Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO.

2, FEBRUARY 2013 619


Fast-Predictive Optimal Control of
NPC Multilevel Converters
J. Dionsio Barros, Member, IEEE, J. Fernando A. Silva, Senior Member, IEEE, and lvio G. A. Jesus
AbstractThe development of high-processing-capability mi-
croprocessors allows the implementation of new digital control
methods for neutral-point-clamped (NPC) multilevel converter in
power-electronic applications. This paper presents a new predic-
tive digital control method for multilevel converters, called fast
predictive. This method computes the optimal vector using the
NPC three-phase multilevel dynamic model equations just once
in each control cycle, while current predictive methods need 27
calculations. The closest vector to the optimal vector is found
by minimizing the distance between each one of the 27 available
vectors to the optimal vector. Space vector modulation could be
also used. The obtained performance is similar to the predictive
optimal control that uses the converter model to nd all the 27 re-
sponses of the multilevel and then searches for the vector that min-
imizes control errors. Relative to predictive optimal control, the
fast predictive improves digital processing speed by at least 150%
in multilevel converters with 27 vectors. This speed improvement
would allow multilevel converters with ve or higher number of
levels (125 instead of 27 vectors) to be controlled using the same
sampling frequency of the three-level inverter. The fast-predictive
controller is used in a multilevel rectier with near-unity power
factor to enforce the ac currents. Fast predictive control is also
used in the rectier dc voltage to reduce sensitivity of the dc
voltage to dc load disturbances. The simulation and experimental
results showthat the fast-predictive controller is able to control the
ac currents of a three-phase multilevel rectier, achieving nearly
1.5% total harmonic distortion while balancing the capacitors dc
voltages. The use of predictive control to regulate the dc voltage
shows an improvement of approximately 7% compared to a pro-
portional-integral controller.
Index TermsDigital control, fast-predictive controller, multi-
level converter, optimal controller, unity power factor.
I. INTRODUCTION
A
PPLICATIONS of power electronics to improve power
quality often use active-power semiconductors operating
at relatively high switching frequencies. The drive and the con-
trol of power converter semiconductors often rely in digital con-
trol and thus require fast digital processors and high sampling
frequencies to implement advanced control techniques [1].
Manuscript received June 9, 2012; accepted June 14, 2012. Date of publica-
tion July 26, 2012; date of current version September 13, 2012. This work was
support in part by the Universidade da Madeira, by the Universidade Tcnica
de Lisboa, Instituto Superior Tcnico, and by the Center for Innovation in
Electrical and Energy Engineering (Cie
3
).
J. D. Barros is with the Exact Sciences and Engineering Competence Centre,
University of Madeira, 9000-390 Funchal, Portugal (e-mail: dbarros@uma.pt).
J. F. A. Silva is with the Department of Electrical and Computer Engineering,
and the Center for Innovation in Electrical and Energy Engineering (Cie
3
),
Instituto Superior Tcnico, Technical University of Lisbon, 1049-001 Lisboa,
Portugal (e-mail: fernandos@alfa.ist.utl.pt).
. G. A. Jesus is with the Secondary School Padre Manuel lvares,
9350-211 Ribeira, Portugal, and also with the Exact Sciences and Engineer-
ing Competence Centre, University of Madeira, 9000-390 Funchal, Portugal
(e-mail: elvio.jesus@gmail.com).
Digital Object Identier 10.1109/TIE.2012.2206352
The evolution of digital processing systems based on micro-
processors and microcontrollers, eld-programmable gate ar-
rays, digital signal processors (DSPs), and application-specic
integrated circuits associated with improvements of the charac-
teristics of power semiconductors has been the key to the rapid
evolution and development of power-electronic applications
and to improve power quality [1][7].
Multilevel converters are characterized by the ability to
supply staircase-like voltage waveforms. This feature enables
converters with a high number of levels to reduce voltage and
current total harmonic distortion (THD), and their architecture
enables high-voltage high-current operation [8][12].
Control methods based on pulsewidth modulation (PWM)
[13], [14], space vector modulation (SVM) [15], [16], and
sliding mode [17] are often used to control the multilevel con-
verters. The PWM strategy is suitably implemented in digital
controllers [18]. Other techniques, such as sliding mode [19],
[20], that can present robustness and improved stability can be
also implemented.
The optimal-predictive digital control method has been re-
cently used in multilevel-converter applications. Based on the
system dynamics model, the optimal-predictive computes, in
real time, the results of applying each one of all the available
vectors and chooses the best vector, which is the one that
minimizes a cost functional of the control errors [20][25].
The choice of the optimal-model-based vector system may
require high computational resources and high-speed real-time
digital control systems, usually limiting the application of this
method to simple load models, converters with limited set
of states or control vectors ( 27), or systems with a rela-
tively low number of calculations to nd the optimal vector.
For example, in a three-level three-phase multilevel converter,
the algorithm to nd the optimal vector must be computed
27 times in each control cycle. The cost functional must be also
computed 27 times, and only the algorithm to nd its minimum
is executed once.
This paper presents a new predictive control method, called
fast predictive, that uses the model equations of the system just
once in each control cycle to predict the ideal optimal vector
needed to control the state space variables. The choice of the
converter output vector is obtained by minimizing the distance
between the ideal vector and the converter available vectors.
Alternatively, an SVM technique can be used.
The fast-predictive digital control method provides similar
results when compared to the optimal-predictive digital control
method but with a very signicant improvement in reducing
the processing time (around 150% for neutral-point-clamped
(NPC) three-phase converters).
0278-0046/$31.00 2012 IEEE
620 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 2, FEBRUARY 2013
Fig. 1. NPC multilevel rectier circuit used in applications to improve the
power quality.
This paper presents the multilevel-converter models (see
Section II) working as rectiers with near-unity power factor.
In Section III, the fast-predictive digital control method is
developed to digitally control the following: 1) the converter dc
voltage; 2) the ac currents; and 3) the balancing of capacitors
dc voltage of the NPC multilevel converter. The design of a dc
voltage proportional-integral (PI) controller (for comparison)
is also included. Section IV compares the optimal-predictive
controller with the fast-predictive controller of the NPC mul-
tilevel converter. The simulation and experimental results of
the NPC multilevel converter working as a rectier with unity
power factor with the digital fast-predictive control is presented
in Section V. The last section points the main conclusions of
this paper.
II. MULTILEVEL RECTIFIER DYNAMIC MODELING
The NPC multilevel rectier is connected to the electrical
power network U
L
through the coupling coils L with loss
resistance R. The dc load Z
Udc
is connected to the capacitor
terminals of the multilevel converter, i.e., C
1
and C
2
, and has
an applied voltage U
dc
= U
C1
+ U
C2
(see Fig. 1).
Assuming ideal semiconductors, the three valid combina-
tions of the binary states of switches S
kj
of each leg k can be
dened as

k
=
_
_
_
1 (S
k1
=1 and S
k2
=1)and(S
k3
=0 and S
k4
=0)
0 (S
k1
=0 and S
k2
=1)and(S
k3
=1 and S
k4
=0)
1 (S
k1
=0 and S
k2
=0)and(S
k3
=1 and S
k4
=1).
(1)
The dependence of the equations of the ac currents (i
1
+
i
2
+ i
3
= 0), represented in the coordinate system 123 [20],
can be eliminated by applying the ClarkConcordia transfor-
mation, resulting in the dynamic model represented in
_

_
di

dt
di

dt
dU
C1
dt
dU
C2
dt
_

_
=
_

R
L
0

1
L

2
L
0
R
L

1
L

2
L

1
C
1

1
C
1
0 0

2
C
2

2
C
2
0 0
_

_
_

_
i

U
C1
U
C2
_

_
+
_

1
L
0 0
0
1
L
0
0 0
1
C
1
0 0
1
C
2
_

_
_
_
U
L
U
L
i
dc
_
_
(2)
where the variables with the states of the semiconductor in ,

i
, and
i
, with i {1, 2}, are given by

i
=
_
2
3
_

i1


i2
2


i3
2
_

i
=
_
2
3
_

3
2

i2

3
2

i3
_
(3)
where

1k
=

k
(
k
+ 1)
2

2k
=

k
(1
k
)
2
. (4)
The converter model equations (2) will be used to design the
fast-predictive controller.
Considering the almost-sinusoidal behavior, the ac currents
and voltages, represented in dq coordinates, have almost con-
stant values for each operating point. This enables the design
of dc voltage controllers for the multilevel rectier. Applying
the Park transform to the model of the multilevel rectier,
represented in coordinate system (2), the model of the
rectier in dq coordinates (5) is obtained. The following will
allow the design of dc voltage controllers:
_

_
di
d
dt
di
q
dt
dU
C1
dt
dU
C2
dt
_

_
=
_

R
L


1d
L

2d
L

R
L

1q
L

2q
L

1d
C
1

1q
C
1
0 0

2d
C
2

2q
C
2
0 0
_

_
_

_
i
d
i
q
U
C1
U
C2
_

_
+
_

1
L
0 0
0
1
L
0
0 0
1
C
1
0 0
1
C
2
_

_
_
_
U
Ld
U
Lq
i
dc
_
_
. (5)
Argument = t is the angular phase of the electrical power
network. The semiconductor states in dq,
id
, and
iq
, with
i {1, 2}, are given by
_

id

iq
_
=
_
cos sin
sin cos
_ _

i
_
. (6)
III. DESIGN OF FAST-PREDICTIVE CONTROLLERS
The control system of the optimal-predictive rectier (see
Fig. 2) consists of a voltage regulator, which determines the
value of the ac current reference i
dqRef
based upon the dc
error voltage e
Udc
using the dynamics of the dc voltage, i.e.,
dU
dc
/dt. The control of the ac current and the capacitor dc
voltage balance is done through the on/off switching of the NPC
multilevel active switches, i.e., S
kj
, so that the ac currents i
1
,
i
2
, and i
3
are nearly sinusoidal to conserve power quality.
A. DC Voltage Control
The dc voltage controller evaluates the deviation of the dc
voltage U
dc
in relation to its reference U
dcRef
to determine the
values of the ac current reference in order to regulate the dc
voltage to a constant value, even with load disturbances. A fast-
predictive dc voltage controller is introduced. For performance
comparison, the well-known PI controller is also designed.
BARROS et al.: FAST-PREDICTIVE OPTIMAL CONTROL OF NPC MULTILEVEL CONVERTERS 621
Fig. 2. NPC multilevel rectier control system.
1) Fast-Predictive Control of DC Voltage: The predictive
controller design is performed based on the equation of the
dc voltage dynamics as follows, which is obtained from the
multilevel rectier model (5):
dU
dc
dt
=
_

1d
C
1
+

2d
C
2
_
i
d

1q
C
1
+

2q
C
2
_
i
q
+
_
1
C
1
+
1
C
2
_
i
dc
.
(7)
The ac current reference is obtained by inverting (7) (reverse
dynamics model). For conciseness, the following explanation
considers that the rectier operates at near-unity power factor
(i
q
i
qRef
= 0). Approximating the derivative of the dc volt-
age U
dc
in (7) by a rst-order behavior in the time interval T
Udc
,
a prediction for the current reference i
dRef
is obtained in
i
dRef

U
dcRef

6U
L
C
U
dcRef
U
dc
T
Udc
+
U
dcRef

6U
L
2i
dc
. (8)
To obtain the fast-predictive control law (8), it was consid-
ered that the capacitance of the multilevel-converter capacitors
are equal, i.e., C
1
C
2
C, and the state of the active semi-
conductors variables, i.e.,
1d
+
2d
, as follows are evaluated
from the steady-state equation of the ac current i
d
in (5),
neglecting the converter semiconductor losses:

1d
+
2d

2

6U
L
U
dc
. (9)
The state of active semiconductors
1d
+
2d
in steady state
depends on the amplitude of the ac voltage U
L
and the dc volt-
age U
dc
, which are sampled and used in the state calculations.
For nonunity power factors, from the active and reactive powers
P = v
d
i
d
+ v
q
i
q
and Q = v
q
i
d
v
d
i
q
, using a dq referential
in which v
q
= 0, i
qRef
= Q
Ref
/v
d
would be obtained, and
term
1q
+
2q
2Li
d
/U
dc
from (5) should be added to (8)
to ensure a nonzero reactive power Q
Ref
.
2) DC Voltage PI Controller: For comparison purposes, the
design of the dc voltage controller with a classical PI controller,
characterized by imposing stationary zero errors, is presented.
From the dynamic model of the dc voltage (7) and consid-
ering that the dc load is a resistance, i.e., R
dc
, the transfer
function in a closed loop results in a second-order system
(s
2
+ 2
n
s +
2
n
) with zero, i.e.,
U
dc
U
dcRef

3U
LRMS
U
dcRef
C
(sK
pUdc
+K
iUdc
)
s
2
+s
_
2
R
dc

3U
LRMS
K
pUdc
U
dcRef
C
_

3U
LRMS
K
iUdc
U
dcRef
C
.
(10)
The PI constants K
pUdc
and K
iUd
must be chosen to place
the zero well to the right of the complex poles and can be
written as
K
pUdc
=
U
dcRef
U
dcRef

n
R
dc
C

3U
LRMS
R
dc
(11)
K
iUdc
=
U
dcRef

2
n
C
2

3U
LRMS
. (12)
The damping factor is usually chosen to be 1/

2, for
fast response, but without signicant overshoot (K
pUdc
0).
The closed-loop natural frequency
n
should be
n
,
where is the fundamental frequency of the electrical power
network.
B. Predictive Control of NPC Multilevel AC Currents
1) Optimal-Predictive Controller of the NPC: Optimal-
predictive controllers [20][25] analyze in real time all the
vectors of the multilevel converter and choose the vector that
minimizes the cost functional of the state variables to follow
the reference values.
In the optimal-predictive controller, the predicted quantities
are the state space variables. The predictive equations are
derived from the direct dynamics system model (2) [20], the
variables being predicted for the next (t
s
+ T) sampling time
interval T, using the previous sampled values t
s
, i.e.,
i
Ref(t
s
+T)
= i

(t
s
)
R
L
i

(t
s
)T
1
L
U
L
(t
s
)T +
1
L
V
i
(t
s
)T
(13)
i
Ref(t
s
+T)
= i

(t
s
)
R
L
i

(t
s
)T
1
L
U
L
(t
s
)T +
1
L
V
i
(t
s
)T
(14)
U
UC1Ref(t
s
+T)
U
UC2Ref(t
s
+T)
= U
UC1
(t
s
) U
UC2
(t
s
) +
1
C
I
UC
(t
s
)T (15)
where the available converter control inputs V
i
(t
s
), V
i
(t
s
),
and I
UC
(t
s
) are dened as follows, obtained considering all
the 27 available NPC converter vectors, given by variables (3)
that represent the states of the active semiconductors:
V
i
(t
s
) =
1
U
C1
+
2
U
C2
(16)
V
i
(t
s
) =
1
U
C1
+
2
U
C2
(17)
I
UC
(t
s
) =(
2

1
)i

+ (
2

1
)i

. (18)
622 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 2, FEBRUARY 2013
Equations (16)(18), together with (13)(15), are computed
27 times in each control cycle.
The cost functional of the optimal-predictive controller, i.e.,
C
OP
(t
s
), shown in (19) found at the bottom of the page,
evaluates the 27 control errors enabling the choice of the best
combination (3) of the active NPC semiconductors for the state
space variables to approach the references.
Weights
i
,
i
, and
UC
allow the user denition of the
priority level of each controlled variable.
The 27 calculations of (13)(19) during a sampling time
interval T need a relatively fast digital processor in order to
be able to predict the values of ac currents and dc voltages of
the converter within a sampling interval that is small enough
[20], [23].
2) Fast-Predictive Controller of the NPC: In the fast-
predictive control method, the dynamic equations are solved
just once to nd the expected optimal vector. The distance
between this expected optimal vector and all the 27 available
vectors is then evaluated, through simple operations to reduce
digital time processing, in order to nd the closest vector to the
expected optimal vector.
Solving the state space dynamic equations of the multilevel
converter (2), the prediction equations (inverse dynamics) to
compute the reference control vectors V
iRef
(t
s
), V
iRef
(t
s
),
and I
UCRef
(t
s
), are obtained as
V
iRef
(t
s
) = [i

(t
s
+ T) i

(t
s
)]

L
T
+ Ri

(t
s
) + U
L
(t
s
) (20)
V
iRef
(t
s
) = [i

(t
s
+ T) i

(t
s
)]

L
T
+ Ri

(t
s
) + U
L
(t
s
) (21)
I
UCRef
(t
s
) = [U
C2
(t
s
+ T) U
C1
(t
s+T
)]
C
T
(22)
where i

(t
s
+ T) = i
Ref
and i

(t
s
+ T) = i
Ref
are
the ac currents, and the capacitor voltages U
C1
(t
s
+ T) =
U
C1Ref
and U
C2
(t
s
+ T) = U
C2Ref
, which must be equal to
their reference values at the next (t
s
+ T) sampling interval.
In each control cycle, the fast-predictive cost functional
C
FP
(t
s
), as shown in (23) found at the bottom of the page,
is applied 27 times to obtain the available vectors V
i
(t
s
),
V
i
(t
s
), and I
UC
(t
s
) [calculated using (16)(18)] closest
to the expected optimal vectors V
iRef
(t
s
), V
iRef
(t
s
), and
I
UCRef
(t
s
) calculated only once from (20)(22)
Weights
V i
,
V i
, and
IUC
are design parameters al-
lowing the designer to assign priorities to the most important
quantities, the ac currents i

and i

, or to the capacitor voltage


balance.
In the fast-predictive control method, the reference vec-
tors V
iRef
(t
s
), V
iRef
(t
s
), and I
UCRef
(t
s
), which depend on
the system inverse model, are calculated just once. This is
a signicant improvement over the more common optimal-
predictive control, where the direct model (13)(15) are used
to determine i
Ref(ts+T)
, i
Ref(ts+T)
and U
C1Ref(ts+T)

U
C2Ref(ts+T)
27 times in each control cycle. The fast-
predictive approach evaluates the cost functional (23) for the
available 27 vectors.
IV. COMPARISON OF FAST- AND OPTIMAL-PREDICTIVE
NPC CONTROLLERS
In this section, a comparison between the fast-predictive
controller and the optimal-predictive NPC multilevel controller
is done. In Fig. 3, a owchart for each controller is represented,
i.e., the optimal predictive and the fast predictive [see Fig. 3(a)
and (b), respectively]. As shown in Fig. 3, the main difference
between these two control methods is the reduction in the
number of calculations done in the main loop. In the optimal-
predictive control method system, the predictive equations are
solved in the main loop, whereas the fast-predictive control
method minimizes the number of operations of the main loop.
The algorithms go through the vectors (see Table I) with
all combinations of states of the active semiconductor of the
multilevel converter [semiconductor leg states
1
,
2
, and
3
,
and vectors V
i
(t
s
) and V
i
(t
s
)]. In the fast-predictive control
method, the ideal control vectors V
iRef
(t
s
), V
iRef
(t
s
), and
I
UCRef
(t
s
) are computed so that the state space variables
follow the reference values. In the main loop, for all combi-
nations of vectors (27 for the case of the three-phase NPC),
the controller reads the available vectors V
i
(t
s
) and V
i
(t
s
)
stored in Table I and selects the best available vector I
UC
(t
s
)
to regulate the capacitor voltages.
In the optimal-predictive control method, the predictive
equations of ac currents i

and i

, together with the error of the


capacitor voltages U
C1Ref
U
C2Ref
, are calculated 27 times.
Then, the cost functional for each available vector is calculated
27 times also.
C
OP
(t
s
) =

(i
Ref
(t
s
+ T) i

(t
s
))
2

i
+
(i
Ref
(t
s
+ T) i

(t
s
))
2

i
+
(U
C2
(t
s
) U
C1
(t
s
))
2

UC
(19)
C
FP
(t
s
) =

(V
iRef
(t
s
) V
i
(t
s
))
2

V i
+
(V
iRef
(t
s
) V
i
(t
s
))
2

V i
+
(I
UCRef
(t
s
) I
UC
(t
s
))
2

IUC
(23)
BARROS et al.: FAST-PREDICTIVE OPTIMAL CONTROL OF NPC MULTILEVEL CONVERTERS 623
Fig. 3. Flowcharts of the (a) optimal- and (b) fast-predictive control methods.
In the fast-predictive control method, it is only necessary
to calculate, in the main loop, the cost functional to nd the
vector closest to the ideal, which was calculated just once [out
of the main cycle see Fig. 3(b)]. This reduces the number of
operations, and the leftover time can be used to improve the
performance of the fast-predictive control method against the
optimal-predictive control method or can enable the predictive
control of ve or plus level inverters (125 vectors or more).
The two predictive methods check, in each sampling step, if
the cost functional of the actual vector is the smallest among the
calculated ones (see Fig. 3). After going through all vectors (see
Table I), the semiconductors are driven with the state vector that
was found to minimize the cost functional.
Table II shows an estimation of the number of operations
that are processed in the digital controller for optimal- and
TABLE I
VECTORS OF THE NPC MULTILEVEL CONVERTER/TBL
TABLE II
NUMBER OF OPERATIONS OF THE FAST- AND OPTIMAL-PREDICTIVE
NPC CONTROL METHODS/TBL
fast-predictive control methods. The right column of Table II
outlines the improvement (in percentage) of the fast-predictive
control method compared to the optimal-predictive control
method.
The results of Table II show that, for three-level inverters,
there is about 45% reduction in the total number of opera-
tions of the fast-predictive control method against the optimal-
predictive method, with the highest percentage in the division
(69%) and multiplication (63%) operations. This improvement
would allow converters with higher number of vectors such
as ve-level inverters (125 vectors) and three-phase three-level
four-wire inverters (81 vectors) to be controlled using the same
processor speeds.
624 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 2, FEBRUARY 2013
TABLE III
CONVERTER PARAMETERS/TBL
Fig. 4. U
dc
voltage of the multilevel rectier with (a) PI and (b) fast-
predictive controls. Vertical: 20 V/div. Horizontal: 100 ms/div
V. SIMULATION AND EXPERIMENTAL RESULTS
This section presents experimental and simulation results
of the NPC multilevel operating as a rectier with near-unity
power factor using the fast-predictive control or the optimal
predictive for ac current control and using the fast-predictive
and PI controllers for dc voltage control. Table III contains the
values of the quantities used in the experimental and simulation
results.
A. NPC Rectier DC Voltage Control
The experimental results of the multilevel rectier U
dc
volt-
age (see Fig. 4) show that, connecting a load of R
dc
/2, in the
fth horizontal division, the U
dc
fast-predictive controller has
a faster response than the U
dc
PI controller. The U
dc
voltage
predictive controller makes the dc voltage almost constant, even
Fig. 5. Currents and voltages of the ac electrical power network in phases 1
(i
1
, U
L1
) and 2 (i
2
, U
L2
). Vertical: 5 A/div and 20 V/div. Horizontal:
10 ms/div. (a) NPC predictive controller. (b) NPC fast-predictive controller.
when a load is abruptly switched to R
dc
/2 [see Fig. 4(b)]. In
the U
dc
voltage PI controller, it has an undervoltage of about
8% [see Fig. 4(a)]. In the steady state, the dc voltages of the
two controllers do not have stationary errors.
Fig. 5 shows the ac currents and voltages of the multilevel
rectier with the U
dc
predictive controller. It is shown that
the ac currents are almost sinusoidal and in phase with the
ac voltage, indicating that the power factor is near unity. The
results of the NPC optimal- and fast-predictive controllers
(Fig. 5(a) and (b), respectively) show that currents follow their
references (obtained from the optimal-predictive dc voltage
controller), with an almost-zero steady-state error (lower than
1%). However, the fast-predictive controller [see Fig. 5(b)]
needs smaller digital processing times.
BARROS et al.: FAST-PREDICTIVE OPTIMAL CONTROL OF NPC MULTILEVEL CONVERTERS 625
Fig. 6. U
dc
voltage of the multilevel rectier in the steady state. Vertical:
20 V/div. Horizontal: 100 ms/div.
Fig. 7. U
dc
voltage and ac current responses to a step change in reference
U
dcRef
. Vertical: 20 V/div and 5 A/div. Horizontal: 100 ms/div. (a) PI U
dc
controller. (b) Fast-predictive U
dc
controller.
Fig. 6 represents the dc voltage U
dc
of the multilevel rectier
with near-unity power factor. The results show that, using
the U
dc
fast-predictive controller, the dc voltage follows the
reference value without stationary error, as in the PI controller.
Fig. 8. Capacitor voltages U
C1
and U
C2
using the NPC (a) optimal- and
(b) fast-predictive methods. Vertical: 10 V/div. Horizontal: 100 ms/div.
Fig. 7 shows the results of the dc voltage U
dc
, as well as
the ac currents i
1
, i
2
, and i
3
multiplied by four and with a
displacement of 5 A, when the dc voltage reference U
dcRef
has a step variation from 120 to 140 V. The results show that
the dc voltages tend to the nal value after an initial decrease,
due to the fact that PWM voltage source rectiers are intrinsic
nonminimum phase systems, using some dc power to increase
the ac currents, before these currents can, in turn, increase the
dc voltage. The PI controller [see Fig. 7(a)] shows a small
overshoot, and the response time is approximately 200 ms.
The fast-predictive controller of the dc voltage [see Fig. 7(b)]
shows a faster response on the dc voltage, approximately 50 ms,
without any overshoot.
B. AC Currents and Capacitor Voltage Balancing of the
NPC Using the Fast-Predictive Controller
Fig. 8(b) depicts the capacitor voltages of the multilevel
rectier using the NPC fast-predictive control method. The
results show that the voltages of capacitors are well balanced.
This balance was observed in all the experiments carried out
with the fast-predictive control method.
626 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 2, FEBRUARY 2013
Fig. 9. Capacitor voltages U
C1
and U
C2
, and U
dc
voltage during a step
change in reference U
dcRef
using the NPC (a) optimal- and (b) fast-predictive
methods. Vertical: 20 V/div horizontal: 100 ms/div.
The experimental results of the fast-predictive method [see
Fig. 8(b)] are similar to the results of the optimal-predictive con-
troller [see Fig. 8(a)], obtained using lower computing times.
Fig. 9 shows the capacitor voltages during a step variation
of the dc voltage from 120 to 140 V to analyze the capacitor
voltages balancing on the transient evolution. The experimental
results of the fast-predictive control method [Fig. 9(b)] and the
optimal-predictive method [Fig. 9(a)] show that the capacitor
voltage are balanced during the transient evolution, each ca-
pacitor showing voltage U
dc
/2. The results show that the fast-
predictive and optimal-predictive control methods perform the
capacitor voltage balancing on the transient mode (see Fig. 9),
as well as in the steady state mode (see Fig. 8).
To analyze the behavior of the processing speed of the
two controllers of the ac current, i.e., the optimal- and fast-
predictive controllers, a C language program was developed
to implement each of the controller algorithm in a DSP, i.e.,
dSPACE 1103. The experimental results show that the choice
of the optimal vector using the optimal-predictive control,
after analyzing the available 27 vectors takes 15 s. Also
using the fast-predictive algorithm this time is able to analyze
70 vectors and choose the optimal vector, which corresponds to
an improvement in the processing time of 160%.
TABLE IV
THD (IN PERCENT) OF AC CURRENT USING THE OPTIMAL- (OP I
AC
)
AND FAST-PREDICTIVE (FP I
AC
) CONTROLLERS IN A RECTIFIER WITH
THE PI DC VOLTAGE CONTROLLER (PI U
DC
) AND THE PREDICTIVE
CONTROLLER (PC U
DC
)/TBL
Considering a ve-level converter [13], with 125 vectors
available, the predictive optimal control takes 32 s to choose
the optimal vector. The fast-predictive algorithm can analyze
315 vectors in 32 s, which corresponds to an improvement in
the processing time by 150%.
The results show that the THD of ac currents of the converter
operating as a near-unity power factor rectier is about 1.5%
in both digital predictive controllers. This shows that the fast-
predictive control method, while being faster in choosing the
optimal vector in the DSP, is able to maintain the same distor-
tion level of the optimal-predictive control method.
Tables IV shows the THD results of the ac currents for
several sampling time intervals T. The results show that
increasing the sampling time interval increases the ac current
THD but with similar results in both predictive controllers of
the NPC. The results of Tables IV also show that the use of a
predictive controller to regulate the dc voltage provides a faster
dynamics and does not signicantly affect the power quality
compared to the PI controller.
The fast-predictive control method has similar performance
compared with optimal-predictive control method, and the im-
plementation of the digital control in a DSP runs faster than an
optimal-predictive control. It is thus an appropriate method to
implement converters with many active switches as multilevel
converters connected back to back and multilevel converters
with several levels and legs, where it is necessary to evaluate
a large set of vectors during the sampling interval to predict
in real time the optimal vector. Although the THD results are
slightly better using the PI dc voltage controller due to the fact
that this controller is slower than the predictive and therefore
acts more slowly in the ac currents, the control designer has the
possibility of tailoring weights
V i
,
V i
, and
IUC
to favor
THD reduction at the cost of a slower dc voltage response.
VI. CONCLUSION
This paper has presented a fast-predictive control method that
is able to reduce the processing time of the implementation in
the DSP of a predictive digital controller for power-electronic
converter applications.
The fast-predictive method solves the converter inverse dy-
namics model just once in each sampling interval to determine
the optimal vector. This method has been compared to the
optimal-predictive control, which solves in each sampling in-
terval the converter direct dynamics model, for all the available
BARROS et al.: FAST-PREDICTIVE OPTIMAL CONTROL OF NPC MULTILEVEL CONVERTERS 627
converter vectors, to determine the cost of each vector. The
fast-predictive method allowed improvements in the order of
150% in the number of vectors that can be analyzed in a DSP
digital controller for a NPC multilevel operating as a rectier
with unity power factor.
The experimental results show that the ac currents have a
THD of 1.5%, the voltages of capacitors in the NPC converter
are well balanced, and the fast-predictive dc voltage regulator
imposes faster speed to control the dc voltage while almost not
affecting the quality of ac currents, compared to a PI dc voltage
converter.
The fast-predictive control method has been presented as the
predictive control solution with similar performance to optimal-
predictive controllers but with a faster run-time digital imple-
mentation. It has been shown that the fast-predictive control
has a great potential to be used in power-quality applications,
having complex system models and using power-electronic
multilevel converters with a high number of levels.
REFERENCES
[1] J. Holtz, Power electronicsA continuing challenge, IEEE Ind. Elec-
tron. Mag., vol. 5, no. 2, pp. 615, Jun. 2011.
[2] A. Malinowski and H. Yu, Comparison of embedded system design for
industrial applications, IEEE Trans Ind. Informat., vol. 7, no. 2, pp. 244
254, May 2011.
[3] A. P. Kazmierkowski, M. Jasinski, and G. Wrona, DSP-based control of
grid-connected power converters operating under grid distortions, IEEE
Trans Ind. Informat., vol. 7, no. 2, pp. 204211, May 2011.
[4] E. Monmasson, L. Idkhajine, M. N. Cirstea, I. Bahri, A. Tisan, and
M. W. Naouar, FPGAs in Industrial control applications, IEEE Trans
Ind. Informat., vol. 7, no. 2, pp. 224243, May 2011.
[5] A. Malinowski and H. Yu, Comparison of embedded system design for
industrial applications, IEEE Trans Ind. Informat., vol. 7, no. 2, pp. 244
254, May 2011.
[6] E. Monmasson, L. Idkhajine, and M. W. Naouar, FPGA-based con-
trollers, IEEE Ind. Electron. Mag., vol. 5, no. 1, pp. 1426, Mar. 2011.
[7] B. Singh, S. Singh, A. Chandra, and K. Al-Haddad, Comprehensive
study of single-phase acdc power factor corrected converters with high-
frequency isolation, IEEE Trans Ind. Informat., vol. 7, no. 4, pp. 540
556, Nov. 2011.
[8] N. Yousefpoor, S. H. Fathi, N. Farokhnia, and H. A. Abyaneh, THD
minimization applied directly on the line-to-line voltage of multilevel
inverters, IEEE Trans. Ind. Electron., vol. 59, no. 1, pp. 373380,
Jan. 2012.
[9] D. Floricau, E. Floricau, and G. Gateau, New multilevel converters with
coupled inductors: Properties and control, IEEE Trans. Ind. Electron.,
vol. 58, no. 12, pp. 53445351, Dec. 2011.
[10] A. Nabae and I. Takahashi, Anewneutral-point-clamped PWMinverter,
IEEE Trans. Ind. Appl., vol. IA-17, no. 5, pp. 518523, Sep. 1981.
[11] T. A. Meynard and H. Foch, Multi-level choppers for high voltage appli-
cations, EPE J., vol. 2, no. 1, pp. 4550, Mar. 1992.
[12] M. Marchesoni, M. Mazzucchelli, and S. Tenconi, A non conventional
power converter for plasma stabilization, in Proc. IEEE Power Electron.
Spec. Conf., Apr. 1988, pp. 122129.
[13] D. G. Holmes and T. A. Lipo, Pulse Width Modulation for Power
Converters. Hoboken, NJ: Wiley, 2003.
[14] A. L. Batschauer, S. A. Mussa, and M. L. Heldwein, Three-phase hy-
brid multilevel inverter based on half-bridge modules, IEEE Trans. Ind.
Electron., vol. 59, no. 2, pp. 668678, Feb. 2012.
[15] M. C. Cavalcanti, A. M. Farias, K. C. Oliveira, F. A. S. Neves, and
J. L. Afonso, Eliminating leakage currents in neutral point clamped
inverters for photovoltaic systems, IEEE Trans. Ind. Electron., vol. 59,
no. 1, pp. 435443, Jan. 2012.
[16] M. P. Kazmierkowski, L. G. Franquelo, J. Rodriguez, M. A. Perez, and
J. I. Leon, High-performance motor drives, IEEE Ind. Electron. Mag.,
vol. 5, no. 3, pp. 626, Sep. 2011.
[17] M. H. Rashid, J. F. Silva, and S. F. Pinto, Power Electronics Handbook,
2nd ed. New York: Academic, 2006, ch. 34.
[18] E. R. C. Silva, E. C. Santos, and C. B. Jacobina, Pulsewidth modulation
strategies, IEEE Ind. Electron. Mag., vol. 5, no. 2, pp. 3745, Jun. 2011.
[19] A. Sabanovic, Variable structure systems with sliding modes in motion
controlA survey, IEEE Trans Ind. Informat., vol. 7, no. 2, pp. 212223,
May 2011.
[20] J. D. Barros and J. F. Silva, Optimal predictive control of three-phase
NPC multilevel converter for power quality applications, IEEE Trans.
Ind. Electron., vol. 53, no. 10, pp. 36703681, Oct. 2008.
[21] K. V. Ling, W. K. Ho, Y. Feng, and B. Wu, Integral-square-error perfor-
mance of multiplexed model predictive control, IEEE Trans Ind. Infor-
mat., vol. 7, no. 2, pp. 196203, May 2011.
[22] J. D. Barros and J. F. Silva, Multilevel optimal predictive dynamic volt-
age restorer, IEEE Trans. Ind. Electron., vol. 57, no. 8, pp. 27472760,
Aug. 2010.
[23] P. Corts, M. P. Kazmierkowski, R. M. Kennel, D. E. Quevedo, and
J. Rodrguez, Predictive control in power electronics and drives, IEEE
Trans. Ind. Electron., vol. 55, no. 12, pp. 43124324, Dec. 2008.
[24] J. Rodrguez, J. Pontt, C. A. Silva, P. Correa, P. Lezana, P. Corts, and
U. Ammann, Predictive current control of voltage source inverter, IEEE
Trans. Ind. Electron., vol. 54, no. 1, pp. 495503, Feb. 2007.
[25] M. A. Perez, J. Rodriguez, E. J. Fuentes, and F. Kammerer, Predic-
tive control of acac modular multilevel converters, IEEE Trans. Ind.
Electron., vol. 59, no. 7, pp. 28322839, Jul. 2012.
J. Dionsio Barros (S04M09) was born in
Madeira Island, Portugal, in 1974. He received the
Dipl.Ing. degree in systems and computer engi-
neering from the University of Madeira, Funchal,
Portugal, in 1998, and the M.Sc. and Ph.D. degrees in
electrical and computer engineering from the Tech-
nical University of Lisbon, Lisboa, Portugal, in 2002
and 2008, respectively.
He is currently an Assistant Professor with the
Exact Sciences and Engineering Competence Centre,
University of Madeira, Funchal, where he is also
a Researcher with the Center for Innovation in Electrical and Energy Engi-
neering. His main research interests are modeling, simulation, and control of
multilevel converters applied to power quality.
Dr. Barros is a member of the Ordem dos Engenheiros, Portugal.
J. Fernando A. Silva (M92SM00) was born in
Mono Portugal in 1956. He received the Dipl.Ing.
degree in electrical engineering and the Ph.D. and
Habil. degrees in electrical and computer engi-
neering from the Universidade Tcnica de Lisboa
(UTL), Lisbon, Portugal, in 1980, 1990, and 2002,
respectively.
Currently, he is an Associate Professor of power
electronics with the Energy group of the Depart-
ment of Electrical and Computer Engineering, UTL.
He teaches power electronics, control of switching
power converter systems, and power quality. As the Leader of the Power
Electronics and Power Quality group of the Center for Innovation in Electrical
and Energy Engineering (Cie
3
), his main research interests include modeling,
simulation, topologies, and advanced control of power-electronic systems and
power quality.
Dr. Silva is a member of the Ordem dos Engenheiros, Portugal.
lvio G. A. Jesus received the Dipl.Ing. degree in
systems and computers engineering and the M.Sc.
degree in telecommunications and networks engi-
neering from the University of Madeira, Funchal,
Portugal, in 2003 and 2011, respectively. He is cur-
rently working toward the Ph.D. degree in electrical
engineering in the Exact Sciences and Engineering
Competence Centre, University of Madeira
He is a Teacher of informatics in the Sec-
ondary School Padre Manuel lvares, Ribeira Brava,
Portugal. His main research interests include the
control of power-electronic technologies and renewable energy systems with
power-electronic converters.

You might also like