Professional Documents
Culture Documents
Control
A Review of Its Applications in Power Electronics
M
odel-based predictive
control (MPC) for power
converters and drives is
a control technique that
has gained attention in
the research communi-
ty. The main reason for
this is that although MPC presents high com-
putational burden, it can easily handle mul-
tivariable case and system constraints and
nonlinearities in a very intuitive way. Taking
advantage of that, MPC has been success-
fully used for different applications such as
an active front end (AFE), power converters
connected to resistor–inductor RL loads,
uninterruptible power supplies, and high-
performance drives for induction machines,
among others. This article provides a review
of the application of MPC in the power elec-
tronics area.
MPC presents a dramatic advance in the
theory of modern automatic control [1]. MPC
was originally studied and applied in the pro-
cess industry, where it has been in use for
decades [2]. Now, predictive control is being
considered in other areas, such as power
electronics and drives [3]–[6]. The reason
for the growing interest in the use of MPC in
this field is the existence of very good mathe-
matical models to predict the behavior of the
©Artville, LLC.
Load
iLc L vSc vdc
– iLc L vSc
∗ P ∗(k + 2)
vdc +
PI
n Sopt
Q ∗(k + 2) n
FIGURE 6 – The power circuit of the AFE.
P p(k + 2) Minimization
vS (k )
Predictive of
J
iL (k ) Model Q p(k + 2) Cost Function
wind turbine for grid integration of J
renewable energies.
The main objective of the control FIGURE 7 – A block diagram of the FCS-MPC control strategy for the AFE.
strategy is to calculate the output in-
verter voltage, v I, abc, to regulate the
output dc-link capacitor voltage to a seven output voltage predictions are delay of the digital implementation of
reference, v *dc, for any connected load calculated, the cost function the control strategy.
and inject into the grid any reactive 2 Another way to perform predic-
g = ^ P * (k + 2) - P p (k + 2) h
power command reference, q *. 2 tive control for the AFE is by using
+ ^Q * (k + 2) - Q p (k + 2)h (2)
There are several alternatives for the P-DPC strategy. The block diagram
designing the control algorithm for is minimized to find the inverter out- of the P-DPC strategy is presented in
an AFE. In general, a cascade control put vector that should be applied in Figure 9. Like GPC, the P-DPC strat-
structure is used. An external control the next sampling period. egy does not take into account the
loop is employed to regulate the dc- Figure 8 presents the experimental switching of power semiconductors;
link voltage. On the other hand, an in- results obtained using this strategy therefore, it provides an exact solu-
ternal control loop is adopted to track [14]. It should be noted that predic- tion to an approximated optimization
the grid currents or the instantaneous tions in instants (k + 2) are used to problem. In addition, P-DPC consid-
active and reactive power references compensate for the control action ers an unconstrained MPC problem.
regarding the states variables used to
develop this controller [18], [19].
MPC has mainly been used as a con-
trol strategy for the inner control loop. 180
Although some works developing grid 170
160
current controllers can be found in the
Voltage (V)
150 vdc
literature, the main approach has been 140
130
the direct power control (DPC) for 120 ∗
vdc
tracking the commands for the instan- 110
100
taneous active and reactive powers, 90
P and Q. The application of FCS-MPC- Time (s)
DPC and predictive DPC (P-DPC) with 400
300
Qs (VAR), Ps (W)
Load
vdc vdc
– iLc L vSc plication, because for grid-connected
∗
vdc + P ∗(k + 2) converters exist highly demanding
PI
Sabc codes that impose strict limits to the
Q ∗(k + 2) n low-order harmonics that can be in-
p(k jected into the grid. FCS-MPC presents
vS (k ) P + 2) vI, abc
P-DPC variable switching frequency; thus, the
Predictive J
iL (k ) Model Q p(k + 2) grid current has a widespread harmon-
PWM-SVM ic spectrum. On the other hand, P-DPC
J
provides constant switching frequen-
FIGURE 9 – A block diagram of the P-DPC strategy for the AFE. cy; thus, the grid current harmonic
spectrum is concentrated around the
switching frequency, which decreases
Thus, an explicit solution can be ob- the switching vectors and the switch- the cost of the output L filter. Figure 10
tained provided the control action ing times; thus, a PWM-SVM modula- shows the experimental results ob-
is applied once the cost function (2) tion strategy is necessary to generate tained using the P-DPC strategy for a
is minimized. Therefore, an optimal the firing pulses. STATCOM application when an instan-
switching vector sequence can be cal- Compared with FCS-MPC, the P-DPC taneous reactive power command
culated. The control strategy provides algorithm uses an external modulator; step is imposed [22].
It should be noted that the outer
control loop to regulate the dc-link ca-
pacitor voltage is usually solved using
800
a conventional proportional-integral
700
600 (PI) controller. However, there are
500 some solutions that replace the PI con-
vdc (V)
5
of series inductors, as shown in Fig-
0 ure 11 (referring to the most common
−5 three-wire configuration without neu-
tral). In such a configuration, ideally,
−10
the APF is able to operate as a control-
0.45 0.47 0.49 0.51 0.53 0.55 lable current generator, drawing from
Time (s)
the mains any set of current wave-
(b)
forms having a null sum. Therefore, an
400
300 vSa iLa × 10 APF is ideally able to compensate the
vSa (V), iLa × 10 (A)
m1 ^ u * p 2
g= v dc (k + 2) - v dc (k + 2) h
v dcR
m2 ^ * p 2 iLa L vSa iSa
+ P S (k + 2) - P S (k + 2) h vdc
PSR iLb L vSb iSb
Sopt(k)
m3 ~ iLc L vSc iSc
+ ^Q *S (k + 2) - Q Sp (k + 2) h2, (3) vdc∗ (k + 2)
PSR vdc∗(k)
ila ilb ilc
Reference
where v *dc, P *S , and Q *S are the refer-
vdc(k) Design P ∗(k + 2) Sopt
ence values of dc-link voltage, active
power, and reactive power, respec- Nonlinear
p p p Load
tively; v dc , P S , and Q S are the pre-
vS, abc(k) P p(k + 2)
dicted values of dc-link voltage, active
power, and reactive power, respec- J
iL, abc(k) Minimization
Q p(k + 2)
tively; PSR and v dcR are, respectively, Predictive of
il, abc(k) Model J Cost Function
the rated values of active power and
v p(k + 2)
the dc-link voltage; m 1, m 2, and m 3 dc
ia (A)
fundamental frequency tracking, de- 0
pending on design considerations.
Finally, the third term (7) is used to –5
reduce the number of commutations
introducing weighting factor, m sw, to –10
keep the power losses below accept- 0 0.02 0.04 0.06 0.08 0.10
Time (s)
able limits. (a)
400
Selective Harmonic
Mitigation Technique
200
An evolution of SHE is the selective
harmonic mitigation (SHM) technique, va (V)
0
which is based on preprogrammed
waveforms not eliminating the low-
order harmonics but reducing the dis- –200
tortion below the limits imposed by a
grid code [30], [31]. –400
0 0.02 0.04 0.06 0.08 0.10
The control objective of the MPC- Time (s)
SHM technique is to follow the voltage (b)
reference, control the harmonic distor- 10 400
tion keeping it below the limits imposed vs
in the grid code, and reduce the switch- 5 200
is
ing losses as much as possible [32].
These three objectives are included in a
vs (V)
is (A)
0 0
cost function similar to that introduced
for the MPC-SHE method. In this case,
–5 –200
the second term (6) has to be modi-
fied, being the sum of the distortion of
–10 –400
those harmonics that have to be kept 0 0.02 0.04 0.06 0.08 0.10
below acceptable limits imposed by the Time (s)
grid code from the second- to the Kth (c)
-harmonic. In fact, the mathematical ex-
FIGURE 16 – The predictive current control of a DMC. (a) The output current. (b) The output volt-
pression of the MPC-SHM cost function age. (c) The input current and grid voltage (phase a) .
is similar to the MPC-SHE one but con-
sidering harmonics up to Kth order,
where K is higher than M. Another dif- higher in the opposite case. This dy- (up to harmonic ten) is always below
ference between both cost functions is namic weighting factors adjustment is the limit imposed. In addition, the
that a weighting factor, m i, is associated made at the fundamental frequency. average switching frequency of the
to each ith-order harmonic distortion The MPC-SHM method has been MPC-SHM method is lower than that
that has to be reduced. This fact pro- tested in a three-phase cascaded H- obtained with the MPC-SHE technique
vides the chance to relax the condition bridge (CHB) multilevel converter dealing with the same number of har-
of eliminating the distortion of those (three H-bridges per phase) connected monics. This phenomenon occurs be-
harmonics. The proposed MPC-SHM to an RL load as depicted in Figure 18. cause the SHM relaxes the conditions
technique can be summarized using the On the other hand, Figure 19 shows of the harmonic distortions compared
block diagram represented in Figure 17, the converter phase output voltage, with the SHE method, making it easier
where a weighting factors tuning block VaN, the line–line voltage, Vab, the load to find better solutions leading to low-
has been added to the previously in- voltage, Van, and the load currents, i a, er power losses.
troduced MPC-SHE method. For each i b, and i c . In addition, the respective
harmonic, the weighting factors tuning harmonic spectra are shown next to Control of Multilevel Inverters
block relaxes the weighting factor, m i, each waveform. A deep analysis of the The FCS-MPC method has been applied
if the ith harmonic distortion is below data shows that the distortion of har- to multilevel converters for multiple
the limit of the grid code, or makes it monics considered in the cost function applications. Among the multilevel
Ra Rb
g = i *a - i ap + i *b - i bp
Rc
p p
+ m dc v C1 - v C2 + m n n c . (8)
n
In the cost function, the first term
is focused on the current tracking,
S1 S2 which is the application of this FCS-
+ Vdc +
MPC method. The second term is pro-
H-Bridgekj portional to the absolute difference
between the voltage predictions of
– S1 S2
both capacitors, so a switching state
that generates smaller differences will
be preferred, leading to a voltage bal-
– ance situation. Finally, the third term
is proportional to the number of com-
FIGURE 18 – The power circuit of an H-bridge multilevel converter (three H-bridges per phase) mutations to get to the next switch-
connected to an RL load. ing state, n c, so a switching state that
0.8
200
0.6
0
0.4
−200 0.2
−400 0
1.79 1.795 1.8 1.805 1.81 1.815 1.82 1.825 1.83 1.835 1.84 0 500 1,000
Time (s) Frequency (Hz)
300 1
Load Voltage van(V)
200
0.8
100
0.6
0
0.4
−100
0.2
−200
−300 0
1.79 1.795 1.8 1.805 1.81 1.815 1.82 1.825 1.83 1.835 1.84 0 500 1,000
Time (s) Frequency (Hz)
15 1
Phase Currents iabc(A)
10
0.8
5
0.6
0
0.4
−5
0.2
−10
−15 0
1.79 1.795 1.8 1.805 1.81 1.815 1.82 1.825 1.83 1.835 1.84 0 500 1,000
Time (s) Frequency (Hz)
FIGURE 19 – The predictive harmonic mitigation phase output voltage, v aN, line voltage, v ab, load voltage, v an, and load currents, i a, i b, i c, for a
seven-level CHB.
p
implies fewer commutations of the g = i *a - i ap + i *b - i b . (9) This fact reduces to seven the num-
power semiconductors will be pre- ber of possible combinations leading
ferred. The weighting factors, m dc and In the CHB inverter case, there are to the same computational cost of a
m n, handle the relation between terms a high number of possible switching conventional three-phase, two-level
dedicated to current reference track- combinations (125 for this topology converter at the expense of solving
ing, voltage balance, and reduction of where the dc voltages of the H-bridges a suboptimal problem and losing dy-
switching frequency [33]. are fixed), so the computational cost namic response.
of the FCS-MPC method can become
CHB Inverter Topology excessively high. In [34], this is solved FC Inverter Topology
The FCS-MPC method can also be ap- by eliminating the redundant switch- A similar FCS-MPC strategy can be ap-
plied to achieve current tracking con- ing states with higher common-mode plied to the FC converter. In [35], a cost
trol in a three-phase CHB converter voltages. In addition, only the last function is presented to achieve cur-
with two H-bridges per phase [34]. In applied switching state and the six rent tracking and floating voltages con-
this case, the cost function only takes states surrounding it in the space-vec- trol of a three-phase FC converter with
into account the current tracking tor diagram are taken into account. two floating capacitors per phase. In
vdc
possible switching combinations to
300 vc2a 24, leading to a high reduction in the
computational cost at the expense of
200 limiting a control degree.
vc1a
To illustrate the good performance
100
of the FCS-MPC method for multilevel
0 converters, the results for the FC in-
Time (ms) verter are represented in Figure 20.
(a) As can be observed, the control objec-
tives, current tracking, and control of
400 the floating voltages are achieved.
Inverter Output Voltage (V)
300
MPC for Inverters with
200
Output LC Filter
Inverters with output LC filter are used
100 when necessary to obtain a sinusoidal
van output voltage with very low harmonic
0 content. This is the case of sensitive
Time (ms)
loads or drives for machines to reduce
(b)
the input voltage harmonics and in-
5
4 THDc =1.5% crease its lifespan or avoid problems
Output Currents (A)
vC, abc(V)
the number of switching states is J = 7. 100
0
Once the seven output voltage predic-
−100
tions are calculated, the cost function
−200
p 2 p 2
−300
g = ^v *C, a - v C, a h + ^v *C, b - v C, b h (12) 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Time (s)
20
is minimized to find the inverter out-
put vector that should be applied in 10
iO, abc(A)
vCa(V)
0 0 and the measurements of stator cur-
−50 −50
rent and mechanical angular speed
−100 −100
of the machine. In [40], this technique
−150 −150
was used together with a three-phase,
−200 −200
0 10 20 30 40 0 10 20 30 40 two-level inverter. In this way, the
Time (ms) Time (ms) number of switching states is J = 7.
(a) (b) Therefore, the predictions associated
to the seven different voltage vectors
FIGURE 25 – The system performance for the GPC strategy for different values of N and m. (a) are evaluated in an optimization stage
Phase a voltage and its reference for N = 5 and m = 0.0. (b) Phase a voltage and its reference to select the optimum switching state,
for N = 6 and m = 1.05.
S, to be applied to the inverter. For
the generation of the electrical torque
the GPC strategy when a linear load is to select the appropriate switching reference, an external control loop
connected to the inverter for different states of the converter. with a PI controller was used.
values of N and m [12]. It should be A third alternative, predictive The results of the PTC of an induc-
noted that correctly tuning the predic- torque control (PTC), has recently tion machine are shown in Figure 27.
tion horizon, N, and the weighting fac- been proposed [40]. In the PTC algo- The speed control is presented in
tor, m, can significantly increase the rithm, a cost function that gathers the Figure 27(a), where the reference step
performance of the system. tracking error of torque and stator changes from –150 to 150 rad/s and from
flux magnitude is employed 150 to –150 rad/s are applied at times
MPC for High-Performance Drives t = 0.5 and t = 3.0 [s], respectively.
* p
For the control of high-performance g = T *e - T ep + m } s - } s , (14) The dynamic performance of the torque
drives, there exist two well-estab- tracking can be observed in Figure 27(b),
lished methods: field-oriented control where T *e, } s *, T ep, and } s p are where it is clear that PTC provides a
and direct torque control (DTC) [39]. the reference and predicted values of very fast response. The stator current
The first one performs a decoupled electrical torque and the magnitude of for phase a is plotted in Figure 27(c),
control of torque and flux by control- the stator flux, respectively. The pa- showing a sinusoidal waveform, even
ling separately the quadrature and di- rameter m is a weighting factor that al- though no current controllers are direct-
rect components of the stator current lows for giving more or less relevance ly implemented.
in a rotating coordinate frame, which to each of the control objectives. Fol-
is synchronous with the rotor flux. lowing the operation principle of FCS- Past, Present, and Future
On the other hand, the DTC ap- MPC, the switching state applied to Challenges of MPC for Power
proach also performs an independent the converter at each sampling period Converters and Drives
control of torque and flux by employ- corresponds to the one that minimiz- MPC has been a good solution for in-
ing nonlinear hysteresis controllers es the cost function. The block dia- dustrial applications for decades [5],
and a logic based on a look-up table gram of PTC is presented in Figure 26. [41]. However, once the simplicity and
good performance of the MPC control-
ler in the power electronics field has
been demonstrated, the question to be
PI Inverter
ω∗ + Te∗ Induction Machine answered is, why is it not already exten-
– Minimization Sopt sively used in the industry?
ω ∗ of As a major challenge, the MPC
|ψ s| Cost Function needs an accurate model of the sys-
p
tem, and this is not usually a simple
p
Te |ψ s| task in highly dynamic systems. How-
J J
ψr ever, in recent years, the modeling of
ia
Predictive ψs complex electrical systems has been
Estimate
Model is ib greatly improved, and this challenge
can be solved. Although more re-
ω search is necessary, it is now possible
to find applications of MPC to power
FIGURE 26 – A block diagram of the PTC of an induction machine. converters where Luenberger and