily taken into
account. one control objective can be used, traditional MPC-based
control algorithms either for grid-connected and electrical-
Index Terms—Predictive models, control design, predic- drives power-converters use the classical cascaded structure
tive control, finite control set, converters, DC-AC power
of an outer PI based dc-bus voltage/speed loop which delivers
converters, smart grids, power quality, reactive power.
the power(current)/torque(current) reference to an inner MPC
active-reactive power/torque(current) loop. As a consequence,
I. I NTRODUCTION the overall system performance is still influenced by the linear
controller-based outer loop, which may present poor dynamic
INGLE-PHASE grid-connected power converters play a
S key role in applications such as integration of renewable
energies, and particularly high-power applications, such as
or overshoot in the outer loop [10].
A model predictive direct speed control (MP-DSC) which
overcomes limitations of cascade linear controllers, was pre-
high-speed railway electrical traction systems [1]. The pop- sented in [11]. This control is mainly based on the use of
ularity of these converters, technically called Active Front an attraction region defined by the MTPA (Maximum Torque
End (AFE) rectifiers, lies in their ability to allow bidirectional per Amp) trajectory, which is used to give importance to
power exchange between ac and dc output terminals while secondary control objectives during transient conditions. In
maintaining high power quality in terms of Power Factor (PF) the same direction, a cascade-free predictive speed control
and Total Harmonics Distortion (THD) on the ac-side [2], [3]. for electrical drives, that uses a simplified equivalent double
The main control objectives for single-phase grid-connected integrator model for representing the dynamic behavior of the
power-converters are: to self-support the dc-bus voltage under motor was proposed in [12]. In that case, it was proved that the
load and grid-voltage variations, and to draw sinusoidal grid- proposed equivalent model was able to successfully represent
current while maintaining the reactive power level around its the expected machine behavior. For grid-connected power
reference. The available literature shows that for classic pulse converters, a soft-constrained MPC formulated as a single
optimization problem was proposed in [13] to avoid the use of
Manuscript received January 30, 2016; revised May 27, 2016 and PI controllers. However, there is no fixed design procedure in
June 28, 2016; accepted June 29, 2016.
P. Acuna and V. G. Agelidis are with The University of New South the soft-constrained MPC to set the upper and lower bounds,
Wales, Australia. (pablo.acuna@unsw.edu.au). and the weighting factor of each control variable included in
R. P. Aguilera is with The University of Technology Sydney (UTS), the cost function. In the same direction, a three-phase AFE
Australia. (raguilera@ieee.org).
A. Ghias is with University of Sharjah, United Arab Emirates. rectifier governed by a cascade-free MPC-based Direct Power
M. Rivera and C. Baier are with Universidad de Talca, Chile. Control (MPC-DPC) was presented in [14]. The major advan-

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

DC-Side Single-Phase Converter AC-Side

tage of this strategy, called dynamic references design, is its ca-
pability to provide suitable references for the grid active power Grid-Filter
1. Power source:
and the dc-bus voltage, while maintaining active and reactive i.e. Renewable Energy and Circuit
Back to Back Converter. Topology
power decoupled. The concept have been described in detail

Grid
2. Electrical Load:
i.e. Energy storages, (see
in [14], however their applicability has not been demonstrated Traction System Fig. 2)
for single-phase AFE rectifiers. In essence, [14] formulates the and so forth.

cost function for a three-phase system in terms of the three-

phase instantaneous active/reactive power definitions. In that Fig. 1. Generalized circuit diagram of a single-phase grid-connected
case, the definition of the three-phase instantaneous predicted power-converter as interface to power sources or electrical loads.
powers is directly considered for designing both the dynamic
references and the cost function. Since the predicted powers
idc i r1
are instantaneous powers in the time domain which does not
iC1 S1 S3
have oscillations (mainly free of second harmonic), they can
be directly used on the cost function formulation. However, +
vC1
for the particular case of single-phase grid-connected power − C1 S2 S4
r L ig
converters, the instantaneous power calculation is not free of + inp
vdc Load +vab − vg
second harmonic. Thus, a straightforward implementation of − iC2
[14] on single-phase systems is not possible. +
S1 S3
vC2
The work at hand presents the design, implementation and − C2
evaluation of a cascade-free FCS-MPC strategy for single- S2 S4
phase grid-connected power-converters that address the issues
i r2
outlined above. The key novelty of this approach is the
compatible reference design, which allows the controller to
Fig. 2. Circuit topology of the single-phase grid-connected NPC con-
transfer the required dc-power from the grid while regulating verter.
the dc-voltage and imposing a desired power factor by only
control the grid-current. To accomplish this, the dynamic
reference design concept which was originally proposed to variables grid-current, ig , and the dc-voltage, vdc , but also the
govern three-phase power converters [14] is adopted and internal capacitor voltages vC1 and vC2 .
extended to single-phase converters. Nevertheless, the optimal By analyzing the schematic circuit shown in Fig. 2 it
control problem is directly formulated on the grid-current is possible to describe the continuous-time dynamic of the
in the original stationary reference frame. As an illustrative injected grid-current, ig , via:
example, the proposed method is used to govern a single-phase
NPC converter working as an AFE rectifier. dig 1
= (vg − rig − vab ), (1)
Additionally, two different cost function formulations to dt L
evaluate the effectivenesses of the proposed algorithm in terms where vg stands for the grid-voltage and vab represents the
of obtaining a spread and imposed grid-current harmonic single-phase inverter voltage. On the other hand, the dynamic
spectrum are adopted [15]. equations for the capacitor voltages are given by:
dvC1 1 1
II. S INGLE -P HASE G RID -C ONNECTED = iC = (ir − idc ) ,
dt C1 1 C1 1
P OWER -C ONVERTER M ODEL (2)
dvC2 1 1
= iC = (−ir2 − idc ) .
In general, single-phase grid-connected power-converters dt C2 2 C2
can be modeled as a four-port power module where its
terminals describe the respective ac-side (grid) and dc-side where iC1 and iC2 are the associated capacitor currents, ir1
connections, as shown in Fig. 1. Depending on the topology and ir2 are internal currents of the NPC converter and idc is
of the internal circuit, the ac-side terminals can behave as the dc-side current.
a multilevel-converter that fulfills the needs for low total Thus, (1) can be transformed into discrete-time form by
harmonic distortion (THD) with reduced dv using the well known classical forward Euler method that
dt . On the other
hand, the dc-side terminals can behave either as an electrical exhibits enough precision to enable the controller to predict
load or power source. In order to control all the required power the future value of the grid-current, i.e.,:
converter variables, a control block external to the converter  
k+1 Ts r Ts

enforces the desired behavior. ig = 1− ig k + vg k − vab k , (3)
L L
The work at hand focuses on the control of a single-phase
converter which is connected to the grid by means of a L where Ts is the sampling period, and the converter output
filter (r represents the parasitic resistor of L). The dc-side is voltage vab can be expressed in terms of switching states (see
connected to a dynamic load as shown in Fig. 2. As a study Table I) as follows:
case, a single-phase NPC converter is considered in this work.
Thus, it is required to control not only the continuous-time vab k = vC1 k (S1 − S3 ) + vC2 k (S2 − S4 ) . (4)

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

Therefore, based on the MPC requirements, the present A. Cost Function Formulation
values of ig , vg , vC1 and vC2 are needed to properly forecast In general, a single control loop over the grid-current
the future behavior of ig . injected by the converter is enough to control the active and
Since the regulation of vC1 and vC2 is also a control reactive power in the grid-side. However, for this particular
objective, the discrete-time form of (2) is obtained following converter, it is also required to regulate the dc-voltage, vdc ,
the same procedure as used for (3), i.e.,: while keeping the capacitor voltage balance, i.e., vC1 = vC2 =
vdc
Ts k Ts k 2 . It is important to emphasize the fact that both variables
vC1 k+1 = ir − idc + vC1 k , are heavily coupled, thus a compatible reference design is
C1 1 C1
(5) necessary in order to include both control targets (grid-current
k+1 Ts Ts k
vC2 = − i r2 k − idc + vC2 k . tracking and dc-voltage regulation) into the MPC grid-current
C2 C2
reference. Therefore, the proposed cost function for the single-
To avoid unnecessary measurements of ir1 and ir2 it is phase grid-connected NPC converter is:
convenient to define the direct relation between the switching
2
states of the NPC converter and ig as [16]: J k = i⋆g k+1 − ig k+1 +
2
SA (SA + 1) − SB (SB + 1) k ig k+1

2
i r1 k = ig ,
2 vC1 − vC2 k+1 , (7)
2 (6)
vdc
2
SA (SA − 1) − SB (SB − 1) k
i r2 k = ig . where ig and v2dc are rated values used to normalize the
2
Consequently, (3) and (5) can be used by the controller to tracking errors. Notice that the dc-voltage control loop is
obtain estimated values for the predictions of the grid-current not explicitly included in (7). This is due to the proposed
and the capacitor voltages respectively. controller that includes a static expression for the dc-voltage
reference embedded in the computation of the optimal state
Sxopt , ∀ x ∈ {1, 2, 3, 4}. For that reason, the design of the grid
III. MPC FOR A S INGLE -P HASE G RID -C ONNECTED current reference i⋆g k+1 which is directly related with the dc-
P OWER C ONVERTER ⋆ k+1
voltage reference vdc is explained in detail in the following
In this section, the proposed cascade-free FCS-MPC algo- subsections.
rithm for a single-phase grid-connected power converter is
derived. The procedure can be divided in three main parts: B. Stationary Reference Frame Grid-Current Definition
cost function formulation, stationary reference frame grid-
Clearly, the dc-side power can be obtained as follows:
current definition and grid-current dynamic references design.
First, the control targets, grid-current tracking in the stationary pdc k = pC1 k + pC2 k + pZdc k ,
reference frame and voltage balancing, are set into the cost (8)
= ir1 k vC1 k − ir2 k vC2 k ,
function. Then, a definition of the grid-current in the stationary
reference frame by means of the required active and reactive where ir1 and ir1 are taken from (6).
power is also derived. The resulting expression along with Regarding the ac-side, the instantaneous active (pk ) and
the dynamic references design concept is used to define the reactive (q k ) power can be calculated by [17]:
grid-current reference in the stationary reference frame and 1 k k

pk = vd id + vqk ikq ,
the dc-voltage references. 2 (9)
k 1 k k

q = vq id − vdk ikq ,
2
TABLE I where v and i are any pair of voltage and current in a given
S WITCHING S TATES∗ OF THE S INGLE - PHASE F ULL - BRIDGE NPC
C ONVERTER common point expressed in their dq components. In this work,
the synchronous frame is aligned to the grid-voltage, vg ,
is considered as the direct component. Thus, the quadrature
j S1 S2 S3 S4 SA SB vab component of the grid-voltage is null, i.e., vg d = v̂g and
0 1 1 0 0 1 -1 vC 1 + vC 2 vg q = 0. Therefore, the grid-side active and reactive power
1 1 1 0 1 1 0 vC 1
can be expressed as follows:
1

2 0 1 0 0 0 -1 vC 2 pg k = vg kd ig kd , (10)
2
3 0 1 0 1 0 0 0 1  
4 1 1 1 1 1 1 0
qg k = − vg kd ig kq . (11)
2
5 0 0 0 0 -1 -1 0 Finally, by utilizing (10) and (11), the expression for the
6 0 1 1 1 0 1 −vC 1 grid-current in the stationary reference frame is obtained by
multiplying each dq current component of ig k by sin (ωt) and
7 0 0 0 1 -1 0 −vC 2
cos (ωt), i.e.,:
8 0 0 1 1 -1 1 −vC 1 − vC 2
2pg k −2qg k
∗
: only selected states. ig k = sin (ωt) + cos (ωt). (12)
vg kd vg kd

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

k+1
From (12), it is possible to confirm that by governing between two consecutive steps, redc can be estimated by
the injected grid-current, ig , the active and reactive power considering the measurement of idc k as:
injected to or absorbed from the grid can be handled. Note vC1 k + vC2 k
k+1
that this step is only an intermediate step that aims to find redc = . (17)
an equivalent mathematical expression for the grid-current. idc k
As will be elucidated, the dynamic references design concept After following the above procedure, an average dc-power
will allow to define suitable references for active and reactive reference pe⋆dc k+1 can be defined as:
power in order to find the equivalent one-step ahead grid-  
pe⋆dc k+1 = ei⋆dc k+1 + ei⋆C1 k+1 veC ⋆ k+1
current reference that matches with the form of (12).  
1
(18)
+ eidc
⋆ k+1
+ eiC2
⋆ k+1
veC2 k+1 .
⋆

C. Grid-Current Dynamic Reference Design

Proceeding accordingly and recalling that grid-connected
A power balance on both power converter’s terminals is single-phase converters present a large second-order harmonic
used to obtain a compatible reference for both dc and ac component in the dc-side variables, the latter pe⋆dc k+1 is then fil-
control targets. However, it is not possible to instantaneously tered with a notch-filter1 centered at 100 Hz (2f ).
Furthermore,
compare both ac- and dc-power. For instance, when SA =SB , −2f
the use of the notch-filter output as a reference pe⋆dc k+1
the rectified currents ir1 and ir2 in (6) becomes zero, yielding provides a current reference i⋆g k+1 free of third harmonic in
to a null dc-power, i.e., pdc = 0, while the ac-power may the stationary frame.
not have a value equal to zero, i.e., pg 6= 0. For this reason To account for the filter inductor loss it is necessary to
a so-called average dc-power reference, pe⋆dc , and average ac estimate the resistor power loss. Thus, it follows that:
active power reference, pe⋆g , are analytically derived to produce
1  2
an average power balance in the power converter. Firstly, it pe⋆k+1
r = r îg k+1
follows that 2
2r  ⋆ k+1
2
2  (19)
pe⋆g k+1 = per k+1 + pe⋆dc k+1 , (13) = 2 peg + qg⋆ k+1
(v̂g k+1 )
where per stands for the filter resistor power loss.
Since pe⋆dc cannot be directly obtained as per (8), it is Here, v̂g k+1 ≈ v̂gk and îg k+1 are the peak values of grid-
necessary to use average values for each term in (8). It is voltage and current respectively.
important to recall that each capacitor voltage can only be Now, it is possible to transfer power from the ac-side by
adjusted by their capacitor current, which takes energy from considering (13), (18) and (19). Thus, the total power balance
the grid. To achieve the power balance while maintaining (13) becomes a quadratic equation in peg k+1 (a2 + bx + c = 0)
a desired behavior in the capacitor voltages, a reference as:
prediction horizon N ⋆ is introduced [14]. The value of N ⋆ 2r

⋆ k+1 2

−2f

2 peg − pe⋆g k+1 + pe⋆dc k+1 +
allows the controller to take enough power from the grid k
v̂g
⋆ k+1 (20)
in order to reach vdc in N ⋆ steps, while limiting each 2r

⋆ k+1 2
capacitor current increment by N1⋆ of the total current required
2 qg = 0,
to lead the dc-voltage reference. Thus, the average next-step v̂gk
capacitor voltage references are given by: where
1  ⋆ k 
a=
2r

c = pe⋆dc k+1
−2f 2r

⋆ k+1 2
⋆ k+1
veC = vC
k
+ v − vC
k
2 , b = −1, +
2 qg .
1 1
N ⋆ C1 1
(14) v̂gk v̂gk
1  
⋆ k+1 k ⋆ k k
veC = vC 2 + vC − vC 2 , Given that pe⋆g k+1 has two distinct√roots, the solution that
2
N⋆ 2
b2 −4ac
⋆ k ⋆ k v⋆ k+1 minimizes the power is equal to −b− 2a , i.e.,:
where vC = vC = dc2 . s
1 2
 !
Therefore, the required average capacitor currents to achieve ⋆ k+1 ρk 4
⋆ k+1 −2f (qg⋆ k+1 )2
these voltage references are: peg = 1− 1 − k pedc + ,
2 ρ ρk


ei⋆C k+1 = C1 veC
⋆ k+1
− vC1 k 2
(21)
1
Ts 1 k (v̂gk )
(15) where ρ = 2r .


ei⋆C k+1 = C2 veC
⋆ k+1
− vC 2
k
. Consequently, after following this procedure, the final ex-
2
Ts 2
pression for the current reference, i⋆g k+1 , which also takes into
Additionally, the required average dc-current for this set account the dc-voltage regulation, is given by:
point is given by:
⋆ k+1
veC +e⋆ k+1
vC p⋆g k+1
2e −2qg⋆ k+1
ei⋆dc k+1 = 1 2
. (16) i⋆g k+1 = sin (ωt) + cos (ωt). (22)
k+1 v̂gk v̂gk
redc
1 A discrete form of the notch-filter can be easily designed as: y k =
Since the actual values of vC1 k and vC2 k are available as
xk b1 + xk−1 b2 + xk−2 b3 − y k−1 a2 − y k−2 a3 , where a1 = 1, a2 =
measurements and the average dc-side power consumption −1.9555, a3 = 0.9565, b1 = a2 , b2 = −1.9555 and b3 = 0.9782 for Ts
modeled by redc can be considered approximately constant = 50µs.

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

⋆
i⋆g k+1 vdc
⋆ k+1
vdc 400
MPC Sxopt vdc
& Single-phase

[V]
Reference (a)
Generation Minimization NPC 300
qg⋆ k+1
of J k Converter
vC1 vC2
200

ωt v̂gk 400
vg
vg k Estimat. vg k ig k vC1 k vC2 k idc k 200

[V],[A]
Measurements (b) 0
-200
Fig. 3. Block diagram of the proposed cascade-free FCS-MPC algo- ig × 10
rithm. -400
400
vab
200
Finally, the block diagram of the proposed MPC algorithm is

[V]
⋆ k+1 0
shown in Fig. 3. The desired dc-voltage reference vdc and (c)
-200
the reactive power reference qg⋆ k+1 are first set and fed into
-400
the reference generation block. This block also receives the
0 0.035 0.07 0.115 0.14
main grid parameters, i.e., peak voltage v̂gk and angle ωt from [s]
a demodulation based single-phase grid-voltage fundamental Fig. 4. Simulation results under dc-voltage step change from 360V to
estimation block2 . With this information and the measured 432V. (a) DC-side voltages, (b) grid-voltage and grid-current and (c)
values of vg k , ig k , vC1 k , vC2 k , and idc k at the actual instant converter output voltage.
k, the reference generation block provides the future current
reference, i⋆g k+1 , to transfer the required dc-power from the
grid while regulating the dc-voltage. Moreover, a desired considering N ⋆ equal to 2f1Ts which amounts to allowing
power factor is achieved by including the reactive power the converter to reach vdc ⋆ k+1
approximately in a half of
reference, qg⋆ k+1 , into i⋆g k+1 . Then, this current reference the fundamental period as explained in [14]. Besides the
i⋆g k+1 is sent to the MPC block. Here, the cost function J k selection of N ⋆ it is also important to define constraints such
described in (7), which includes the current tracking error and as maximum power to be extracted from the grid when any
the capacitor voltage balance, is minimized at each sampling disturbance in the system is introduced. The selection of this
instant k. Thus, the optimal switching combination Sxopt , which maximum power allows the controller to reach the desired
minimizes J k , is applied at the next instant k + 1. voltage reference while avoiding overshoots. In this case,

pe⋆g k+1 is limited in eq. (22) according to pe⋆g k+1  ≤ pg max ,
IV. R ESULTS where pg max can be selected as the maximum power extracted
Simulation results have first been used to show the influence from the grid.
of the reference prediction horizon N ⋆ . Then, experimental Figure 4 details the effect of choosing N ⋆ and pg max equal
results have been conducted by focusing the attention on to 200 and 3252 W respectively. As can be clearly appreciated,
six particular operating conditions: voltage sag in the grid, the dc-voltage exhibits an inherent second harmonic compo-
dc-load step change, volt-ampere reactive (VAR) tracking nent in steady state while its dynamic behavior during the step
⋆ k+1
performance, dc-voltage tracking performance, cost function change in vdc is governed by the selection of N ⋆ and the
evaluation and uncertainties evaluation. The results presented 20 A peak saturation imposed by pg max . Notice that due the
in this section have been obtained using the following param- saturation imposed by pg max , the controller decides to limit the
eters: Grid voltage vg = 230 V, grid frequency f = 50 Hz, grid-current until the dc-voltage reach the desired voltage ref-
⋆ erence approximate in three fundamental periods. Also, both
dc-voltage ref. vdc = 360 V, dc-capacitors C1,2 = 4450 µF,
dc-load rdc = 69 Ω, filter inductor L = 3.43 mH (0.056 pu), capacitor voltages maintain the same behavior. Henceforth,
filter resistance r = 0.1 Ω, ref. pred. hor. N ⋆ = 200, sampling due to the good performance obtained, both factors will be
period Ts = 50 µs, Vbase = 230 V and Sbase = 2.75 kVA. kept at these values in the remaining experimental tests.
For comparison purpose, the same step change is performed
A. Simulation Results when the NPC converter is governed by a traditional cascaded
FCS-MPC and a traditional Resonant-Proportional Controller
Simulated results have been obtained by means of
(RPC) adopted from [19] and [20] respectively. The close
MATLAB-Simulink and PLECS. The MPC algorithm was
loop performance obtained when using the proposed algorithm
programmed using C-code within an S-function block. Thus,
considering N ⋆ equal to 20 in order to obtain a fast dynamic
the simulated algorithm can be easily later implemented in
response is shown in Fig. 5(a). The sub-indexes u and c
a digital control platform for experimental validation. The
stand for application of unconstrained and constrained active
single-phase NPC converter presented in Fig. 2 was tested with
power references respectively. Results for FCS-MPC with a
a linear load rated at 2.75 kW. Simulations were performed
bandwidth (BW) equal to 70 Hz and a damping factor (ζ) of
2 The interested reader is referred to [18] for further information about 0.707 are depicted in Fig. 5(b). Here, it can be observed that
single-phase grid-voltage fundamental parameter estimation methods. large BW produces fast dynamic response but with overshoots

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

4
500 vdc u 0.15
⋆ s.t.
vdc

Settling time [s]

3
[V]
(a)

[%]
400 0.1
vdcc
2

1/vdc
⋆
300 0.05 vdc (2f )
1
500 vdcu
⋆
vdc 0 0
0.5 1 1.5 2
[V]

(b) 400
vdc c [×C1,2 ]
Fig. 6. Sensitivity analysis for settling time (s.t.) under dc-voltage
300 step change from 360V to 432V and steady state second harmonic
component (vdc (2f ) ) with respect to dc-capacitances C1,2 .
500 vdcu
⋆
vdc 400
[V]

(c) 400
200 vg
vdcc

[V],[A]
(a) 0
300
0 0.06 0.12 0.18 0.24 -200 ig × 10
[s] -400
Fig. 5. Simulation results under dc-voltage step change from 360V to ⋆
432V. (a) cascade-free FCS-MPC, (b) cascaded FCS-MPC (PI-based 400 vdc
with BW = 70 Hz and ζ = 0.707) (c) PI-based with BW = 70 Hz plus
RPC with BW = 1.5 kHz and ζ = 0.707.
vdc

[V]
(b) 300

200 vC1 vC2

if the PI controller is not limited. To mitigate this, the PI is
x6 x7
limited by the maximum allowed current to be extracted from 400
the dc-side. In that case, the PI controller behaves slightly vab
200
slower than the proposed algorithm. Note that in practice,
[V]

0
no limitations on the references signals and/or PI controller (c)
-200
outputs may lead in high overcurrents. Finally, a RPC current
-400
controller along with a PI controller is tested. Since the 0 0.035 0.07 0.115 0.14
switching frequency of the RPC current controller is set to [s]
6 kHz and the BW to 1.5 kHz, the behavior of the dc-voltage Fig. 7. Experimental results under grid-voltage sag of 30% during 40ms.
loop remains similar with cascaded FCS-MPC as shown in (a) Grid-voltage and grid-current, (b) dc-side voltages and (c) converter
Fig. 5(c). Clearly, an optimization of the PI controller may be output voltage.
carried out to improve the dynamic response. However, the
proposed method presents the capability of including several
proposed controller was developed for k + 1, it is necessary
control objectives in the same cost function while easily
to apply delay compensation for practical implementation.
handling system constraints. The details about the PI design
Although delay compensation using estimated values for k + 1
have been omitted for the sake of brevity, but can be found in
and predictions in k+2 was applied, this step was intentionally
[20].
omitted for the sake of brevity. For this case, the complete
Additionally, Fig. 6 depicts the changes in settling time
control loop is executed by the controller every 50 µs, while
(s.t.) under the same dc-voltage step change for dc-capacitance
the optimal switching state is computed in 12 µs. An average
C1,2 variations. It can be noticed that as C1,2 decreases, the
switching frequency of 4.67 kHz was obtained.
controller is able to provide faster settling times. However,
1) Grid-voltage Sag: First, the NPC converter is operated
there is a trade-off between how many times the actual value
at rated conditions with unity power factor (qg⋆ = 0), low total
of C1,2 is smaller than its nominal value and the amount of
harmonic distortion (THDi = 3.6%) and balancing for both
steady state second harmonic component in vdc (vdc (2f ) ). This
capacitor voltages. The normal condition is suddenly altered,
trade-off limits the operation of the controller to be inside a
by changing peak value of the grid v̂g from 100% to 70%
range where vdc (2f ) takes a desired value, i.e. tolerable vdc
during 40ms. Figure 7 shows that the grid-current present an
ripple. For that reason, besides the controller tolerates this
increment to compensate the variation on vg while keeping
change without knowing the actual C1,2 value, most of single-
the dc-voltage around its reference.
phase converters still use bulky dc-side capacitors that reduce
2) Load Step Change: Second, a load step change from
vdc (2f ) .
rdc = 69 Ω to rdc = 47 Ω, equivalent to increasing the active
power from 1.64 kW to 2.75 kW was produced to evaluate
B. Experimental Results the dynamic behavior of vdc . As observed in Fig. 8(b) the
The effectiveness of the proposed MPC algorithm is tested proposed controller is able to maintain the desired value of vdc ,
in a 2.75 kVA single-phase experimental setup. The parameters despite the load disturbance. Moreover, the capacitor voltages
of the setup are same as used in simulations. Since the remain balanced.

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

400 400
vg vg
[V],[A] 200 200

[V],[A]
(a) 0 (a) 0
-200 -200
ig × 10 ig × 10
-400 -400
⋆
400 ⋆
vdc 400 vdc
vdc
vdc
[V]

[V]
(b) 300 (b) 300
vC1 vC2
200 vC1 vC2 200

400 400
vab vab
200 200

[V]
[V]

0 0
(c) (c)
-200 -200
-400 -400
0 0.035 0.07 0.115 0.14 0 0.035 0.07 0.115 0.14
[s] [s]
Fig. 8. Experimental results under load step change from rdc = 79 Ω to Fig. 10. Experimental results under dc-voltage step change from 360V
rdc = 47 Ω. (a) Grid-voltage and grid-current, (b) dc-side voltages and to 432V. (a) Grid-voltage and grid-current, (b) dc-side voltages and (c)
(c) converter output voltage. converter output voltage.

400
200
vg 5) Cost Function Evaluation: As mentioned in the intro-
[V],[A]

0 duction, suitable cost function formulations can be adopted

(a)
to permit the operation with spread and imposed grid-current
-200
ig × 10 harmonic spectrum. The proposed algorithm was formulated
-400
using a typical quadratic error cost function as per eq. (7)
⋆ which gives a non-fixed harmonic spectrum as shown in
400 vdc
vdc Fig. 11(c). However, the cost function proposed in eq. (7) can
[V]

(b) 300 be transformed into eq. (23) as explained in [15] to obtain a

vC1 vC2 nearly fixed harmonic spectrum.
200

2 
2000 Jwk =F (k) i⋆g k+1 − ig k+1 +
qg⋆
qg 2
ig
2
[VAR]

0
vC1 k+1 − vC2 k+1 . (23)

(c) vdc 2
2
-2000 The function F (k) is a band-stop filter which is designed to
0 0.035 0.07 0.115 0.14 fix the harmonic spectrum of the grid-current around a desired
[s]
frequency. Results obtained using the so-called frequency
Fig. 9. Experimental results under VAR step change from −1500 VAR weighted cost function Jwk centered at 2250 Hz are shown in
to 1500 VAR. (a) Grid-voltage and grid-current, (b) dc-side voltages and
(c) reactive power. Fig. 11(b). Notice that due to the reduction of the switching
frequency the value of L was increased twice to decrease the
ripple in the grid-current. Here, the corresponding spectrum
3) VAR Tracking Performance: Next, one test was per- (see Fig. 11(d)) shows that a nearly fixed frequency can be
formed to evaluate the tracking performance of the VAR achieved without manipulating the equations that belong to the
reference. Fig. 9 depicts the experimental results under qg⋆ step proposed systematic approach, i.e. i⋆g k+1 , vC1 k+1 and vC2 k+1 .
change from −1500 VAR to 1500 VAR. Consequently, the Additionally, the same test used to evaluate the dc-voltage
grid-current changes from lagging the grid-voltage to leading tracking performance was performed to confirm the flexibility
it while maintaining the dc-voltage around its reference. Espe- to adopt different cost functions. In this case, a step change in
cially, Fig. 9(a) shows that the phase angle of the grid-current the vdc reference was produced from 360 V to 432 V. As can
quickly turns 180◦ at t = 0.04 s, barely affecting the capacitor be clearly appreciated in Fig. 12(b), the proposed controller
voltage balance. reaches the desired voltage reference while imposing a nearly
4) DC-voltage Tracking Performance: A step change in the fixed switching frequency which produces an improved wave-
vdc reference was produced from 360 V to 432 V which is form of the converter output voltage as shown in Fig. 12(c).
equivalent to increase the active power by a factor of 1.44. 6) Sensitivity Analysis: Finally, a set of filter inductance
As can be clearly appreciated in Fig. 10(b), the proposed value variation tests were performed to experimentally ex-
controller reaches the desired voltage reference as expected. amine the tracking capabilities. These tests were carried out

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

20 30

25
ig THD50
[A]
(a) 0
i⋆g 20

[%]
10
-20
20 5
WTHD eig

ig 0
[A]

0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5

(b)
0 i⋆g
% of rated filter inductance [÷100]

-20 Fig. 13. Steady state distortion indexes of the grid-current under filter
0 0.005 0.01 0.015 0.02 inductance value (L) variation.
[s]
5
4 h1 this case it is recommended to avoid uncertainties greater than
% of h1

3
(c) 2.25 or less than 0.75 times the rated value in order to maintain
2
e.g. a THD50 below 5%. The effect of L-variations over the
1
close-loop dynamic performance remains unaltered. In these
5 x6 cases, the MPC saturates its output forcing ig to follow the
4 dynamic of the plant, which behaves as a pure integrator due to
h1
% of h1

3 r << XL . Since, the variation of the r value did not represent

(d) 2 a problem for the proposed controller, these variations where
1 not registered.
0 10 20 30 40 50
Harmonic Order V. C ONCLUSION
Fig. 11. Steady state evaluation for different cost functions: (a) Grid- A cascade-free FCS-MPC algorithm to govern single-phase
current for a typical quadratic error cost function (see eq. (7)), (b) grid- grid-connected power converters has been presented and an-
current for a frequency weighted cost function centered at 2250 Hz (see
eq. (23)), (c) grid-current harmonic spectrum for (a) and (d) grid-current alyzed. The key novelty of this approach is the compatible
harmonic spectrum for (b). reference design that directly allows to formulate the opti-
mal control problem using the grid-current in the original
400
vg stationary reference frame. As a result of the design pro-
200
cedure, no cascaded control loops are required, making the
[V],[A]

(a) 0 proposed cascade-free FCS-MPC algorithm easy to implement

-200 in practice. Compared with traditional implementations such
ig × 10
-400 as cascaded FCS-MPC, the proposed algorithm presents the
⋆
vdc capability of including several control targets in the same cost
400 vdc function while easily handling system constraints. Simulation
and experimental results have shown that the grid-current, dc-
[V]

(b) 300
vC1 vC2 voltage and capacitor voltages are kept around their references
200 despite changes in the operating conditions.

400
vab R EFERENCES
200
[1] A. Youssef, S. El Khil, and I. Slama-Belkhodja, “State observer-based
[V]

0
(c) sensor fault detection and isolation, and fault tolerant control of a single-
-200 phase pwm rectifier for electric railway traction,” IEEE Trans. Power
-400 Electron., vol. 28, DOI 10.1109/TPEL.2013.2257862, no. 12, pp. 5842–
0 0.035 0.07 0.115 0.14 5853, Dec. 2013.
[2] C.-H. Chang, Y.-H. Lin, Y.-M. Chen, and Y.-R. Chang,
[s]
“Simplified reactive power control for single-phase grid-connected
Fig. 12. Experimental results under dc-voltage step change from 360V photovoltaic inverters,” IEEE Trans. Ind. Electron., vol. 61,
to 432V using a frequency weighted cost function centered at 2250 Hz. DOI 10.1109/TIE.2013.2271600, no. 5, pp. 2286–2296, May. 2014.
(a) Grid-voltage and grid-current, (b) dc-side voltages and (c) converter [3] M. Monfared, S. Golestan, and J. M. Guerrero, “Analysis, design,
output voltage. and experimental verification of a synchronous reference frame voltage
control for single-phase inverters,” IEEE Trans. Ind. Electron., vol. 61,
DOI 10.1109/TIE.2013.2238878, no. 1, pp. 258–269, Jan. 2014.
[4] Z. Yao and L. Xiao, “Control of single-phase grid-connected in-
to verify the converter limits which produces a maximum verters with nonlinear loads,” IEEE Trans. Ind. Electron., vol. 60,
permissible steady state distortion index of the grid-current DOI 10.1109/TIE.2011.2174535, no. 4, pp. 1384–1389, Apr. 2013.
created from unknown filter parameters. Three well-known [5] S. Xu, J. Wang, and J. Xu, “A current decoupling parallel con-
trol strategy of single-phase inverter with voltage and current
distortion indexes were evaluated – average tracking error dual closed-loop feedback,” IEEE Trans. Ind. Electron., vol. 60,
⋆
eig = |ig − ig |, THD50 and WTHD – as shown in Fig. 13. In DOI 10.1109/TIE.2011.2161660, no. 4, pp. 1306–1313, Apr. 2013.

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2599478, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

[6] M. Vasiladiotis and A. Rufer, “Dynamic analysis and state feed-

back voltage control of single-phase active rectifiers with dc- Ricardo P. Aguilera (S’01–M12) received his
link resonant filters,” IEEE Trans. Power Electron., vol. 29, M.Sc. degree in Electronics Engineering from
DOI 10.1109/TPEL.2013.2294909, no. 10, pp. 5620–5633, Oct. 2014. the Universidad Tecnica Federico Santa Maria
[7] M. Karimi-Ghartemani, S. A. Khajehoddin, P. Jain, and A. Bakhshai, (UTFSM), Chile, 2007, and Ph.D. degree in
“A systematic approach to dc-bus control design in single-phase grid- Electrical Engineering from The University of
connected renewable converters,” IEEE Trans. Power Electron., vol. 28, Newcastle (UN), Australia, 2012. From 2012 to
DOI 10.1109/TPEL.2012.2222672, no. 7, pp. 3158–3166, Jul. 2013. 2013, he was a Research Academic at UN,
[8] S. Vazquez, A. Marquez, R. P. Aguilera, D. Quevedo, J. I. Leon, and where he was part of the Centre for Complex
L. G. Franquelo, “Predictive optimal switching sequence direct power Dynamic Systems and Control. From 2014 to
control for grid-connected power converters,” IEEE Trans. Ind. Electron., 2016, he was a Senior Research Associate at
vol. 62, DOI 10.1109/TIE.2014.2351378, no. 4, pp. 2010–2020, Apr. The University of New South Wales (UNSW),
2015. Australia, where he was part of the Australian Energy Research Institute
[9] S. Vazquez, J. I. Leon, L. G. Franquelo, J. Rodriguez, H. A. Young, (AERI). Since 2016, he joined The University of Technology Sydney
A. Marquez, and P. Zanchetta, “Model predictive control: A review of (UTS), Australia, where he currently holds a Lecturer position. His
its applications in power electronics,” IEEE Ind. Electron. Mag., vol. 8, main research interests include power electronics, and theoretical and
DOI 10.1109/MIE.2013.2290138, no. 1, pp. 16–31, Mar. 2014. practical aspects on model predictive control.
[10] W. Song, J. Ma, L. Zhou, and X. Feng, “Deadbeat predictive power
Amer M.Y.M. Ghias (S’10-M’14) received the B.
control of single phase three level neutral-point-clamped converters
Sc. degree in electrical engineering from Saint
using space-vector modulation for electric railway traction,” IEEE Trans.
Cloud State University, USA, in 2001, the M.
Power Electron., vol. 31, DOI 10.1109/TPEL.2015.2400924, no. 01, pp.
Eng. degree in telecommunications from Univer-
721–732, Jan. 2016.
sity of Limerick, Ireland, in 2006, and the Ph.D.
[11] M. Preindl and S. Bolognani, “Model predictive direct speed control with
degree in electrical engineering from the Univer-
finite control set of pmsm drive systems,” IEEE Trans. Power Electron.,
sity of New South Wales (UNSW), Australia, in
vol. 28, DOI 10.1109/TPEL.2012.2204277, no. 2, pp. 1007–1015, Feb.
2014. He is currently Assistant Professor at the
2013.
University of Sharjah, Electrical and Computer
[12] E. Fuentes, D. Kalise, J. Rodrguez, and R. M. Kennel, “Cascade-free
Engineering faculty, UAE. His research inter-
predictive speed control for electrical drives,” IEEE Trans. Ind. Electron.,
ests include model predictive control of power
vol. 61, DOI 10.1109/TIE.2013.2272280, no. 5, pp. 2176–2184, May.
electronics converter, hybrid energy storage, fault-tolerant converter,
2014.
modulations and voltage balancing techniques for multilevel converter.
[13] K. G. Pavlou, M. Vasiladiotis, and S. N. Manias, “Constrained model
predictive control strategy for single-phase switch-mode rectifiers,” IET
Power Electron., vol. 5, DOI 10.1049/iet-pel.2010.0253, no. 1, pp. 31– Marco Rivera (S’09–M’11) received his B.Sc.
40, Jan. 2012. in Electronics Engineering and M.Sc. in Electri-
[14] D. E. Quevedo, R. P. Aguilera, M. A. Perez, P. Cortes, and cal Engineering from the Universidad de Con-
R. Lizana, “Model predictive control of an afe rectifier with cepción, Chile in 2007 and 2008, respectively.
dynamic references,” IEEE Trans. Power Electron., vol. 27, He recieved his Ph.D. degree at the Department
DOI 10.1109/TPEL.2011.2179672, no. 7, pp. 3128 –3136, Jul. of Electronics Engineering, Universidad Técnica
2012. Federico Santa Marı́a (UTFSM), in Valparaı́so,
[15] P. Cortes, J. Rodriguez, D. E. Quevedo, and C. Silva, “Predictive current Chile, in 2011. He is currently Associate Pro-
control strategy with imposed load current spectrum,” IEEE Trans. fessor in the Department of Industrial Technolo-
Power Electron., vol. 23, DOI 10.1109/TPEL.2007.915605, no. 2, pp. gies at Universidad de Talca, Curicó, Chile. His
612–618, Mar. 2008. main research areas are digital control applied
[16] B.-R. Lin and D.-J. Chen, “Single-phase neutral point clamped ac/dc to power electronics, matrix converters, predictive control and control of
converter with the function of power factor corrector and active power converters for renewable energy applications.
filter,” IEE Proceedings - Electric Power Applications, vol. 149,
DOI 10.1049/ip-epa:20020023, no. 1, pp. 19–30, Jan. 2002.
[17] H. Akagi, E. H. Watanabe, and M. Aredes, Instantaneous power theory Carlos R. Baier (S’08) He received the B.Sc.,
and applications to power conditioning. John Wiley & Sons, 2007, M.Sc. and D.sc. degrees in electrical engineer-
vol. 31. ing from the University of Concepcion, Concep-
[18] M. Reza, M. Ciobotaru, and V. G. Agelidis, “A modified de- cion, Chile, in 2004, 2006, and 2010, respec-
modulation technique for single-phase grid voltage fundamen- tively. Since 2009, he has been a Professor
tal parameter estimation,” IEEE Trans. Ind. Electron., vol. 62, in the Department of Industrial Technologies,
DOI 10.1109/TIE.2014.2366720, no. 6, pp. 3705–3713, Jun. 2015. University of Talca, Talca, Chile, where he is
[19] P. Acuna, L. Moran, M. Rivera, R. Aguilera, R. Burgos, and V. G. Age- teaching in the areas of industrial electronics.
lidis, “A single-objective predictive control method for a multivariable His research interests include improved control
single-phase three-level npc converter-based active power filter,” IEEE techniques for multicell converters, new multi-
Trans. Ind. Electron., vol. 62, DOI 10.1109/TIE.2015.2393556, no. 7, level topologies to inject power into the grid and
pp. 4598–4607, Jul. 2015. high energy efficient improvements for medium-voltage converters.
[20] P. Lezana, C. A. Silva, J. Rodriguez, and M. A. Perez, “Zero-steady-
state-error input-current controller for regenerative multilevel converters
based on single-phase cells,” IEEE Trans. Ind. Electron., vol. 54, Vasssilios G. Agelidis (SM’00–F’16) received
DOI 10.1109/TIE.2007.891994, no. 2, pp. 733–740, Apr. 2007. the B.Eng. degree in Electrical Engineering from
the Democritus University of Thrace, Greece, in
Pablo Acuna (M’12) received the B.Sc. in 1988, the M.S. degree in Applied Science from
Electronics Engineering, the Electronics Engi- Concordia University, Montreal, QC, Canada, in
neering Professional, and the Ph.D. degrees 1992, and the Ph.D. degree in Electrical Engi-
in Electrical Engineering from the University of neering from Curtin University, Perth, Australia,
Concepcion, Chile, in 2004, 2007, and 2013 in 1997. He has done extensive research in
respectively. He is currently Research Associate the field of smarter grid infrastructure and sus-
at University of New South Wales, Australia. His tainable energy systems incorporating solar and
research interests include electrical power con- wind energy sources. He is currently Professor
version systems and its applications to industry, of Power Engineering at the School of Electrical Engineering and
transportation and utility. Telecomunications at the University of New South Wales (UNSW).

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.