Professional Documents
Culture Documents
School of Electrical Engineering & Computer Science, The University of Newcastle, Australia;
e-mails: {raguilera, dquevedo}@ieee.org.
2
Depto. de Ingeniera Elctrica, Universidad Tcnica Federico Santa Mara, Valparaso, Chile; e-mail: pablo.lezana@usm.cl.
I. I NTRODUCTION
In general, model predictive control (MPC) for power
converters can be classied into two major categories
according to the nature of the system input, Explicit MPC
and Finite Control Set MPC (FCS-MPC) [1].
Explicit MPC [2] considers the duty cycle (or modulation index) as control input. Therefore, for this kind
of predictive strategy, the system input belongs to a
bounded continuous control set, e.g. d(t) [0, 1]. To track
references, a cost function which considers the tracking
error at each sampling instant is used. To obtain an
optimal input, the state-space is divided in several partitions, e.g. Ri Rn . Thus, an optimal local controller,
i (x), is obtained for each state-space partition Ri . This
procedure is carried out off-line. Afterwards, a lookup
table is used to implement the optimal controller to be
applied. Consequently, the on-line algorithm is focused
on determining which region Ri the system-state belongs.
Regarding the advantages of this predictive strategy,
Explicit MPC presents a good performance during the
steady state. In this case, a zero-average tracking error can
be achieved as well as a constant switching frequency in
the electrical variables, e.g. output converter voltage and
load current. Additionally, depending on the modulation
strategy, a x number of switch commutations can also
be obtained. The main drawback of this predictive formulation is related to the off-line computation complexity. The derived controller may consist of several local
controllers, yielding a lookup table with several entries.
To overcome this situation, some suboptimal solutions
DS3c.5-1
y {1, . . . , n}.
(1)
(8)
where
(x, u) = xT Qx + uT Ru,
is the stage cost, in which Q and R are positive denite
matrices, and
Vf (x) = xT P x,
(4)
(5)
(3)
the
(6)
u
(7)
(11)
(12)
DS3c.5-2
(13)
A. FCS-MPC of an FCC
As already mentioned, we rst need to obtain a
discrete-time model of the FCC. Considering the system
state as x[k] = [vc1 [k] vc2 [k] ia [k]]T and the control
input as u[k] = [S1 [k] S2 [k] S3 [k]]T , equations (2) to
(7) can be easily discretized and rewritten in a matrix
form (see [17], [18]) as:
x[k + 1] = Ax[k] + B(x[k])u[k],
where
1
A = 0
0
0
1
0
(14)
0
0 ,
ka
(19)
.
(16)
e[k] vc2 [k] vc2
iL [k] iL [k]
In this case, we chose a prediction horizon of N = 1.
Thus, the quadratic cost function (9) becomes:
V (x[k], u[k]) = e[k]T Qe[k] + e[k + 1]T P e[k + 1] (17)
(20)
where R = 0 and
K = (GT P G + R)1 GT P F.
P = Q = diag{1 , 2 , 1}.
Xf {x Rn : E (x[k]) U} .
(21)
DS3c.5-3
Table I
M AIN CONVERTER AND CONTROL PARAMETERS .
Now, we dene when to switch between the two predictive formulations in practice. At each sampling instant
k, based on the system error presented in (16), we evaluate
the following system deviation:
J[k] = e[k]T Qe[k].
If J[k] is higher than a lower bound, JL , i.e. J[k] >
JL , we will apply FCS-MPC to lead the system towards
the reference. Once, J[k] < JL , we will apply the LQR
presented in (20).
To avoid unnecessaries switching between these two
predictive techniques during a transition, the FCS-MPC
only will be applied again as soon as the system deviation
is higher than an upper bound, JH , i.e. J[k] > JH . Notice
that the upper bound, JH , is related to the terminal region,
Xf , in (21). Thus, it must be designed in order to satisfy
that when J[k] < JL = x[k] Xf .
Parameter
Value
Vdc
R
L
C1 =C2
fs
fc
400 V
40
10 mH
110 F
4000 Hz
1333 Hz
IV. R ESULTS
The most relevant converter and control parameters
are detailed in Table I. For the FCS-MPC strategy the
sampling time, h, is set as 250s, i.e. fs = 4 kHz.
Thus, for this predictive strategy, the maximum switch
commutation frequency and the output voltage switching
frequency is 2 kHz. On the other hand, to implement the
Explicit MPC solution, a PS-PWM along with the LQR is
considered. Here, the carrier frequency is set as fc = 1.33
kHz. Thus, each switch will commutate at fc . However,
as a positive consequence of using PS-PWM, the output
voltage switching frequency will be 4 kHz, see [22].
The weighting factors used by FCS-MPC are chosen
as 1 = 2 = 0.01. Thus, P = Q = diag{0.01, 0.01, 1}
and R = 0. Using these weighting matrices, we nally
obtain the optimal controller gain presented in (20) as
K = 6.5896.
A. Start-up performance
One of the most demanding test for a control scheme
of an FCC is the start-up process without pre-charging
the oating capacitors. Fig. 2 shows the performance of
a pure LQR+PWM scheme under this condition with an
output current reference given by:
ia = ia + ia
= 5 + 4 sin(250 t).
Under this control strategy, there is not a closed-loop
control of the ying capacitor voltages. Therefore, this
strategy relies in the natural balancing (open-loop control)
yielded by the modulation strategy PS-PWM. As can
be clearly observed in Fig. 2, those voltages tend to
their desired values but with a very low dynamic and a
signicant overshoot. Finally, after approximately 300ms,
the oating capacitor voltages reach their nominal values
of 266V and 133V respectively. As a consequence of the
start-up dynamic of the oating voltages, signicant highfrequency distortion is presented in the output current.
Nevertheless, its fundamental component follows the reference.
Figure 3. Start-up with the proposed scheme: (a) FCC inner voltages;
(b) Output current; (c) Output voltage; (d) Joint error function.
DS3c.5-4
Current reference
f S1
f S2
f S3
FCS-MPC
1A
1.5A
2A
2.5A
3A
3.5A
4A
4.5A
5A
1.49kHz
1.46kHz
1.39kHz
1.34kHz
1.09kHz
0.93kHz
1.1kHz
0.79kHz
0.65kHz
1.55kHz
1.52kHz
1.45kHz
1.43kHz
1.36kHz
1.1kHz
0.99kHz
0.75kHz
0.68kHz
1.61kHz
1.48kHz
1.44kHz
1.42kHz
1.2kHz
1.01kHz
0.83kHz
0.7kHz
0.59kHz
Prop. Scheme
Table II
AVERAGE S WITCHING F REQUENCIES FOR D IFFERENT C URRENT
R EFERENCE VALUES
Any
1.33kHz
1.33kHz
1.33kHz
Figure 5. Current reference step response:: (a) FCC inner voltages; (b)
Output current; (c) Output voltage; (d) Joint error function.
B. Steady-state performance
One of the main drawbacks of FCS-MPC is related
with its performance in the frequency domain. As shown
in Fig. 4 the output current spectrum obtained is wide
spread over the range 500-3000Hz, moreover, the spectrum signicatively changes with the current reference.
This can excite no modeled resonances, and make difcult
the design of lters to reduce the harmonic pollution. The
proposed technique uses a PS-PWM scheme in steady
state, therefore, the harmonics are well dened around
nfc , where n is the number of internal cells. In this case
n = 3 and fc = 1333Hz.
A signicant advantage of modulator-based control
strategies comes from the fact that the semiconductor
switching frequency is well known. In this case it corresponds to fc = 1.33kHz. Additionally, the switching
losses are equally spread among the semiconductors with
no additional effort. These characteristics are no present in
FCS-MPC unless additional modications in the quality
function are included.
Table II shows the average switching frequency, f Sx ,
for switches S1 S3 as a function of the current reference. As can be observed, S2 presents, in general, a
higher switching frequency when compared to S1 and S3 .
Moreover, these switching frequencies change depending
on the current reference magnitude.
Figure 6. Current reference step response: (a) FCC inner voltages; (b)
Output current; (c) Output voltage; (d) Joint error function.
DS3c.5-5
FCS-MPC
0.15
Proposed Scheme
0.15
i*=5+4sin(2p50t)
i =5+4sin(2p50t)
*
0.1
0.1
0.05
0.05
(a)
0.15
i =5+2sin(2p50t)
0.15
i =5+2sin(2p50t)
0.1
0.1
0.05
0.05
0
0.5
0.5
6
7
Frequency [kHz] (b)
Frequency [kHz]
Figure 4. Output current spectrum obtained with a standard FCS-MPC algorithm and the proposed scheme for: (a) i = 5 + 4 sin(250t); (b)
i = 5 + 2 sin(250t).
R EFERENCES
[1] P. Corts, M. P. Kazmierkowski, R. M. Kennel, D. E. Quevedo,
and J. Rodrguez, Predictive Control in Power Electronics and
Drives, Industrial Electronics, IEEE Transactions on, vol. 55,
no. 12, pp. 43124324, Dec. 2008.
[2] A. Linder and R. Kennel, Model Predictive Control for Electrical
Drives, in Power Electronics Specialists Conference, 2005. PESC
05. IEEE 36th, 2005, pp. 17931799.
[3] T. A. Johansen, I. Petersen, and O. Slupphaug, Explicit suboptimal linear quadratic regulation with state and input constraints, Automatica, vol. 38, pp. 10991111, 2002.
[4] N. Ameen, B. Galal, R. M. Kennel, and R. Kanchan, The
polynomial approximation of the explicit solution of model-based
predictive controller for drive applications, Predictive Control
of Electrical Drives and Power Electronics (PRECEDE), 2011
Workshop on, pp. 7681, 2011.
[5] A. Grancharova, T. A. Johansen, and P. Tndel, Computational
Aspects of Approximate Explicit Nonlinear Model Predictive Control, ser. Lecture Notes in Control and Information Sciences,
R. Findeisen, F. Allgwer, and L. T. Biegler, Eds.
Berlin,
Heidelberg: Springer Berlin Heidelberg, 2007, vol. 358.
[6] S. Kouro, P. Corts, R. Vargas, U. Ammann, and J. Rodriguez,
Model Predictive ControlA Simple and Powerful Method to
Control Power Converters, Industrial Electronics, IEEE Transactions on, vol. 56, no. 6, pp. 18261838, Jun. 2009.
[7] A. Linder and R. Kennel, Direct model predictive control - a
new direct predictive control strategy for electrical drives, in
Power Electronics and Applications, 2005 European Conference
on, 2005, p. 10.
[8] R. Vargas, U. Ammann, J. Rodriguez, and J. Pontt, Predictive
Strategy to Control Common-Mode Voltage in Loads Fed by
Matrix Converters, Industrial Electronics, IEEE Transactions on,
vol. 55, no. 12, pp. 43724380, 2008.
[9] R. Vargas, U. Ammann, and J. Rodriguez, Predictive Approach
to Increase Efciency and Reduce Switching Losses on Matrix
Converters, Power Electronics, IEEE Transactions on, vol. 24,
no. 4, pp. 894902, 2009.
[10] D. E. Quevedo, R. P. Aguilera, M. A. Prez, P. Corts, and
R. Lizana, Model Predictive Control of an AFE Rectier With
Dynamic References, Power Electronics, IEEE Transactions on,
vol. 27, no. 7, pp. 31283136, 2012.
[11] P. Lezana, R. Aguilera, and D. Quevedo, Steady-state issues
with nite control set model predictive control, in Industrial
Electronics, 2009. IECON 09. 35th Annual Conference of IEEE,
2009, pp. 17761781.
[12] J. Rawlings and D. Mayne, Model Predictive Control: Theory and
Design. Nob Hill Publishing, 2009.
[13] T. Meynard, M. Fadel, and N. Aouda, Modeling of multilevel
converters, Industrial Electronics, IEEE Transactions on, vol. 44,
no. 3, pp. 356364, 1997.
DS3c.5-6
DS3c.5-7