Sei sulla pagina 1di 8

Variable Sampling Time Finite Control-Set Model

Predictive Current Control for Voltage Source Inverters



Nils Hoffmann*, Markus Andresen*, Friedrich W. Fuchs*, Lucian Asiminoaei** and Paul B. Thgersen***
*Institute for Power Electronics and Electrical Drives
Christian-Albrechts-University of Kiel
D-24143 Kiel, Germany
nho@tf.uni-kiel.de
**Danfoss Solar Inverters A/S
DK-64000 Snderborg, Denmark
las@danfoss.com
***KK-Electronic A/S
DK-9220 Aalborg, Denmark
patho@kk-electronic.com

AbstractThis work introduces the control concept of variable
sampling time finite control-set model predictive control (FCS-
MPC). The new control concept is introduced in theory based on
a review of the conventional FCS-MPC concepts performed with
a constant sampling time. Based on the partitioning of the
sampling instant in multiple smaller sampling instants it is
possible to optimize, besides the switching states, the switching
states turn-on times. Therefore, the proposed variable sampling
time FCS-MPC sets both: the switching state and the related
turn-on times. To utilize the available calculation power for
longer sampling instants an adaptation of the control- and
prediction horizon to the sampling time is proposed. The
theoretical control concepts are applied to the control of a grid
connected two-level voltage-source converter where a simple L-
type linefilter is used to demonstrate the control performance of
the variable sampling time FCS-MPC algorithms in the
laboratory environment.
NOMENCALTURE
N
R
Control horizon
N
P
Prediction horizon
I
S
Sampling time
X|
k-1
Variable at sampling instant t = (k-1)Ts
X|
k
Variable at sampling instant t = kTs
X|
k+1
Variable at sampling instant t = (k+1)Ts
I. INTRODUCTION
In the last decades model predictive control (MPC)
emerged tremendously in the field of power electronic
engineering. The enormous success of MPC for power
electronic applications is traced back to the distinguished
features provided by this control approach. Compared to
conventional approaches MPC offers to include complex
constrained optimization problems, model nonlinearities and
discrete event-based problem formulations in the controller
design process. Furthermore, the achievable control
performance of MPC algorithms appears to be excellent and
very convincing in terms of reference- and disturbance
rejection as well as robustness in relation to model- and
parameter uncertainties [1,2].
One of the major concerns about using MPC in industrial
relevant power electronic applications is the high computation
burden that is inherent to the complex control formulation.
Nevertheless, academic and industrial researchers spent high
efforts to overcome this aforementioned issue making the MPC
approaches more practicable, more reliable and thus more cost
efficient to a wide range of power electronic applications [3,4].
MPC approaches are classified into continuous control-set
(CCS) MPC approaches and into finite control-set (FCS) MPC
approaches. In this context continuous control-set refers to a
theoretically infinite choice of possible control actuating
variables whereas finite control-set refers to a limited (finite)
choice of possible control actuating variables. In power
electronic applications CCS-MPC approaches are often used in
connection with modulators (e.g. pulse-width modulation) to
transform the reference output waveforms to the representative
switching states and duty cycles of the semiconductor devices
[5,6]. In contrast to that, FCS-MPC approaches are influencing
the switching states of power electronic converters directly.
The set of possible switching states, and thus of possible set of
output values for the control law formulation depends on the
used converter topology [7,8].
Another approach to classify MPC algorithms uses the way
how the control performance is optimized. The control
performance can be either optimized prior (offline) or during
(online) the real-time control calculations. MPC algorithms
that are using offline optimized control performances are
referred to as explicit MPC algorithms whereas MPC
algorithms that are using online optimization are referred to as
linear MPC algorithms. A study of recently published articles
about MPC reveals that explicit (CCS-) MPC algorithms are
mostly applied when more than one system state variable is
controlled which is especially interesting when systems with
resonances are studied [6,9]. Linear (FCS-) MPC is more likely
applied when complex converter topologies with a limited
number of system states are controlled [10-12].
This work focuses on linear FCS-MPC used for the current
control of voltage-source inverters (VSIs). In [7] the
underlying principle of FCS-MPC applied to the current
control for VSIs is explained. The performance of FCS-MPC
based current control is compared with hysteresis- and PWM-
based current control and it is concluded to be comparable very
well to these classical control methods. A detailed guideline for
the implementation of FCS-MPC to a state-of-the art DSP is
presented. The work presented in [13] basically continues this
work presenting a graph algorithm included to the FCS-MPC
978-1-4673-0803-8/12/$31.00 2012 IEEE 2215
based current controller to reduce the inverters switching losses
in steady state converter operation. The focus of [13] is set to
high power applications. The graph algorithm introduced to the
FCS-MPC based current control leads to a switching behavior
where only one inverter-leg is switched in each control period.
This work aims to contribute to the FCS-MPC based
current control of VSIs. In contrast to the FCS-MPC
algorithms presented in literature so far, where a fixed
sampling time and thus constant pulse-width resolution is used,
the proposed FCS-MPC varies the sampling time based on the
predicted system behavior and the applied cost functions
leading to variable pulse-widths applied by the FCS-MPC
based current controller. Therefore, compared to the
conventional FCS-MPC algorithms that are operated with a
constant sampling time the proposed FCS-MPC control that is
operated with a variable sampling time optimizes both, the
switching states and the related turn-on times. This new FCS-
MPC approach is referred to as variable sampling time FCS-
MPC. Further, this work presents a possibility to improve the
FCS-MPC control performance by including additional
sampling time adaptive control functions to utilize the
available computation power. The proposed variable sampling
time FCS-MPC approach is applied to the current control of a
grid connected two-level VSI with a simple L filter to
demonstrate the achievable control performance of variable
sampling time FCS-MPC. This converter topology is chosen
because of its industrial relevance and widespread applications.
However, the studies presented here are theoretically not
limited to this two-level inverter topology. The proposed
control principle can be applied to every voltage-source
inverter topology where an average output voltage level is
achieved by variation of the duty cycle ratios of switched
semiconductor devices such as switched-mode power supplies,
DC-DC converters or multi-level converters.
This work is structured as follows: In the second chapter a
system description is presented. The third chapter reviews the
conventional constant sampling time FCS-MPC approach and
the proposed variable sampling-time FCS-MPC approach is
introduced in the fourth chapter. A measurement study is
presented in the fifth chapter. The paper is closed by a detailed
conclusion presented in the last chapter.
Fig. 1: Block-diagram of three-phase three-wire grid connected voltage-
source converter with L filter type line-filter
II. SYSTEM DESCRIPTION AND CONTROL BLOCK DIAGRAM
In Fig. 1 the block diagram of a three-phase three-wire grid
connected voltage source inverter with an L filter is presented.
An active load is indicated emphasizing that the grid connected
converter operates under non constant, time dependent load
conditions.
The inverter is composed of three inverter phase legs
whereas each phase leg consists of two insulated gate bipolar
transistors (IGBTs) with anti parallel diodes. To avoid fault
conditions due to a short circuit in the inverters phase leg it is
not allowed switching the complementary IGBTs of each
phase-leg at the same time. Taking this into account, eight
possible switching states of the two-level VSI can be
identified. These eight possible switching states lead to seven
different output voltage states which are summarized in Table
I. Based on the corresponding output voltage level the
presented space vectors (SV) are separated into active state
(SV1 SV6) and zero state (SV0, SV7) space vectors.
TABLE I
SWITCHING STATE DEFINITIONS AND CONVERTER OUTPUT VOLTAGES
Switching
vector
Switch states
Converter output voltage
U

Conv
S1S4 S3S6 S5S2 UConv, UConv,
SV0 0 1 0 1 0 1 0 0
SV1 1 0 0 1 0 1 2/3 UDC 0
SV2 1 0 1 0 0 1 1/3 UDC 1/3 UDC
SV3 0 1 1 0 0 1 -1/3 UDC 1/3 UDC
SV4 0 1 1 0 1 0 -2/3 UDC 0
SV5 0 1 0 1 1 0 -1/3 UDC -1/3 UDC
SV6 1 0 0 1 1 0 1/3 UDC -1/3 UDC
SV7 1 0 1 0 1 0 0 0
Notation: 0 Switch Sx open, 1 Switch Sx closed

To demonstrate the principles of the proposed variable
pulse-width FCS-MPC approach the control scheme of the
voltage-oriented control (VOC) is used [14]. The VOC scheme
is composed of cascaded control-loops with an inner current
control loop and an outer DC link voltage control loop. By
aligning the controlled converter currents to the phase angle of
the fundamental grid voltage the active and the reactive power
drawn or fed-in to the mains can be controlled independently.
The phase angle of the grid voltage phasor is achieved by
means of phase locked loop (PLL) technique.
III. CONSTANT SAMPLING TIME FCS-MPC
Several approaches have been published introducing the
general idea of FCS-MPC [2,7,13,15-17] and specifically for
three-phase VSI applications [7,8,13,18-20]. To provide a
deeper understanding of the underlying principles of FCS-
MPC the application of a constant sampling time FCS-MPC
based current controller for a grid connected VSI is reviewed
in the next sections.
A. System equations of converter model
A three-phase three-wire grid connected VSI as it has
already been illustrated in Fig. 1 is considered. Applying
Kirchhoffs law to the L filter type line-filter and using the
Clarke () transformation the line side filter dynamics are
expressed as presented in (1). (X denotes a space vector)
2216
J
Jt
I
Con
u[
(t) =
1
I
L
[u
Lnc
u[
(t) -u
Con
u[
(t) -R
L
I
Con
u[
(t)
(1)
Using the Park (dq) transformation to calculate the line
filter dynamics in the rotating dq reference frame (with the
angular line-frequency
Line
) the equation (2) holds place
whereas a coupling term between the d- and q-current
components occurs.
J
Jt
I
Con
dq
(t) =
1
I
L
[u
Lnc
dq
(t) -u
Con
dq
(t) -R
L
I
Con
dq
(t)
- ]
Lnc
I
Con
dq
(t)
(2)
For the sake of simplicity it is assumed that the converter
currents I
Conv
, and the line voltages U
Line
are measured. Thus,
every systems state of (2) is known except the converter output
voltages U
Conv
. However, utilizing the principle of a finite
control-set with respect to the possible switching states of the
two-level VSI (see Table I) the converters output voltage can
be expressed dependent on the inverters switching state and the
DC link voltage U
DC
(3) where the switching states S
1
, S
3
and
S
5
address the calculated switching states of the former
calculation instant (here: t = kT
s
). (Notation of switching
states: see Table I)
_
u
Con,u
(k)
u
Con,[
(k)
_ = _
2
S
, -
1
S
, -
1
S
,
u
1
S
, -
1
S
,
_ u
C
_
S
1
(k -1)
S
3
(k - 1)
S
5
(k - 1)
_
(3)
B. Prediction of system states
The system equations of the grid connected VSI (1)-(3) are
now used to achieve a prediction equation of the controlled
system states. A convenient state space model formulation of
the d- and q-current line filter dynamics is used for illustration
purposes (4).
J
Jt
_
i
Con
d
(t)
i
Con
q
(t)
_ =
l
l
l
l -
R
L
I
L

Lnc
-
Lnc
-
R
L
I
L
1
1
1
1
_
i
Con
d
(t)
i
Con
q
(t)
_ +
1
I
L
_
u
Inc
d
(t)
u
Inc
q
(t)
_
-
1
I
L
_
u
Con
d
(t)
u
Con
q
(t)
_
(4)
Using explicit Euler approximation to derive the derivative
converter current dynamics between two sampling instants the
discrete prediction equation is calculated (5).
_
i
Con
d
(k + 1)
i
Con
q
(k + 1)
_ =
l
l
l
l1 - I
S
R
L
I
L

Lnc
I
S
-
Lnc
I
S
1 - I
S
R
L
I
L
1
1
1
1
_
i
Con
d
(k)
i
Con
q
(k)
_
+
I
S
I
L
_
u
Lnc
d
(k)
u
Lnc
q
(k)
_ -
I
S
I
L
_
u
Con
d
(k)
u
Con
q
(k)
_
(5)
The current evolution for each possible switching state of
the two-level VSI can be predicted combining (3) and (5). The
prediction principle is used in an iterative manner leading to
current predictions for multiple sampling instants to the future.
C. Cost function and switching optimization
The presented prediction equations are used to optimize the
control performance. Different optimization goals can be
included in the switching state optimization of FCS-MPC
based current controllers [17]. The optimization goals are
dependent on the specific control performance requirements set
by the converter application. The optimization goal of this
work is set to switching loss reduction with an acceptable
range of harmonic current distortion and reference current
tracking in the same time. To achieve this optimization goal a
quadratic cost function is introduced to the switching state
optimization (6). The future (predicted) current control
deviations are used to evaluate the costs of each possible
switching state:
C = |i
Con
d
(k + i) -i
Con
d,c]
(k +i)]
2
+|i
Con
q
(k +i) - i
Con
q,c]
(k + i)]
2

(6)
D. Switching frequency reduction due to switching state
restrictions
The former paragraphs present the principles of FCS-MPC
operating with constant sampling time. Basically, an
appropriate choice of the cost-function is used to optimize the
control performance. The converters topology, i.e. the possible
and allowed switching state combinations, is considered in the
prediction law formulation.
In [13] another possibility to influence the switching
performance of the chosen two-level VSI converter topology is
presented. A switching state flow graph is included to the
current prediction algorithms to further manipulate the
inverters switching behavior. The related switching state flow-
graph is depicted in Fig. 2. The flow graph is designed to allow
only one switching transition at maximum in all of the three
inverter legs in each sampling period (i.e. sampling time).
Fig. 2 Switching state flow-graph to allow at maximum one switching
transition in sampling instant based on [13]
As aforementioned the control performance goal is set to
switching loss reduction with an acceptable range of harmonic
current distortion and reference current tracking in the same
time. To achieve this goal a quadratic cost function is used. To
further reduce the inverters switching losses with an acceptable
amount of harmonic current distortion the discussed flow graph
is used for the converter output current prediction.
E. Calculation burden
The presented principles of FCS-MPC reveal a significant
amount of calculations involved. To reduce the basic
calculation burden of these FCS-MPC algorithms different
approaches have been presented in literature [4,19]. An
appropriate choice of the reference frame (here the rotating dq
reference frame is chosen) for the FCS-MPC control problem
2217
formulation [19] as well as an adaptation of mathematical
programming techniques [4] reduces the basic calculation
burden. Here a two-level VSI is used with a switching state
flow graph that allows one switching transition at maximum in
all of the three inverter legs for each sampling period. Thus
four possible switching states have to be considered in each
sampling time instant for the current prediction and cost
function evaluation. This leads to the basic calculation burden
presented in (7).
0(N
R
, N
P
) = 4
N
R
+4 (N
P
-N
R
) (7)
IV. VARIABLE SAMPLING TIME FCS-MPC
The principles of constant sampling time FCS-MPC based
current control are reviewed in the former paragraphs. Except
the FCS-MPC based concept of model predictive direct torque
control proposed by Geyer et. al. [15,16] (here the switching
performance is optimized based on state extrapolation with
given switching boundaries) these reviewed methods have in
common that the system is sampled and the switching states
are updated with a fixed sampling time. Thus, giving a
maximum control horizon N
R
and prediction horizon N
P
, the
voltage pulse width set by the FCS-MPC controller is restricted
by the applied sampling time resolution. When the control
algorithms are executed with a fixed sampling time, the
switching states are optimized in each sampling instant leading
to: First, a high computational burden even when no switching
transitions occur between multiple sampling periods and
second, unnecessary (non-optimal) switching transitions that
occur because of model- and parameters uncertainties when the
receding horizon concept is utilized. To overcome these
drawbacks of a constant sampling time FCS-MPC a variable
sampling time FCS-MPC is proposed. The variation of the
sampling time is included in the FCS-MPC prediction law
formulation leading to a new degree of freedom for control
performance optimization during online control processing.
A. Concept
In Fig. 3 the functional principle of the proposed variable
sampling time FCS-MPC is highlighted. The functional
concept of this control is explained in five steps:
Step 1: At an arbitrary sampling instant k the system states
are sampled and the systems switching state S|
(k-1)
is updated
due to the FCS-MPC calculations of the former sampling
instant k-1. Then, for a given minimum sampling time T
s,min

and maximum sampling time T
s,max
, the next sampling instant
k+1 is divided in multiple partitions. In the exemplary
illustration the sampling instant k+1 is divided into five
equidistant partitions (L = 5).
Step 2: For a given control horizon N
R
|
k
and prediction
horizon N
P
|
k
(which was set in the former sampling instant k-1)
the converter currents are predicted using the same principles
that have already been introduced for constant sampling time
FCS-MPC but in this case for the different possible switching
instant partitions L. The fragmentation of the maximal possible
sampling time into multiple smaller partitions is used to
optimize the turn-on times in addition to the switching state
optimization. The related costs for each of these predicted
currents are then evaluated for every possible partition and
switching state. To reduce the calculation burden only the
switching patterns with a minimum cost for each considered
partition are compared with the minimum costs of every other
considered partition.
Step 3: In the next step the system switching state and the
related turn-on time is updated based on the calculated optimal
switching pattern and the related turn-on times. Only the first
switching state is applied. Since the related turn-on time is
optimized and thus known in advance also the counter value
for the hardware interrupt to set the next (optimized) sampling
instant is set by the FCS-MPC based current controller.
Step 4: The calculation steps of the FCS-MPC with
variable sampling time presented so far reveal that besides the
optimal switching pattern also the turn-on times for each
switching state are optimized. Thus, due to the predictive
nature of the FCS-MPC algorithm, the switching state and the
turn-on time for the next (future) sampling instant is known in
advance. This knowledge is used to utilize the available
calculation power for the varying sampling times. For
increasing sampling times also the available calculation time
increases. Thus, more calculations can be performed for these
increased sampling times. Here, the prediction and the control
horizon are varied based on the available calculation time.
Fig. 3: Functional principle of FCS-MPC with variable sampling
time
2218
Step 5: To realize a control state feedback the receding
horizon concept is utilized: In the next sampling instant only
the first optimized switching state and the related turn-on time
is applied to the system. Then the steps 1 to 4 are performed
again based on the sampled system states for this next
sampling instant.
B. Partitioning of sampling instants
The presented idea of the partitioning of a given maximum
sampling time T
s,max
into multiple smaller sampling instants
provides the possibility to optimize the switching states and the
related turn-on times for each switching state. By predicting
the current waveforms for each possible switching state and
each possible turn-on time the control performance can be
optimized by evaluating the related costs for each of these
switching states and turn-on times. From the whole set of
possible switching states and turn-on times that switching
sequence is chosen that minimizes the related cost function
which thus leads to an optimal control performance. In the
chosen example the partitioning of the maximum allowed
sampling time is chosen to be equidistant. However, the
partitioning is not limited to be equidistant. Different non-
equidistant partitioning patterns are possible and the
appropriate choice is dependent on the specific requirements
set by the application of the power converter. Further, the
selection of the minimum and maximum sampling times and
the number of partitions is only restricted by the available
online calculation power. Compared to a FCS-MPC performed
with a constant sampling time this variable sampling time
offers an additional degree of freedom for control performance
optimization.
C. Sampling time adaptive control functions
Due to the variable sampling time and the predictive nature
of the proposed FCS-MPC the sampling time of the next
(future) sampling periods is known in advance. Assuming a
fixed calculation power, an increased sampling time leads to an
increased calculation time and thus to a higher number of
calculations that have to be performed in one sampling period.
This knowledge is used to utilize the available calculation
power by including sampling time adaptive control functions.
One intuitive choice is to adapt the control horizon to the
variation of the sampling time (i.e. to the number of partitions
L). For longer sampling periods a higher control horizon is
applied to improve the quality of the predictions. For shorter
sampling periods a lower control horizon has to be applied for
the FCS-MPC calculations. The sampling time adaptive
adjustment of the control horizon for the system used in this
work is summarized in (8).
N
R
= _
1 i I = 1
2 i I 2

(8)
In this work the focus of the control performance
optimization is set to the minimization of the switching losses.
However, different sampling time adaptive control functions
can be implemented to the variable sampling time FCS-MPC
to utilize the available calculation power. In [21] a FCS-MPC
based selective harmonic elimination for multilevel converters
is introduced. There, the proposed algorithm uses sliding
discrete Fourier transformations to calculate the individual
harmonic components that are optimized by the FCS-MPC
controller in real time. For this control approach an adaptation
of the harmonic components considered for the spectral
analysis is another promising example for a sampling time
adaptive control function.
D. Calculation burden
The presented principle of variable sampling time FCS-
MPC uses a partitioning of the predicted system states between
a minimal and maximal sampling time. A sampling frequency
adaptive adaptation of the control horizon is proposed to utilize
the available calculation power. The resulting calculation costs
0 are summarized in (9).
0(N
R
, N
P
, I) = I|
k
4
N
R
|
k
+I|
k
4 (N
P
|
k
- N
R
|
k
) (9)
Compared with the calculation costs presented for the
constant sampling time FCS-MPC the calculation costs for the
variable sampling time FCS-MPC are increased by the number
of partitions used. When a sampling time adaptive adjustment
of the control horizon is implemented, the resulting calculation
costs are dependent on the number of partitions and vary
during the real-time control calculations.
TABLE II
MEASUREMENT SYSTEM PARAMETERS
Symbol Quantity Value (per unit)
ULL Line-to-Line Voltage (rms) 400 V (1.0)
UDC DC-link voltage 650 V (1.63)
Angular line frequency 2 50 Hz (1.0)
IL Rated converter current (rms) 31 A (1.0)
LL Filter inductance 5 mH (0.21)
RL Filter resistance 50 m (0.01)
CDC DC-link capacitance 2200 F (5.0)
TS Sampling time (variable) 40 s 200 s
V. MEASUREMENT STUDY
A measurement study is carried out to examine the variable
sampling time FCS-MPC under laboratory conditions. To
compare the achievable control performance of the proposed
variable sampling time FSC-MPC with the conventional
constant sampling time FCS-MPC additional measurements
using identical system conditions are carried out. To study the
effect of the proposed sampling time adaptive control functions
the measurements for the variable sampling time FCS-MPC
are presented with and without adaption of the control and
prediction horizon to the variable sampling time.
A. Test-setup description
A 22 kVA laboratory system is used to verify the proposed
control concepts. The system is composed of a grid and a
2219
motor connected two-level VSI. An L filter is used to connect
the grid side converter to the mains. The control algorithms are
implemented on dSpace DS1006 board where custom written
(DWO, digital waveform output) codes are used to generate the
pulse signals and the variable (hardware interrupt-based)
sampling time for the control of the grid side converter. The
general system parameters are summarized in Table II. All
presented measurements are performed on the same
experimental setup without changing the system parameters.
The sampling time for the constant sampling time FCS-MPC is
set equal to the minimum sampling time of the variable
sampling time FCS-MPC (here: T
s,min
= 40s).
B. Steady-state control performance
In Fig. 4 the measurement results for the steady-state
performance of constant sampling time FCS-MPC are
presented. The measured converter current waveforms and the
related switching pattern of each inverter leg (S
a
,S
b
,S
c
) are
highlighted in Fig. 4 (a) and (b). A total harmonic current
distortion (THD
I
) of 3.4 % and an effective switching
frequency (average switching frequency over one fundamental
period) of 2.93 kHz are measured.
In Fig. 5 the measurement results for the variable FCS-
MPC without using sampling time adaptive control functions
are presented. For these measurements a minimal sampling
time T
s,min
of 40 s and a maximum sampling time of T
s,max
of
200 s with a partitioning L of 5 is chosen. The measured
current waveforms and the switching patterns reveal a THD
I
of
3.75 % and an effective switching frequency of 2.87 kHz, see
Fig. 5 (a),(b). In Fig. 5 (c) the measured variation of the
sampling time is presented. The maximum sampling time (L =
Ts/T
s,min
= 5) is not reached in the presented experimental
results. The sampling time is increased every time two current
phases cross each other.
In Fig. 6 the measurement results for the variable FCS-
MPC with the proposed adaptation of the control and
prediction horizon to the variable sampling time is presented.
The measured current waveforms and the switching patterns
reveal a THD
I
of 3.85% and an effective switching frequency
of 2.85kHz, see Fig. 6 (a),(b). In Fig. 5 (c) the measured
variation of the sampling time is presented. The maximum
sampling time (L = Ts/T
s,min
= 5) is reached only once in the
presented experimental results. The control- and prediction are
varied based on the applied sampling time; see Fig. 6 (d).
The presented measurement results of the steady state
performances of the constant and variable sampling time FCS-
MPC based current control reveal that it is possible to reduce
the effective switching frequency of the inverter by an and the
related turn-on times. Further, the measurement results proof

(a)

(b)

(c)
N
R
, N
P

(d)
Fig. 4: Measurement results: steady-state analysis
of FCS-MPC with constant sampling time: (a)
converter currents, (b) switching pattern, (c)
sampling frequency and (d) control horizon

(a)
(b)
(c)
(d)
Fig. 5: Measurement results: steady-state analysis
of FCS-MPC with variable sampling time: (a)
converter currents, (b) switching pattern, (c)
sampling frequency and (d) control horizon

I
c
o
n
v
,a
b
c
[
A
]
(a)
(b)
(c)
(d)
Fig. 6: Measurement results: steady-state analysis
of FCS-MPC with variable sampling time and
sampling time dependent control horizon
adaptation: (a) converter currents, (b) switching
pattern, (c) sampling frequency and (d) control
horizon
2220
the proposed concept of applying sampling time adaptive
control functions (here the adaptation of the control and
prediction horizon to the sampling time) to be a powerful tool
to improve the control performance of variable sampling time
FCS-MPC. The effect of the sampling time adaptive variation
of the prediction horizon can be seen by comparing the
measured effective switching frequencies and the THD
I
of the
three concepts. While the effective switching frequency is
decreased from the FCS-MPC with constant sampling time to
the FCS-MPC with variable sampling time without and with
adaptation of the control/prediction horizon the THD
I
values
are increased respectively. Since the three control algorithms
are performed with the same (minimum) sampling time, the
effect of the switching state and turn-on time optimization can
be seen from the decreased effective switching frequencies.
The effect of the sampling time adaptive adjustment of the
control/prediction horizon can be seen by an additional
reduction of the effective switching frequencies.
C. Dynamic control performance
To study the dynamic performance of the three control
concepts additional measurements are carried out. For each of
these measurements reactive power reference steps were
performed and the related q-current component step responses
are evaluated concerning the measured settling times t
st
and
number of switching transitions needed to adjust the reference
values.
The step response (reactive current step of 20 A) and the
related switching pattern (S
a
,S
b
,S
c
) of each inverter leg of the
constant sampling time FCS-MPC are presented in Fig. 7 (a)
and (b). A settling time of 0.32 ms is measured where 6
switching transitions are needed to adjust the reference value.
In Fig. 8 the dynamic performance of the FCS-MPC based
current controller with variable sampling time and without an
adaptation of the control/prediction horizon is presented. The
measured step response, see Fig. 8 (a), leads to a settling time
of 0.32 ms where 2 switching transitions are needed to adjust
the new reference value, see Fig. 8 (b). From Fig. 8 (c) it can
be seen that the maximum sampling time and thus the
maximum amount of partitions (here L= 5) is used by the
FCS-MPC controller to optimize the dynamic control
performance.
Fig. 9 presents the dynamic measurement results of the
proposed variable sampling time FCS-MPC with an

(a)

(b)

(c)

(d)
Fig. 7: Measurement results: dynamic analysis of
FCS-MPC with constant sampling time: (a)
converter currents in dq-frame, (b) switching
pattern, (c) sampling frequency and (d) control
horizon

2 1 0 1 2 3 4
10
0
10
20
30
40
t [ms]
iconv,q
iconv,d
tst = 0.32ms
(a)
(b)
(c)
(d)
Fig. 8: Measurement results: dynamic analysis of
FCS-MPC with variable sampling time: (a)
converter currents in dq-frame, (b) switching
pattern, (c) sampling frequency and (d) control
horizon

(a)
(b)
(c)
(d)
Fig. 9: Measurement results: dynamic analysis of
FCS-MPC with variable sampling time and
sampling time dependent control horizon
adaptation: (a) converter currents in dq-frame, (b)
switching pattern, (c) sampling frequency and (d)
control horizon
2221
adaptation of the control/prediction-horizon to the sampling
time. The step response analysis reveals a sampling time of
0.24 ms, see Fig. 9 (a) with 2 switching transitions needed to
adjust the reference value; see Fig. 9 (b). Further, the
maximum sampling time is fully utilized during the reference
step; see Fig. 9 (c), which results from an increased control-
and prediction horizon adapted to utilize the available
calculation power.
The presented measurement results of the dynamic
performance of the variable sampling time FCS-MPC with
adaptation of the control /prediction horizon reveal an
improved dynamic control performance in relation to the
settling time and number of switching transitions needed to
adjust the reference value in comparison to the constant
sampling time FCS-MPC.
VI. CONCLUSION
The concept of variable sampling time FCS-MPC is
introduced in this work. The new control concept is
introduced in theory and compared to the conventional
constant sampling time FCS-MPC based control algorithms.
Based on a partitioning of the sampling instants into multiple
smaller sampling instants it is possible to optimize, besides the
switching states, the related switching turn-on times. To
utilize the available calculation power when longer sampling
instants are applied to the system an adaptation of the control-
and prediction horizon to the sampling time is proposed. The
theoretical control concepts are applied to the control of a grid
connected two-level voltage source converter where a simple
L filter is used to demonstrate the control performance of the
variable sampling time FCS-MPC algorithms. The presented
measurements, performed on a 22 kVA laboratory system,
reveal the excellent control performance of the proposed
control concept, especially during dynamic control conditions.
ACKNOWLEDGMENT
This work has been financed by the Ministry of Schleswig-
Holstein and the European Union and is operated under
Cewind e.G. Center of Excellence for Wind energy Schleswig-
Holstein.
REFERENCES
[1] P. Cortes, M. P. Kazmierkowski, R. M. Kennel, D. E. Quevedo, and J.
Rodriguez, "Predictive Control in Power Electronics and Drives," Industrial
Electronics, IEEE Transactions on, vol. 55, no. 12, pp. 4312-4324, Dec.2008.
[2] S. Kouro, P. Cortes, R. Vargas, U. Ammann, and J. Rodriguez, "Model
Predictive Control - A Simple and Powerful Method to Control Power
Converters," Industrial Electronics, IEEE Transactions on, vol. 56, no. 6, pp.
1826-1838, June2009.
[3] R. P. Aguilera and D. E. Quevedo, "On stability of Finite Control Set
MPC strategy for Multicell Converters," in Proc. Industrial Technology
(ICIT), 2010 IEEE International Conference on, 2010, pp. 1277-1282.
[4] T. Geyer, "Computationally Efficient Model Predictive Direct Torque
Control," Power Electronics, IEEE Transactions on, vol. 26, no. 10, pp. 2804-
2816, Oct.2011.
[5] H. Abu-Rub, J. Guzinski, Z. Krzeminski, and H. A. Toliyat, "Predictive
current control of voltage-source inverters," Industrial Electronics, IEEE
Transactions on, vol. 51, no. 3, pp. 585-593, June2004.
[6] S. Mariethoz and M. Morari, "Explicit Model-Predictive Control of a
PWM Inverter With an LCL Filter," Industrial Electronics, IEEE
Transactions on, vol. 56, no. 2, pp. 389-399, Feb.2009.
[7] J. Rodriguez, P. Jorge, A. S. Csar, C. Pablo, L. Pablo, C. Patricio, and
A. Ulrich, "Predictive Current Control of a Voltage Source Inverter,"
Industrial Electronics, IEEE Transactions on, vol. 54, no. 1, pp. 495-503,
Feb.2007.
[8] T. Geyer, "Model Predictive Direct Current Control for multi-level
converters," in Proc. Energy Conversion Congress and Exposition (ECCE),
2010 IEEE, 2010, pp. 4305-4312.
[9] M. Cychowski, K. Szabat, and T. Orlowska-Kowalska, "Constrained
Model Predictive Control of the Drive System With Mechanical Elasticity,"
Industrial Electronics, IEEE Transactions on, vol. 56, no. 6, pp. 1963-1973,
June2009.
[10] R. Vargas, U. Ammann, B. Hudoffsky, J. Rodriguez, and P. Wheeler,
"Predictive Torque Control of an Induction Machine Fed by a Matrix
Converter With Reactive Input Power Control," Power Electronics, IEEE
Transactions on, vol. 25, no. 6, pp. 1426-1438, June2010.
[11] R. Vargas, P. Cortes, U. Ammann, J. Rodriguez, and J. Pontt,
"Predictive Control of a Three-Phase Neutral-Point-Clamped Inverter,"
Industrial Electronics, IEEE Transactions on, vol. 54, no. 5, pp. 2697-2705,
Oct.2007.
[12] T. Geyer and S. Mastellone, "Model predictive direct torque control of
a five-level ANPC converter drive system," in Proc. Energy Conversion
Congress and Exposition (ECCE), 2011 IEEE, 2011, pp. 363-370.
[13] M. Preindl, E. Schaltz, and P. Thogersen, "Switching Frequency
Reduction Using Model Predictive Direct Current Control for High-Power
Voltage Source Inverters," Industrial Electronics, IEEE Transactions on, vol.
58, no. 7, pp. 2826-2835, July2011.
[14] M. P. Kazmierkowski, R. Krishan, and F. Blaabjerg, Control in Power
Electronics - Selected Problems. London: Academic Press - An imprint of
Elsevier Science, 2002.
[15] T. Geyer, G. Papafotiou, and M. Morari, "Model Predictive Direct
Torque Control Part I: Concept, Algorithm, and Analysis," Industrial
Electronics, IEEE Transactions on, vol. 56, no. 6, pp. 1894-1905, June2009.
[16] G. Papafotiou, J. Kley, K. G. Papadopoulos, P. Bohren, and M. Morari,
"Model Predictive Direct Torque Control Part II: Implementation and
Experimental Evaluation," Industrial Electronics, IEEE Transactions on, vol.
56, no. 6, pp. 1906-1915, June2009.
[17] P. Cortes, J. Rodriguez, R. Vargas, and U. Ammann, "Cost Function-
Based Predictive Control for Power Converters," in Proc. IEEE Industrial
Electronics, IECON 2006 - 32nd Annual Conference on, 2006, pp. 2268-
2273.
[18] H. Miranda, R. Teodorescu, P. Rodriguez, and L. Helle, "Model
predictive current control for high-power grid-connected converters with
output LCL filter," in Proc. Industrial Electronics, 2009. IECON '09. 35th
Annual Conference of IEEE, 2009, pp. 633-638.
[19] S. Vazquez, J. I. Leon, L. G. Franquelo, J. M. Carrasco, E. Dominguez,
P. Cortes, and J. Rodriguez, "Comparison between FS-MPC control strategy
for an UPS inverter application in alpha/beta and abc frames," in Proc.
Industrial Electronics (ISIE), 2010 IEEE International Symposium on, 2010,
pp. 3133-3138.
[20] M. Preindl and E. Schaltz, "Sensorless Model Predictive Direct Current
Control Using Novel Second-Order PLL Observer for PMSM Drive
Systems," Industrial Electronics, IEEE Transactions on, vol. 58, no. 9, pp.
4087-4095, Sept.2011.
[21] S. Kouro, B. La Rocca, P. Cortes, S. Alepuz, W. Bin, and J. Rodriguez,
"Predictive control based selective harmonic elimination with low switching
frequency for multilevel converters," in Proc. Energy Conversion Congress
and Exposition, 2009. ECCE 2009. IEEE, 2009, pp. 3130-3136.

2222

Potrebbero piacerti anche