IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO.

1, MARCH 2014 3

A New Power Conversion System for Megawatt

PMSG Wind Turbines Using Four-Level
Converters and a Simple Control Scheme
Based on Two-Step Model Predictive
Strategy—Part I: Modeling
and Theoretical Analysis
Venkata Yaramasu, Student Member, IEEE, Bin Wu, Fellow, IEEE,
Marco Rivera, Member, IEEE, and Jose Rodriguez, Fellow, IEEE

Abstract— A new power conversion system is explored in this (MV), megawatt-level, multilevel converter, permanent magnet
paper aiming wind turbines rated at the megawatt level. The synchronous machine, power conversion, predictive control,
proposed configuration consists of a medium-voltage, permanent renewable energy, state-space model, wind energy.
magnet synchronous generator connected to a low-cost three-
phase diode bridge rectifier, a dc–dc four-level boost converter as
I. I NTRODUCTION
the intermediate stage, and a four-level diode-clamped inverter
on the grid-side. The dc-link capacitor voltages are balanced
by the boost converter, and thus the control complexity for
the grid-tied inverter is greatly simplified. To control the boost
T HE AGE of multi-megawatt (MW) wind turbines has
arrived and currently 7.5 MW units are available in the
market [1], [2]. The future generation of turbines is anticipated
converter and grid-tied inverter, a simple method based on a
two-step model predictive strategy is presented. In the first to be in the range of 10–15 MW. At these power levels, the
part of this paper, the continuous- and discrete-time modeling use of medium-voltage (MV) technology is a better suited,
of the proposed power conversion system is analyzed. The efficient, economical, and promising approach [1]–[8]. Many
control objectives such as maximum power point tracking, dc-link MV power conversion systems are reported in [3], [6], and
capacitor voltages balancing, regulation of net dc-bus voltage, [9]–[13] for permanent magnet synchronous generator
reactive power generation, lower switching frequency operation,
and common-mode voltage minimization are considered in the (PMSG)-based wind turbines. The commercial solutions
design of controller. based on back-to-back (BTB) connected neutral-point-clamped
(NPC) converters [14], [15] are already available in the market
Index Terms— AC–dc power conversion, boost converter,
common-mode voltage (CMV), current control, dc–ac power by ABB (PCS6000) and Converteam (MV7000).
conversion, dc–dc power conversion, dc-link capacitor voltages The power flow in wind turbines is unidirectional, i.e.,
balancing, digital control, direct-driven, four-level boost, grid- from the generator to the grid. Moreover, the PMSG does
connected, maximum power point tracking, medium voltage not require a magnetizing current, unlike the squirrel cage
induction generator [3]. Having this freedom (which is an
Manuscript received June 3, 2013; revised October 25, 2013; accepted advantage), an uncontrolled diode rectifier can be used for
November 25, 2013. Date of publication December 12, 2013; date of current
version January 29, 2014. This work was supported in part by the Natural generator-side power conversion [16], [17]. This helps reduce
Sciences and Engineering Research Council of Canada through Wind Energy cost and complexity compared with active front-end rectifiers.
Strategic Network Project 3.1, in part by the Fondecyt Initiation into Research The use of passive front-end is also commercialized at the
under Grant 11121492, in part by CONICYT PCCI 12048 Project, in part by
the Universidad Técnica Federico Santa María, and in part by the NPRP under megawatt-level and low-voltage (LV) (<1 kV) by Clipper
Grant 4-077-2-028 through the Qatar National Research Fund. Recommended (Liberty C89), Vensys (V70/77), and Gold Wind (GW70/77).
for publication by Associate Editor Wenzhong Gao. Considering the state of the wind-energy market and its
V. Yaramasu and B. Wu are with the Department of Electrical and Computer
Engineering, Ryerson University, Toronto, ON M5B 2K3, Canada (e-mail: trend toward MV technology, the initiative of this paper is
[email protected]; [email protected]). to combine passive front-end converters with grid-side mul-
M. Rivera is with the Department of Industrial Technologies, Universidad tilevel inverters. To meet the future generation of 1–10 kV
de Talca, Curico 3460000, Chile (e-mail: [email protected]).
J. Rodriguez is with the Department of Electronics Engineering, Uni- class (as defined in [18]) PMSG wind turbines, a four-level
versidad Técnica Federico Santa María, Valparaíso 2390123, Chile (e-mail: diode-clamped inverter (FLDCI) is proposed at the grid-side.
[email protected]). Compared with NPC inverters, the FLDCI offers many benefits
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. such as [7], [8], [19]: smaller dv/dt’s, which reduce electro-
Digital Object Identifier 10.1109/JESTPE.2013.2294920 magnetic interference, a reduced device voltage rating (one-
2168-6777 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
4 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 1, MARCH 2014

third of dc-bus voltage), grid-friendly waveforms, increased achieve the aforementioned benefits is the requirement for
equivalent switching frequency (three-times the device switch- higher computational capacity. However, it is not a major
ing frequency), reduced grid filter requirement, and better grid- obstacle as many fast digital control platforms (DSPs and
code compliance [20]. In addition, the FLDCI can be directly FPGAs) are readily available in the present market. The
connected to a three-phase collection point without any step- FCS-MPC methods usually consider a one-step prediction
up transformer and this feature leads to lower operational horizon to reduce the computational burden. In [31], two-
and maintenance costs [11], [21]. The balancing of dc-link step prediction is proposed through simulations to improve
capacitor voltages with the FLDCI is much more difficult and the control performance, while minimizing the number of
complicated compared with the NPC converter [19], [22]. To calculations. In this paper, a simplified two-step prediction
overcome this problem, a four-level boost converter (FLBC) horizon is considered to achieve optimal performance.
is proposed, with which the balancing of the dc-link capacitor In the proposed power conversion system, the control
voltages can be achieved during all operating conditions. objectives for the FLBC include the maximum power point
Since the capacitor voltages are balanced by the FLBC, a tracking (MPPT) and balancing of dc-link capacitor voltages,
simple control system can be designed for the FLDCI. As and the FLDCI regulates net dc-bus voltage and grid reactive
shown in Fig. 1, the proposed power conversion system uses power. To improve the efficiency of power conversion at the
a three-phase diode rectifier at the generator-side, a dc–dc megawatt-level, the semiconductor device switching frequency
FLBC at the intermediate stage and a dc–ac FLDCI as a grid is minimized to below 1 kHZ. The grid-tied inverter switch-
interfacing unit. ing actions produce common-mode voltage (CMV). This
The works in [6] and [12] propose using a LV six-phase gen- CMV causes electromagnetic interference and leakage currents
erator, series connected two-level converters on the generator- [32], [33], and should thus be minimized. To achieve all
side, and NPC inverter on the grid-side. Estay et al. [13] these goals, in the first part of this paper, the complete
proposed a similar configuration, but with series connected system modeling is presented in continuous-time. For the
two-level boost converters on the generator-side. The main purpose of digital implementation, all the continuous-time
idea is to promote multiphase machines at the megawatt power models are converted to discrete-time domain. The complete
level to achieve lower torque ripple, better fault tolerance, system control is proposed using predictive control, where,
and a lower per-phase power/voltage rating compared with all control objectives are expressed as two independent cost
conventional three-phase machines. The proposed configura- functions. During each sampling interval, the cost functions
tion shown in Fig. 1 can be used with multiphase machines are minimized, and meaning that the control goals are met.
as well to achieve the aforementioned advantages. A nine- The switching states that minimize these cost functions have
phase generator with one-third power/voltage rating per phase, been chosen and applied to the FLBC and FLDCI directly.
followed by the series connection of diode-bridge rectifiers, This paper is organized as follows: in Section II, the config-
can be used to achieve the topology given in Fig. 1. uration and modeling of FLBC and FLDCI are introduced. The
In modern industrial controls, predictive control technology discrete-time conversion for two-step prediction is analyzed
is one of the major success stories and this methodology has in Section III. In Section IV, the proposed control scheme is
been used in hundreds of real-world applications. The finite presented. Finally, in Section V, conclusions are drawn.
control-set model predictive control (FCS-MPC) has become a
simple, intuitive, powerful, and promising alternative to control II. C ONFIGURATION AND M ODELING
power converters [7], [8], [23]–[28]. This method eliminates OF THE P ROPOSED S YSTEM
the need for linear PI regulators and modulators. The lower The configuration of the proposed variable-speed wind
switching frequency operation is a mandatory requirement at energy system is shown in Fig. 1. It consists of a direct-driven
the megawatt-level to minimize switching losses and to allow PMSG, a diode rectifier, a FLBC, and a FLDCI. The inverter
proper heat dissipation [7]. The pulse width/space vector mod- is connected to the three-phase grid through an L filter. It is
ulation (PWM/SVM) techniques produce significantly lower important to note that both the generator and grid are rated
order harmonics when operated below 1 kHZ and this causes for MV operation. The output voltage of the diode rectifier
poor power quality [8]. Much research has been carried out changes with respect to the varying wind speed conditions
and is still being carried out to solve this issue [29]. Since and it represents an unregulated dc-link. A regulated dc-link
a linear controller is applied to the power converter, which is used at the inverter terminals to ensure proper operation of
is nonlinear in nature, the performance is unsymmetrical and the system.
varies according to the operating conditions. The nonlinear The FLBC is composed of three active switches and
nature of the power converter becomes more predominant four diodes. The voltage rating of these devices is
at lower switching frequency operations, and under such one-third of those used in a standard boost converter,
conditions the FCS-MPC strategy provides better steady-state and thus the cost of this FLBC is approximately the
results and a faster transient response compared with classical same as a standard boost converter [34]. The FLDCI
controllers [8], [30]. uses 18 active switches and 18 clamping diodes. The
One of the best features of FCS-MPC strategy is that active switches can be realized using the MV-IGBT
system nonlinearities and limitations can be incorporated or IGCT. Three capacitors are used in the regulated dc-link
straight into the model. The constraint handling is naturally and the total dc-link voltage is divided among these capac-
accommodated by this method. The price that is paid to itors [v c1 = v c2 = v c3 = (v dc /3)]. The regulated dc-link
YARAMASU et al.: NEW POWER CONVERSION SYSTEM FOR MEGAWATT PMSG WIND TURBINES—PART I 5

Fig. 1. Configuration of proposed high-power PMSG wind energy system.

TABLE I a complementary manner [7], [8], [19].

S WITCHING S TATES AND I NVERTER AC-S IDE V OLTAGES ( j = a, b, c) To simplify the analysis, the following assumptions are
made:
⎫
Rag = Rbg = Rcg = Rg ⎪ ⎪
⎬
L ag = L bg = L cg = L g
. (1)
v ag + v bg + v cg = 0 ⎪ ⎪
⎭
i ag + i bg + i cg = 0

The inverter ac-side voltages (with respect to the negative

isolates the FLBC and FLDCI, and thus the operation of one dc-rail, N) can be expressed in terms of switching signals and
converter does not affect the other. As shall be shown in the dc-side voltages as follows:
following subsections, this feature facilitates easier modeling ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
and control. v aN S1a S2a S3a
⎣v bN ⎦ = v c1 ⎣ S1b ⎦ + v c2 ⎣ S2b ⎦ + v c3 ⎣ S3b ⎦ (2)
v cN S1c S2c S3c
A. Continuous-Time Modeling of Four-Level Inverter
The FLDCI as a grid-interfacing unit needs to control the where v c1 , v c2 , and v c3 are dc-link capacitor voltages. S1a ,
net dc-bus voltage and reactive power requested by the grid S2a , S3a , S1b , S2b , S3b , S1c , S2c , and S3c are the switching
operator [7], [8]. These objectives can be fulfilled by properly signals of FLDCI.
controlling the grid currents. For this reason, in this section, The synchronous frame (dq) grid currents can be expressed
the modeling of grid currents is analyzed in detail. in terms inverter voltages, grid voltages and filter inductance as
The model of FLDCI was previously analyzed in [35] and
considers a resistive load. In this paper, a simplified modeling d i dg 1

− ωg i qg = v di − v dg − Rg i dg
is presented because the inverter does not need to control dt Lg
the dc-link capacitor voltages. In addition, the modeling is d i qg 1

+ ωg i dg = v qi − v qg − Rg i qg (3)
presented in synchronous reference frame (SRF) to facilitate dt Lg
easier calculation of reference control variables. For a three-
phase FLDCI, 64 (levelphase = 43 ) switching combinations where L g and Rg represent grid filter inductance and its
and 37 voltage vectors (levelphase − [level − 1]phase = 43 − 33 ) internal resistance, respectively. ωg (= 2π f g ) is grid angular
are available [7]. The switching states and the corresponding frequency. ωg i qg and ωg i dg are induced voltages due to the
inverter ac-side voltages are shown in Table I, where, the transformation of filter inductance from abc to dq frame.
S j represents the level of inverter voltage vector. At any v dg and v qg are d- and q-axis grid voltages, respectively. i dg
switching instant, only three active switches conduct. The and i qg are d- and q-axis grid currents, respectively. v di and
switch pairs (S1 j , S 1 j ), (S2 j , S 2 j ), and (S3 j , S 3 j ) operate in v qi are d- and q-axis inverter voltages, respectively.
6 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 1, MARCH 2014

The abc to dq transformation is performed as follows: By applying Kirchhoff voltage and current laws to all the
⎡ ⎤ five modes given in Fig. 2, the dynamic equations for the
 xa
xd inductor current, i dc can be obtained as
= [K ] ⎣x b ⎦ (4)
xq d i dc v in − Rdc i dc
xc Mode-I : =
 dt L dc
2 cos(θg ) cos(θg − 2π/3) cos(θg − 4π/3)
[K ] = d i dc v in − Rdc i dc − v c2
3 −sin(θg ) −sin(θg − 2π/3) −sin(θg − 4π/3) Mode-II : =
dt L dc
(5)
d i dc v in − Rdc i dc − v c2 − v c3
Mode-III : =
where x d and x q are variables in a synchronous reference dt L dc
frame, and x a , x b , and x c are natural frame variables. θg is d i dc v in − Rdc i dc − v c1 − v c2
Mode-IV : =
the grid-voltage angle and it can be obtained by a SRF phase- dt L dc
locked loop (PLL) [5]. d i dc v in − Rdc i dc − v c1 − v c2 − v c3
The SRF grid currents in (3) can be expressed in state-space Mode-V : = . (8)
dt L dc
form as follows [7], [8]: Similarly, the dc-link capacitor voltage dynamics can be
  −Rg   1  obtained as
d i dg ωg i dg Lg 0 v di
= Lg
−R + d v c1 −i o1 d v c2 −i o2 d v c3 −i o3
dt i qg −ωg L gg i qg 0 L1g v qi I: = , = , =
   dt C1 dt C2 dt C3
  
Qi d v c1 −i o1 d v c2 i dc − i o2 d v c3 −i o3
P
 −1 i  II : = , = , =
dt C1 dt C2 dt C3
Lg 0 v dg d v c1 −i o1 d v c2 i dc − i o2 d v c3 i dc − i o3
+ −1 (6)
0 Lg
v qg III : = , = , =
   dt C1 dt C2 dt C3
Ri d v c1 i dc − i o1 d v c2 i dc − i o2 d v c3 −i o3
IV : = = , =
dt C1 dt C2 dt C3
where v di and v qi are estimated using (2), (4) and (5). From
d v c1 i dc − i o1 d v c2 i dc − i o2 d v c3 i dc − i o3
(6), it can be noticed that the grid current dynamics are related V: = = , = .
to the parameters of the system, measured grid voltages and dt C1 dt C2 dt C3
currents, and estimated inverter voltages. (9)
The CMV can be represented as the voltage between By combining the dynamic equations in (8) and (9), the
the neutral point of the grid (n) and negative dc-rail (N) continuous-time system describing the inductor current and
[33], [36] dc-link capacitor voltages can be obtained as
⎡ R (1−S ) (1−S ) (1−S ) ⎤
v aN + v bN + v cN ⎡ ⎤ − L dc
dc
− L dc1 f − L dc2 f − L dc3 f
v nN = v cm = . (7) i dc ⎢ ⎥
3 d ⎢ ⎥ ⎢ (1−S1 f ) ⎥
⎢v c1 ⎥ = ⎢ C1 0 0 0 ⎥
The systems in (2) and (7) shows that CMV is related dt ⎣ v ⎦ ⎢ (1−S2 f ) ⎥
to the switching signals. By properly selecting a switching
c2 ⎣ C2 0 0 0 ⎦
v c3 (1−S3 f )
state among 64, the SRF grid currents can be controlled while C3 0 0 0
  
minimizing the CMV. Pf
⎡ ⎤ ⎡ 1
⎤⎡ ⎤
i dc 0 L dc0 0 v in
B. Continuous-Time Modeling of FLBC ⎢v c1 ⎥ ⎢ ⎢ 0 − 1
0 0 ⎥⎢ ⎥
⎥ i
×⎢ ⎥
⎣v c2 ⎦ + ⎢
C 1 ⎥ ⎢ ⎥ (10)
o1
The MPPT can be effectively achieved by controlling the ⎣ 0 0 − C12 0 ⎦ ⎣i o2 ⎦
inductor current of a FLBC. The dc-link capacitor voltages v c3 0 0 0 − C13 i o2
can also be balanced by the FLBC with the proper selection   
of switching states. To facilitate the controller design, the Qf
modeling of inductor current, and dc-link capacitor voltages where S1 f , S2 f , and S3 f are the switching signals applied to
is performed in this section. the FLBC active switches S1 , S2 , and S3 , respectively.
The model of standard boost converter is presented in [37] In deriving the above dynamic system, the inverter is
for a single-phase power factor correction application. Similar replaced by three virtual resistive loads as shown in Fig. 3.
modeling is performed for the FLBC, treating an unregulated The measurement of output currents, i o1, i o2, and i o3 is not
dc-link as the input voltage source and a four-level inverter recommended as it increases the complexity as well as cost.
as three virtual resistive loads. As shown in Fig. 2, the Rather, these variables can be estimated using the measured
boost converter operates in five modes [34]. In comparison grid currents and FLDCI switching signals. By referring to
with the standard boost converter, which operates in mode I Fig. 3, the relationship between the output currents and inverter
and V only, the additional modes help in the charging and currents can be shown as
discharging of each capacitor. The FLBC switching states
i o1 = i p
and the corresponding status of active switches, diodes, and
capacitors are summarized in Table II. The Rdc represents the i o2 = i o1 + i x = i p + i x
internal resistance of the inductor L dc . i o3 = i o2 + i y = i p + i x + i y . (11)
YARAMASU et al.: NEW POWER CONVERSION SYSTEM FOR MEGAWATT PMSG WIND TURBINES—PART I 7

Fig. 2. Operating modes for the FLBC (a) mode-I, (b) mode-II, (c) mode-III, (d) mode-IV, and (e) mode-V.

TABLE II switching signals. This is shown as follows:

S UMMARY OF FLBC S WITCHING S TATES ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ip S1a S2a S3a S1b S2b S3b
⎣i x ⎦ = ⎣ S 1a S2a S3a ⎦ i ag + ⎣ S 1b S2b S3b ⎦ i bg
iy S 1a S 2a S3a S 1b S 2b S3b
⎡ ⎤
S1c S2c S3c
+ ⎣ S 1c S2c S3c ⎦ i cg . (13)
S 1c S 2c S3c
As shown in this section, the proposed approach does not
require the dynamic model and parameters of the PMSG.
In addition, the PMSG rotor position is also not required.
This greatly simplifies the complexity of the control system.
From (10), it can be understood that the inductor current and
dc-link capacitor voltages can be effectively controlled by the
optimal selection of switching signals.

III. D ISCRETE -T IME M ODELING AND

T WO -S TEP P REDICTION
The FCS-MPC is an advanced, truly model based, and
optimization control strategy [40]. For the digital imple-
mentation of FCS-MPC algorithm, a discrete-time model is
needed. Microprocessor-based hardware helps in the real-time
implementation of such models. To accomplish this, in this
section, all the continuous-time systems are converted to the
discrete-time domain.
The discrete-time dynamic system for SRF grid currents can
Fig. 3. Representation of virtual resistors and FLDCI.
be obtained from (6) as follows:
  
i dg (k + h) i dg (k + h − 1) v di (k + h − 1)
= +ϒ
The inverter currents i p , i x , and i y can be estimated based i qg (k + h) i qg (k + h − 1) v qi (k + h − 1)

on inverter voltage vector levels Sa , Sb , and Sc , and measured v (k + h − 1)
+ dg (14)
grid currents as [7], [8], [35], [38], [39] v qg (k + h − 1)
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ where
ip Sa == 3 Sb == 3 Sc == 3
⎣i x ⎦= ⎣ Sa == 2⎦ i ag + ⎣ Sb == 2⎦ i bg + ⎣ Sc == 2⎦ i cg .
 = ePi Ts , ϒ = Pi −1 ( − I2x2 )Qi
iy Sa == 1 Sb == 1 Sc == 1
 = Pi −1 ( − I2x2 )Ri . (15)
(12)
In the above systems, h is prediction horizon and Ts is
By substituting the Sa , Sb , and Sc values given in Table I, controller sampling time. The continuous-time system for
the above expression can be described in terms of FLDCI the boost converter, which is described in (10), can also be
8 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 1, MARCH 2014

have not yet been analyzed through experimental tests. In

this paper, a simplified two-step prediction is extended for
the control of proposed power converters and this concept is
shown in Fig. 5(a) and (b). Similar to one-step prediction,
this approach uses 5 and 64 switching states for FLBC and
FLDCI, respectively. However, the predictions are projected
to (k + 2) sampling instant, and with this approach the
advantages of lower computations of one-step prediction and
optimal performance of two-step prediction can be combined.
Fig. 4. Prediction of control variables using one-step prediction for (a) FLBC
The discrete-time model for the two-step prediction of grid
and (b) four-level inverter. currents is as follows:
  
i dg (k + 2) i (k + 1) v (k)
=  dg + ϒ di
i qg (k + 2) i qg (k + 1) v qi (k)

v (k + 1)
+ dg (18)
v qg (k + 1)
where v dg (k + 1) and v qg (k + 1) are extrapolated [25], [26],
[30] grid voltages. The i dg (k + 1) and i dg (k + 1) are predicted
grid currents in (k + 1) sampling instant, and they can be
obtained as
   
i dg(k + 1) i (k) v (k) v (k)
Fig. 5. Prediction of control variables using simplified two-step prediction =  dg + ϒ di +  dg (19)
for (a) FLBC and (b) four-level inverter. i qg(k + 1) i qg (k) v qi (k) v qg (k)
where v dg (k) and v qg (k) are measured grid voltages. i dg (k) and
converted to discrete-time domain as i qg (k) are measured grid currents. By observing (18) and (19),
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
i dc (k + h) i dc (k + h − 1) v in (k + h − 1) it can be observed that the same inverter voltages [v di (k),
⎢v c1 (k + h)⎥ ⎢v c1 (k + h − 1)⎥ ⎢i o1(k + h − 1)⎥ v qi (k)] are used in (k + 1) and (k + 2) prediction of grid
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣v c2 (k + h)⎦ =  ⎣v c2 (k + h − 1)⎦+ ⎣i o2 (k + h − 1)⎦ currents. Compared with one-step prediction, the number of
v c3 (k + h) v c3 (k + h − 1) i o3(k + h − 1) calculations are higher, but significantly lower compared with
(16) the standard two-step prediction.
Similarly, the two-step prediction model for the FLB con-
where verter can be obtained as shown in (20), at the top of the next
page. The gating signals S 1 f , S 2 f , and S 3 f are complementary
 = ePf Ts , = Pf −1 ( − I4x4 )Qf . (17)
of S1 f , S2 f , and S3 f , respectively. v in (k + 1), i o1(k + 1),
The parameter matrix, Pf contains gating signals S1 f , S2 f , i o2 (k + 1), and i o3 (k + 1) are extrapolated quantities. φi j with
and S3 f . These signals are set to zero during the offline i = 1, 2, 3, 4 and j = 1, 2, 3, 4 are the elements of the matrix
discrete-time conversion.  defined in (17). i dc (k + 1), v c1 (k + 1), v c2 (k + 1), and
In Figs. 4 and 5, Sf and Si represents the vector of switch- v c3 (k + 1) are predicted quantities during (k + 1) sampling
ing signals applied to the FLBC and FLDCI, respectively. instant and they can be obtained, as shown in (21), at the top
In Fig. 4(a) and (b), prediction of control variables at (k + 1) of the next page. v in (k), i dc (k), v c1 (k), v c2 (k), and v c3 (k) are
uses 5 and 64 switching states for the FLBC and FLDCI, measured variables and i o1 (k), i o2 (k), and i o3 (k) are calculated
respectively. This is essentially called one-step prediction in terms of grid currents and FLDCI gating signals [see (13)].
(h = 1) and this methodology is often used [23], [35], [37] Similar to the two-step prediction of FLDCI, the (k + 1) and
to simplify analysis and digital implementation computations. (k + 2) prediction are carried out using same FLBC switching
[41]–[43] suggest using two-step prediction (h = 2) for signals [S1 f (k), S2 f (k), and S3 f (k)].
optimal performance. But this leads to higher computational
burden for FLBC and FLDCI caused by 25 (52 ) and 4096 (642 ) IV. P ROPOSED M ODEL P REDICTIVE C ONTROL S TRATEGY
predictions, respectively. To reduce the number of calculations The proposed control scheme using two-step prediction
while achieving better performance compared with one-step strategy is shown in Fig. 6. As shown in Fig. 6(a), independent
prediction, a simplified two-step prediction is analyzed in [31]. control loops are developed for FLBC and FLDCI and the
The control delay caused by the digital signal processor control objectives are set for both of them. This is further
is a very important factor that needs to be considered in clarified in the following subsections.
the real-time implementation of predictive control algorithms.
However, the analysis in [31] has been presented through
MATLAB simulations by ignoring this delay. The complete A. Control System for FLBC
analysis in terms of complexity of controller, increase in the The control system for FLBC is depicted in Fig. 6(b).
number of calculations, and additional computational capacity One of the important requirements in variable speed wind
required and improvement in the performance of converter energy systems is harvesting the maximum possible energy
YARAMASU et al.: NEW POWER CONVERSION SYSTEM FOR MEGAWATT PMSG WIND TURBINES—PART I 9

⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
i dc (k + 2) φ11 φ12 S 1 f (k) φ13 S 2 f (k) φ14 S 3 f (k) i dc (k + 1) v in (k + 1)
⎢v c1 (k + 2)⎥ ⎢φ21 S 1 f (k) φ22 φ23 φ24 ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥=⎢ ⎥ ⎢v c1 (k + 1)⎥ + ⎢i o1 (k + 1)⎥ (20)
⎣v c2 (k + 2)⎦ ⎣φ31 S 2 f (k) φ32 φ33 φ34 ⎦ ⎣v c2 (k + 1)⎦ ⎣i o2 (k + 1)⎦
v c3 (k + 2) φ41 S 3 f (k) φ42 φ43 φ44 v c3 (k + 1) i o3 (k + 1)
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
i dc (k + 1) φ11 φ12 S 1 f (k) φ13 S 2 f (k) φ14 S 3 f (k) i dc (k) v in (k)
⎢v c1 (k + 1)⎥ ⎢φ21 S 1 f (k) φ22 φ23 φ24 ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢v c1 (k)⎥ + ⎢i o1 (k)⎥
⎣v c2 (k + 1)⎦ = ⎣φ31 S 2 f (k) φ32 φ33 φ34 ⎦ ⎣ v c2 (k) ⎦ ⎣i o2 (k) ⎦ (21)
v c3 (k + 1) φ41 S 3 f (k) φ42 φ43 φ44 v c3 (k) i o3 (k)

from the wind. For this purpose, MPPT algorithms are used. where λswc, f is the corresponding weighting factor. Sxf,op (k)
In this paper, a simple optimal tip-speed-ratio MPPT algorithm is optimal FLBC switching signal in the previous iteration.
[4], [44] is used. Other advanced and sophisticated MPPT This switching frequency minimization algorithm anticipates
algorithms can be readily applied to this power conversion and minimizes the number of switch changes compared with
system. The MPPT algorithm provides reference speed to the the previous sampling instant. When λswc, f > 0, the switching
generator based on the varying wind speed conditions. A PI frequency reduction can be obtained. Therefore, the optimiza-
controller generates the input active power reference, Pin∗ , as tion of weighting factor λswc, f can start from 0 and increase
follows: to a higher value.
 
∗  All these control objectives for the FLBC are merged into
Pin∗ (k) = K p + K i /S ωm (k) − ωm (k) (22) a cost function as
where K p and K i are the PI controller proportional and
g f (k) = gtrack, f (k) + gbal, f (k) + gswc, f (k). (27)
integral gains. The generation of Pin∗ is the same for classical
and proposed control methods, but the regulation of Pin is The control algorithm for the digital implementation two-
different with these methods. By tracking to Pin∗ , the MPPT step prediction for the FLBC is shown in Fig. 7(a). The
can be achieved, with which the generator speed, current, and algorithm is initialized by setting the switching state number
electromagnetic torque can be regulated. This control objective i to 0 and optimal g value to ∞, and then the algorithm
is expressed as the following cost function: enters the loop. The measured quantities and 5 switching states

2 are used in the prediction of (k + 1) instant variables. The
gtrack, f (k) = Pin∗ (k + 2) − Pin (k + 2) (23)
predicted variables at (k + 1) instant and the extrapolated
where Pin∗ (k + 2) is the extrapolated reference variable. quantities are used in the prediction of (k + 2) variables. It
The two-step prediction for the input active power is is important to note that the number of predictions at (k + 1)
obtained using extrapolated input voltage, v in (k + 2), and and (k +2) instants are 5 only. The cost function minimization
predicted dc-link current, i dc (k + 2) [according to (20)]. This is based on the (k + 2) variables only. The switching state,
is depicted as follows: which produces minimal value of gop , is chosen and applied
to the FLBC gating terminals directly. The current regulators
Pin (k + 2) = v in (k + 2) · i dc (k + 2). (24) and modulation are eliminated with the proposed approach.
Another important control objective is to balance the dc-link
capacitor voltages. This is expressed as follows: B. Control System for Four-Level Inverter
gbal, f (k) = λbal, f ∗ [v c1 (k + 2) − v c2 (k + 2)]2 The FLDCI control system is shown in the Fig. 6(c). The
control objectives for the FLDCI include regulation of grid
+λbal, f ∗ [v c2 (k + 2) − v c3 (k + 2)]2
active and reactive powers. The grid active and reactive powers
+λbal, f ∗ [v c1 (k + 2) − v c3 (k + 2)]2 (25) can be expressed as [4], [8]
where λbal, f is the weighting factor used for the capacitor 3 
voltages balancing task. There is no empirical formula for Pg = v dg i dg + v qg i qg
2
the selection of a weighting factor, but the guidelines given 3 
in [45] can be used. Since the dc-link capacitor voltages Qg = v qg i dg − v dg i qg . (28)
2
balancing must be accomplished along with the Pin tracking,
For grid-voltage orientation, the d-axis of the SRF is aligned
one can start initial optimization with a high value of λbal, f
with the grid voltage vector and thus v qg = 0. With this,
and decrease this value in small steps until the drift in capacitor
the active and reactive power expressions in (28) can be
voltages reaches 2% of its nominal dc-link voltage.
simplified as
The device switching frequency reduction is penalized in
⎫
the cost function, gswc, f 3
Pg = + v dg i dg ⎪⎬
 2
gswc, f (k) = λswc, f ∗ | Sxf (k) − Sxf,op (k) | (26) for v qg = 0. (29)
3 ⎪
⎭
x=1,2,3
Q g = − v dg i qg
2
10 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 1, MARCH 2014

Fig. 6. Proposed model predictive control scheme for four-level converters based PMSG WECS. (a) Block diagram of complete control system with control
objectives and measurements. (b) Control system for four-level boost converter. (c) Control system for four-level diode-clamped inverter.

From (29), it can be observed that the active and reactive to (29). This is demonstrated as follows:
powers fed to the grid can be regulated by controlling the d and ∗
 
∗ 
i dg (k) = K p + K i /S v dc (k) − v dc (k)
q-axis currents, respectively. Through the regulation of net dc-
bus voltage by a PI controller, the reference d-axis current is ∗
Q ∗g (k)
i qg (k) = (30)
generated. The reference q-axis current is obtained according −1.5 v dg (k)
YARAMASU et al.: NEW POWER CONVERSION SYSTEM FOR MEGAWATT PMSG WIND TURBINES—PART I 11

Fig. 7. Proposed two-step predictive control algorithm for (a) FLBC and (b) FLDCI.

∗ is net dc-bus voltage reference, and for grid-tied

where v dc obtained according to (18). Since the reference tracking of grid
applications this value is constant and usually set to 3.062 currents is the primary objective, the corresponding weighting
times grid phase voltage [4]. The Q ∗g is the reference reactive factor is one, as shown in (31).
power that needs to be provided by a wind turbine to support The switching frequency reduction objective is expressed as
the grid voltage. The Q ∗g is zero for unity power factor,  
gswc,i (k) = λswc,i ∗ | Sxj (k) − Sxj,op(k) | (32)
negative for leading power factor and positive for lagging
∗ and i ∗ x=1,2,3 j =a,b,c
power factor. It should be noted that the generation i dg qg
is similar to the classical voltage oriented control [4]. where λswc,i is a weighting factor. Sxj,op (k) is the optimal
The above two control objectives can be met by forcing the FLDCI switching signal in the previous sampling instant. The
grid currents to track to their references. This is defined as explanation given earlier for the selection of λswc, f value also
 2 applies to the λswc,i .
∗
gtrack,i (k) = i dg (k + 2) − i dg (k + 2) The CMV can be minimized by penalizing the following
 2 cost function:
∗
+ i qg (k + 2) − i qg (k + 2) (31) gcmv,i (k) = λcmv,i ∗ | v cm (k) | (33)
where ∗ (k
+ 2) and
i dg ∗ (k + 2)
i qg are the extrapolated reference where λcmv,i is weighting factor for CMV minimization. The
currents. The two-step prediction for the grid currents are CMV minimization is the secondary control objective, thus,
12 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 1, MARCH 2014

the selection of λcmv,i value can start from 0 and progress to [6] M. Duran, S. Kouro, B. Wu, E. Levi, F. Barrero, and S. Alepuz,
higher values. “Six-phase PMSG wind energy conversion system based on medium-
voltage multilevel converter,” in Proc. IEEE 14th Eur. Conf. Power
The final cost function for the FLDCI is defined as Electron. Appl., Sep. 2011, pp. 1–10.
[7] V. Yaramasu, B. Wu, and J. Chen, “Model predictive control of
gi (k) = gtrack,i (k) + gswc,i (k) + gcmv,i (k). (34) grid-tied four-level diode-clamped inverters for high power wind
energy conversion systems,” IEEE Trans. Power Electron., 2013,
The control algorithm for the FLDCI is shown in Fig. 7(b) doi: 10.1109/TPEL.2013.2276120.
and it is similar to the FLBC algorithm in terms of imple- [8] V. Yaramasu and B. Wu, “A model predictive decoupled active
and reactive power control for high power grid-connected four-
mentation. During each iteration, the future behavior of the level diode-clamped inverters,” IEEE Trans. Ind. Electron., 2013,
control variables are predicted using the 64 switching states. doi: 10.1109/TIE.2013.2278959.
The switching state (among 64), which minimizes the cost [9] F. Blaabjerg, M. Liserre, and K. Ma, “Power electronics converters
for wind turbine systems,” IEEE Trans. Ind. Appl., vol. 48, no. 2,
function gi (k), is chosen and applied to the FLDCI. The linear pp. 708–719, Mar. 2012.
regulators (PI) are eliminated in the internal current control [10] J. Li, S. Bhattacharya, and A. Huang, “A new nine-level active NPC
loops and there is no need to design the pulse width or space (ANPC) converter for grid connection of large wind turbines for
distributed generation,” IEEE Trans. Power Electron., vol. 26, no. 3,
vector modulation. pp. 961–972, Mar. 2011.
[11] S. Gjerde, P. Olsen, K. Ljokelsoy, and T. Undeland, “Control and fault
V. C ONCLUSION handling in a modular series connected converter for a transformerless
100 kV low weight offshore wind turbine,” IEEE Trans. Ind. Appl.,
The major contributions of this paper are summarized as 2013, doi: 10.1109/TIA.2013.2272032.
follows. [12] H. S. Che, E. Levi, M. Jones, M. Duran, W.-P. Hew, and N. Abd Rahim,
1) A simple, low cost and promising MV power conversion “Operation of a six-phase induction machine using series-connected
machine-side converters,” IEEE Trans. Ind. Electron., vol. 61, no. 1,
system consisting of diode rectifier, FLBC, and FLDCI pp. 164–176, Jan. 2014.
is proposed for megawatt-level PMSG wind turbines. [13] G. Estay, L. Vattuone, S. Kouro, M. Duran, and B. Wu, “Dual-boost-
The proposed topology combines the advantages of low NPC converter for a dual three-phase PMSG wind energy conversion
system,” in Proc. IEEE Int. Conf. PEDES, Dec. 2012, pp. 1–6.
cost passive front-end and efficient multilevel operation.
[14] A. Faulstich, J. Stinke, and F. Wittwer, “Medium voltage converter for
2) The grid-tied inverter need not to balance the dc-link permanent magnet wind power generators up to 5 MW,” in Proc. IEEE
capacitor voltages, and this facilitates a simplified con- Eur. Conf. Power Electron. Appl., vol. 1. Sep. 2005, pp. 1–9.
trol system for the inverter. [15] S. Alepuz, A. Calle, S. Busquets-Monge, S. Kouro, and B. Wu, “Use of
stored energy in PMSG rotor inertia for low-voltage ride-through in
3) The continuous- and discrete-time modeling of the back-to-back NPC converter-based wind power systems,” IEEE Trans.
proposed power conversion is presented. The control Ind. Electron., vol. 60, no. 5, pp. 1787–1796, May 2013.
objectives such as MPPT, dc-link capacitor voltages bal- [16] A. Di Gerlando, G. Foglia, M. Iacchetti, and R. Perini, “Analysis
ancing, regulation of net dc-bus voltage, reactive power and test of diode rectifier solutions in grid-connected wind energy
conversion systems employing modular permanent-magnet synchronous
generation, lower switching frequency operation, and generators,” IEEE Trans. Ind. Electron., vol. 59, no. 5, pp. 2135–2146,
CMV minimization are modeled in terms of switching May 2012.
signals of the proposed power electronic converters. [17] J. Wang, D. Xu, B. Wu, and Z. Luo, “A low-cost rectifier topology for
variable-speed high-power PMSG wind turbines,” IEEE Trans. Power
The proposed approach omits the need for wind turbine Electron., vol. 26, no. 8, pp. 2192–2200, Aug. 2011.
and generator models. The proposed methodology can [18] K. Ma, F. Blaabjerg, and M. Liserre, “Thermal analysis of multi-
be easily extended to passive load and motor drive level grid-side converters for 10-MW wind turbines under low-voltage
applications. ride through,” IEEE Trans. Ind. Appl., vol. 49, no. 2, pp. 909–921,
Mar./Apr. 2013.
4) A simplified two-step model predictive strategy is pro- [19] B. Wu, High-Power Converters and AC Drives, 1st ed. New York, NY,
posed to achieve optimal performance compared with USA: Wiley, 2006.
the one-step prediction. The complete system model and [20] E. ON Netz Gmbh, Bayreuth, Germany. (2006, Apr.). Grid
control algorithms are presented and compared with the Code—High and Extra High Voltage [Online]. Available:
http://www.thewindpower.net
standard one-step prediction. [21] W. Erdman and M. Behnke, Low Wind Speed Turbine Project Phase II:
The Application of Medium-Voltage Electrical Apparatus to the Class
of Variable Speed Multi-Megawatt Low Wind Speed Turbines. Golden,
R EFERENCES CO, USA: NREL, 2012.
[22] J. Pou, R. Pindado, and D. Boroyevich, “Voltage-balance limits in four-
[1] M. Liserre, R. Cardenas, M. Molinas, and J. Rodriguez, “Overview of level diode-clamped converters with passive front ends,” IEEE Trans.
multi-MW wind turbines and wind parks,” IEEE Trans. Ind. Electron., Ind. Electron., vol. 52, no. 1, pp. 190–196, Feb. 2005.
vol. 58, no. 4, pp. 1081–1095, Apr. 2011. [23] J. Rodriguez, M. P. Kazmierkowski, J. R. Espinoza, P. Zanchetta,
[2] F. Blaabjerg and K. Ma, “Future on power electronics for wind turbine H. Abu-Rub, H. A. Young, et al., “State of the art of finite control
systems,” IEEE J. Emerging Sel. Topics Power Electron., vol. 1, no. 3, set model predictive control in power electronics,” IEEE Trans. Ind.
pp. 139–152, Sep. 2013. Informat., vol. 9, no. 2, pp. 1003–1016, May 2013.
[3] V. Yaramasu and B. Wu, “Three-level boost converter based medium [24] M. Rivera, V. Yaramasu, A. Llor, J. Rodriguez, B. Wu, and
voltage megawatt PMSG wind energy conversion systems,” in Proc. M. Fadel, “Digital predictive current control of a three-phase four-leg
IEEE ECCE, Phoenix, AZ, USA, Sep. 2011, pp. 561–567. inverter,” IEEE Trans. Ind. Electron., vol. 60, no. 11, pp. 4903–4912,
[4] B. Wu, Y. Lang, N. Zargari, and S. Kouro, Power Conversion and Con- Nov. 2013.
trol of Wind Energy Systems (IEEE Press Series on Power Engineering), [25] V. Yaramasu, M. Rivera, B. Wu, and J. Rodriguez, “Model predictive
1st ed. Hoboken, NJ, USA: Wiley, Jul. 2011. current control of two-level four-leg inverters—Part I: Concept, algo-
[5] R. Teodorescu, M. Liserre, and P. Rodriguez, Grid Converters for Pho- rithm and simulation analysis,” IEEE Trans. Power Electron., vol. 28,
tovoltaic and Wind Power Systems. Chichester, U.K.: Wiley, Jan. 2011. no. 7, pp. 3459–3468, Jul. 2013.
YARAMASU et al.: NEW POWER CONVERSION SYSTEM FOR MEGAWATT PMSG WIND TURBINES—PART I 13

[26] M. Rivera, V. Yaramasu, J. Rodriguez, and B. Wu, “Model predictive Venkata Yaramasu (S’08) received the B.Tech.
current control of two-level four-leg inverters—Part II: Experimental degree in electrical and electronics engineering from
implementation and validation,” IEEE Trans. Power Electron., vol. 28, Jawaharlal Nehru Technological University, Hyder-
no. 7, pp. 3469–3478, Jul. 2013. abad, India, in 2005, and the M.E. degree in elec-
[27] V. Yaramasu, B. Wu, M. Rivera, and J. Rodriguez, “Predictive cur- trical engineering from the S. G. S. Institute of
rent control and DC-link capacitor voltages balancing for four-leg Technology and Science, Indore, India, in 2008. He
NPC inverters,” in Proc. IEEE ISIE, Taipei, Taiwan, May 2013, is currently pursuing the Ph.D. degree in electrical
pp. 1–6. engineering with Ryerson University, Toronto, ON,
[28] M. Rivera, J. Rodriguez, V. Yaramasu, and B. Wu, “Predictive load Canada.
voltage and capacitor balancing control for a four-leg NPC inverter,” in His current research interests include renewable
Proc. IEEE 15th Int. Power Electron. Motion Control Conf., Novi Sad, energy system, high power converters, and predictive
Serbia, Sep. 2012, pp. 1–5. control.
Mr. Yaramasu received six Best Student Paper Awards and two first prizes
[29] W. Song, X. Feng, and K. Smedley, “A carrier-based PWM strategy
in national level technical quiz competitions. He received the Best Poster
with the offset voltage injection for single-phase three-level neutral-
Award at the NSERC–WESNet Annual Meeting in 2010, the Best Teaching
point-clamped converters,” IEEE Trans. Power Electron., vol. 28, no. 3,
Assistant Award from the Faculty of Engineering and Architectural Science
pp. 1083–1095, Mar. 2012.
in 2010, the Student Research Awards from the Toronto Hydro and Connect
[30] J. Rodriguez and P. Cortes, Predictive Control of Power Converters and Canada in 2010, 2011, and 2012, and the Research Excellence Awards from
Electrical Drives, 1st ed. Chichester, U.K.: Wiley, Mar. 2012. the Electrical and Computer Engineering Department in 2012 and 2013.
[31] V. Yaramasu, B. Wu, M. Rivera, and J. Rodriguez, “Enhanced model
predictive voltage control of four-leg inverters with switching frequency
reduction for standalone power systems,” in Proc. IEEE 15th Int.
Power Electron. Motion Control Conf., Novi Sad, Serbia, Sep. 2012, Bin Wu (S’89–M’92–SM’99–F’08) received the
pp. 1–5. Ph.D. degree in electrical and computer engineer-
[32] C.-C. Hou, C.-C. Shih, P.-T. Cheng, and A. M. Hava, “Common- ing from the University of Toronto, Toronto, ON,
mode voltage reduction pulsewidth modulation techniques for three- Canada, in 1993.
phase grid-connected converters,” IEEE Trans. Power Electron., vol. 28, He joined Ryerson University, Toronto, ON,
no. 4, pp. 1971–1979, Apr. 2013. Canada, after being with Rockwell Automation,
[33] P. Pairodamonchai, S. Suwankawin, and S. Sangwongwanich, “Design Cambridge, ON, Canada as a Senior Engineer, where
and implementation of a hybrid output EMI filter for high-frequency he is currently a Professor and NSERC/Rockwell
common-mode voltage compensation in PWM inverters,” IEEE Trans. Industrial Research Chair of power electronics and
Ind. Appl., vol. 45, no. 5, pp. 1647–1659, Oct. 2009. electric drives. He has published more than 280
[34] K. Corzine and S. Majeethia, “Analysis of a novel four-level DC/DC technical papers, authored or co-authored two Wiley-
boost converter,” IEEE Trans. Ind. Appl., vol. 36, no. 5, pp. 1342–1350, IEEE Press books, and holds more than 20 issued/pending patents in the
Sep./Oct. 2000. area of power conversion, advanced controls, adjustable-speed drives, and
[35] P. Cortes, J. Rodriguez, S. Alepuz, S. Busquets-Monge, and renewable energy systems.
J. Bordonau, “Finite-states model predictive control of a four-level Dr. Wu is a fellow of the Engineering Institute of Canada and the Canadian
diode-clamped inverter,” in Proc. IEEE PESC, Rhodes, Greece, Academy of Engineering. He received the Gold Medal of the Governor
Jun. 2008, pp. 2203–2208. General of Canada, the Premiers Research Excellence Award, the Ryerson
[36] R. Vargas, U. Ammann, J. Rodriguez, and J. Pontt, “Predictive strategy Distinguished Scholar Award, the Ryerson Research Chair Award, and the
to control common-mode voltage in loads fed by matrix convert- NSERC Synergy Award for Innovation.
ers,” IEEE Trans. Ind. Electron., vol. 55, no. 12, pp. 4372–4380,
Dec. 2008.
[37] M. Perez, J. Rodriguez, and A. Coccia, “Predictive current control in
Marco Rivera (S’09–M’11) received the B.Sc.
a single phase PFC boost rectifier,” in Proc. IEEE ICIT, Gippsland,
degree in electronics engineering and the M.Sc.
Australia, Feb. 2009, pp. 1–6.
degree in electrical engineering from Universidad
[38] G. Sinha and T. Lipo, “A four-level inverter based drive with a passive de Concepción, Biobío, Chile, in 2007 and 2008,
front end,” IEEE Trans. Power Electron., vol. 15, no. 2, pp. 285–294, respectively, and the Ph.D. degree from the Depart-
Mar. 2000. ment of Electronics Engineering, Universidad Téc-
[39] M. Saeedifard, R. Iravani, and J. Pou, “A space vector modulation nica Federico Santa María, Valparaíso, Chile, in
approach for a back-to-back connected four-level converter,” in Proc. 2011.
IEEE PESC, Orlando, FL, USA, Jun. 2007, pp. 2043–2049. He was a Post-Doctoral Fellow with Universidad
[40] G. Goodwin, D. Carrasco, and M. Seron, “Predictive control: A his- Técnica Federico Santa María in 2011 and 2012,
torical perspective,” Int. J. Robust Nonlinear Control, vol. 22, no. 12, and he is currently a Professor with Universidad de
pp. 1296–1313, Aug. 2012. Talca, Talca, Chile. His current research interests include matrix converters,
[41] V. Yaramasu, B. Wu, M. Rivera, J. Rodriguez, and A. Wilson, predictive and digital controls for high-power drives, four-leg converters,
“Cost-function based predictive voltage control of two-level four- renewable energies, and development of high performance control platforms
leg inverters using two step prediction horizon for standalone power based on field-programmable gate arrays.
systems,” in Proc. IEEE APEC, Orlando, FL, USA, Feb. 2012,
pp. 128–135.
[42] P. Cortes, J. Rodriguez, S. Vazquez, and L. Franquelo, “Predictive Jose Rodriguez (M’81–SM’94–F’10) received the
control of a three-phase UPS inverter using two steps prediction Engineer degree in electrical engineering from Uni-
horizon,” in Proc. IEEE ICIT, Viña del Mar, Chile, Mar. 2010, versidad Técnica Federico Santa María, Valparaíso,
pp. 1283–1288. Chile, in 1977, and the Dr.-Ing. degree in electrical
[43] T. Geyer, “Generalized model predictive direct torque control: Long engineering from the University of Erlangen, Erlan-
prediction horizons and minimization of switching losses,” in Proc. IEEE gen, Germany, in 1985.
CCC, Shanghai, China, Dec. 2009, pp. 6799–6804. He has been with the Department of Electronics
[44] K. Tan and S. Islam, “Optimum control strategies in energy conversion Engineering, Universidad Técnica Federico Santa
of PMSG wind turbine system without mechanical sensors,” IEEE Trans. María, since 1977, where he is currently a Full Pro-
Energy Convers., vol. 19, no. 2, pp. 392–399, Jun. 2004. fessor and Rector. He has co-authored more than 350
[45] P. Cortes, S. Kouro, B. La Rocca, R. Vargas, J. Rodriguez, J. Leon, et al., journal and conference papers. His current research
“Guidelines for weighting factors design in model predictive control of interests include multilevel inverters, new converter topologies, control of
power converters and drives,” in Proc. IEEE ICIT, Gippsland, Australia, power converters, and adjustable-speed drives.
Feb. 2009, pp. 1–7. Dr. Rodriguez is a member of the Chilean Academy of Engineering.