This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.

Model Predictive Control of Grid-Tied Four-Level

Diode-Clamped Inverters for High Power
Wind Energy Conversion Systems
Venkata Yaramasu, Student Member, IEEE, Bin Wu, Fellow Member, IEEE and Jin Chen, Member, IEEE
Department of Electrical and Computer Engineering, Ryerson University, Toronto, ON, CANADA
E-mail: [email protected]; [email protected]; [email protected].

Abstract—In this paper, a model predictive control of medium Rxg Grid filter resistance of phase x = a, b, c
voltage, grid-tied four-level diode-clamped inverter for use in Lxg Grid filter inductance of phase x = a, b, c
high power wind energy conversion systems is presented. The θg Grid voltage angle
control objectives such as regulation of net DC-link voltage, Ts Sampling time
reactive power generation to meet the grid operator request, λdc , λswc Weighting factors
DC-link capacitor voltages balancing and switching frequency
minimization are included in the cost function. The latter
objective is an important requirement for high power converters, I. I NTRODUCTION
and this has been achieved without any modifications to the Due to the increased penetration of wind energy into the
software or hardware configuration. The future behavior of the
grid currents and DC-link voltages are predicted for all the energy market in recent years, the turbine manufacturers
possible switching states using the discrete-time model of the are actively involved in the research and development of
inverter, DC-link, inductive filter and grid. The switching state megawatt-level wind turbines [1]. The full-scale power con-
that minimizes the cost function has been chosen and applied to verters are used in variable speed wind energy conversion
the inverter directly. During different operating conditions, the systems (WECS) to harvest more energy, increase efficiency
switching frequency is regulated between 750 and 850Hz with
the help of a lookup table based weighting factor, and thus the and to meet the grid codes [2], [3]. For the WECS greater than
proposed methodology appears as a promising tool to control 2MW power level, the multilevel diode-clamped converters
the high power grid-tied inverters. The results obtained through operating at a medium voltage (MV) of 3000-4000V offer cost
the MATLAB simulations on 4MVA/4000V system and dSPACE effective, compact, reliable and efficient solutions compared to
DS1103 based experiments on 5kW/208V prototype are in a close the low voltage (<1000V) cascaded two-level converters [4]–
relationship, and thus validate the proposed control strategy.
[7]. In comparison to the most popular three-level inverter [7],
Index Terms—Current control, DC-AC power conversion, DC- [8], the four-level inverter for high power WECS offers added
link capacitor voltages balancing, Digital control, Diode-clamped benefits such as: medium voltage operation with greatly re-
converters, Discrete time signals, Finite control set Model predic-
tive control, Grid-connected, Multilevel inverters, Wind energy duced device voltage rating [9], increased equivalent switching
frequency [10], lower values for grid-side filter, grid friendly
waveforms, and grid code compliance.
The well-established classical control techniques for grid-
N OMENCLATURE tied inverters are virtual flux-oriented control [11], direct
A, Bi , Bg Parameters matrix dimension 2 × 2 power control [12], [13] and decoupled voltage-oriented con-
Φ, Γi , Γg Parameters matrix dimension 2 × 2 trol (VOC) [1], [7], [14]. These control techniques use PI reg-
ig Grid current vector [iag ibg icg ]T ulators in their internal current control loop and a modulation
vg Grid voltage vector [vag vbg vcg ]T stage to generate the gating signals. A complex modulation
vi Inverter voltage vector [van vbn vcn ]T technique along with the DC-link capacitor voltages balancing
vc Capacitor voltage vector [vc1 vc2 vc3 ]T [15], [16] is required to apply these techniques to four-
Sg , Pg , Qg Grid apparent, active and reactive powers level inverters. The most common modulation techniques for
vdg , vqg Grid d and q-axis voltages four-level inverters are: carrier-based pulse width modulation
idg , iqg Grid d and q-axis currents (PWM) [17], [18] and space vector modulation (SVM) [19],
i∗dg , i∗qg Grid d and q-axis reference currents [20]. The device switching frequency reduction to below 1kHZ
vdi , vqi Inverter d and q-axis voltages is another important issue at the megawatt-level to minimize
vdc Net DC-link voltage the switching losses and to allow proper heat dissipation [21].
vdc
∗
Reference DC-link voltage The PWM and SVM techniques produce significantly lower
vcj Voltage across DC capacitors, j = 1, 2, 3 order harmonics when operated below 1kHZ and this causes
icj Current in DC capacitors, j = 1, 2, 3 poor power quality [22].
Cdcj Capacitance of DC capacitors, j = 1, 2, 3 Recently, the finite-control set model predictive control
Sjx Switching states, j = 1, 2, 3, x = a, b, c (FCS-MPC) has emerged as a simple and powerful tool to
vxg /ixg Grid voltage/current of phase x = a, b, c control the power converters and drives [23]–[32]. This method

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

offers a fast dynamic response and the ability to include current to control the reactive power, and (iv) balancing of
nonlinearities and constraints in the design of the controller DC-link capacitor voltages [1]. The proposed control scheme
[33]. This predictive strategy has been analyzed before through predicts the future behavior of the d and q-axis grid currents
simulations for low power four-level inverters with resistive and DC-link capacitor voltages for each valid switching state
load [34]. In the grid-tied wind energy systems, the active of the inverter using the measured grid currents and DC-link
power changes randomly in proportion to the wind speed. voltages. These predictions are evaluated using a cost function
The reactive power also changes as per the grid operator and the switching state which minimizes the cost function
request. The inverter switching frequency varies significantly is selected and applied to the inverter directly. The classical
with these changes in active and reactive powers. This variable VOC defacto requirements for internal current control loops
switching frequency operation causes difficulty in the grid and modulation are eliminated by this scheme. To validate the
filter design [35] and thus regulation of switching frequency proposed method, simulations are carried out on a high power
to certain limits is an important requirement. This issue has (4MVA/4000V) system using MATLAB/Simulink software
been deeply analyzed and a look-up table based weighting and then experimental tests are conducted on a low power
factor selection is proposed. With this approach, the switching prototype (5kW/208V) using dSPACE DS1103 controller.
frequency is regulated between 750 and 850 Hz during all This paper is organized as follows: in section II, the
the operating conditions. It is important to note that the modeling of grid-currents and DC-link capacitor voltages
proposed approach doesn’t need any software or hardware is presented, followed by the explanation of the proposed
reconfiguration. control strategy in section III. The simulation analysis and
The predictive controller design in stationary reference experimental implementation are discussed in sections IV and
frame (αβ) has been carried out in many works [34], [36]– V respectively, and finally in section VI, conclusions are
[38]. For grid-tied applications, the generation of d and q- drawn.
axis reference currents and regulation of net DC-link voltage
becomes easier with the synchronous reference frame (dq) II. G RID -T IED F OUR -L EVEL I NVERTER M ODEL
approach. Moreover the controller tracks to set-reference point, A medium voltage permanent magnet synchronous genera-
rather than sinusoidal tracking [39]. In this paper, modeling tor based WECS with (passive or active) front-end converter
and control of grid currents is performed in synchronous and grid-tied four-level diode-clamped inverter is shown in
reference frame. Fig. 1. The inverter is composed of 18 active switches and
For high power grid-tied inverters, four variables need to be 18 clamping diodes. The clamping diodes withstand different
tightly controlled: (i) regulation of net DC-link voltage, (ii) d- reverse blocking voltages, and if the voltage rating for all the
axis grid current to control the active power, (iii) q-axis grid clamping diodes is selected to be same as the active switches,

idc P(3) ii3

S1a S1b S1c

+ ic3
Front-end vc3 cdc3 S2a S2b S2c
converter −
S3a S3b S3c
ii2
z2 (2)
+ ic2
vdc vc2 cdc2 a b c
−
ii1
z1 (1)
Gear box PMSG S 1a S 1b S 1c
(optional) (MV) ic1
+
Wind vc1 cdc1 S 2a S 2b S 2c
Turbine −
S 3a S 3b S 3c
ii0
N(0)
iag ibg icg

Rag Rbg Rcg

Lag Lbg Lcg

+ + +
vag vbg vcg
− − −
n

Figure 1. Topology of grid-tied four-level diode-clamped inverter used in the medium voltage wind energy systems.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

then series connection of two diodes is necessary [10] as A. Modeling of Grid Currents in Synchronous Frame
shown in Fig. 1. The grid currents can be given in terms of inverter voltages,
The switching states and the corresponding inverter terminal grid voltages and filter inductance in natural frame (abc) as
voltages are shown in Table I, where it can be noted that: [41]:
(a) only three switches conduct at any time, and (b) switch d ig 1
= [vi − vg − Rg ig ]. (3)
pairs (S1x , S 1x ), (S2x , S 2x ) and (S3x , S 3x ) operate in a dt Lg
complementary manner [10]. A total of 37 voltage vectors with
In order to facilitate the easier design and control, the mea-
64 (43 ) switching combinations are available for the four-level
sured natural frame grid voltages and currents are converted to
inverter as shown in Fig. 2.
synchronous frame. The above natural frame currents can be
Table I converted to synchronous frame (dq) and expressed in state-
S WITCHING S TATES AND I NVERTER V OLTAGES (x = a, b, c) [10], [34]. space form as follows [14]:
       
d idg idg vdi v
Sx S1x S2x S3x S 1x S 2x S 3x vxN =A + Bi + Bg dg , (4)
dt qgi i qg vqi vqg
3 1 1 1 0 0 0 vc1 + vc2 + vc3
2 0 1 1 1 0 0 vc1 + vc2 where,
1 0 0 1 1 1 0 vc1
" # " #
−Rg 1
Lg ωg Lg 0
0 0 0 0 1 1 1 0 A= , Bi = 1 , Bg = −Bi . (5)
−ωg
−Rg
Lg
0 Lg

The vdi and vqi are the estimated d- and q-axis inverter
vbN β voltages respectively. They are obtained from the natural frame
v25 v24 v23 v22
030 130 230 330 inverter voltages (van , vbn , and vcn ) given in (2).
The discrete-time model for d and q-axis grid currents can
v26 v11 v10 v9 v21
131 231 331
be obtained from (4) for one-step prediction as follows:
031 020 120 220 320        
idg (k + 1) i (k) v (k) v (k)
v27 v12 v3 v2 v8 v20 = Φ dg + Γi di + Γg dg ,
032 132 232
121
332
211 321 310
iqg (k + 1) iqg (k) vqi (k) vqg (k)
021 010 110 210
(6)
v28 v13 v4 v0 v1 v7 v19 v
000
033 133 233
122 111 322
211 311 300
aN where,
022 011 222
100 200 α
333
v29 v14 v5 v6 v18 v36 Φ= eATs ,
012 001 101 201 Z Ts
023 123 112 212 312 301
223 323
Γi = eATs Bi dτ =A−1 (Φ − I2x2 )Bi ,
v30 v15 v16 v17 v35 0
(7)
013 002 102 202 302
Z Ts
113 213 313
Γg = eATs Bg dτ =A−1 (Φ − I2x2 )Bg .
v31 v32 v33 v34 0
003 103 203 303
vcN B. Modeling of DC-Link Capacitor Voltages
Figure 2. Space vector diagram and switching states for the four-level inverter The DC-link capacitor voltages can be expressed in terms
[40]. of the DC-link capacitor currents as follows [38], [42], [43]:
Z t
1
From Table. I, the voltage in any phase-x of the inverter, vcj (t) = vcj (0) + icj (τ ) dτ. j = 1, 2, 3 (8)
Cdcj 0+
measured from the negative point of the DC-link (N ) can be
expressed in terms of switching signals and DC-link capacitor Refereing to Fig. 1, the DC-link capacitor currents can be
voltages as [20]: derived from idc , ii1 , ii2 and ii3 as follows [20]:
vxN = (Sx == 0) · (0) ic3 = idc − ii3 ,



+ (Sx == 1) · (vc1 ) ic2 = ic3 − ii2 = idc − ii3 − ii2 , (9)


x = a, b, c (1)
+ (Sx == 2) · (vc1 + vc2 ) 
 ic1 = ic2 − ii1 = idc − ii3 − ii2 − ii1 .

+ (Sx == 3) · (vc1 + vc2 + vc3 ).

The objective of the controller is to maintain equal energy
These inverter voltages can be formulated with respect to among the capacitors. With this condition, the net DC current
the grid neutral (n) as [33]: idc becomes nullified [44] and the above expression can be
modified as:
2 1 1 ic3 = −ii3 ,
van = + vaN − vbN − vcN ,
3 3 3 ic2 = −ii3 − ii2 , (10)
1 2 1
vbn = − vaN + vbN − vcN , (2) ic1 = −ii3 − ii2 − ii1 .
3 3 3
1 1 2 The currents ii1 , ii2 and ii3 can be expressed in terms of
vcn = − vaN − vbN + vcN .
3 3 3 switching signals and measured grid currents as follows and

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

thus there is no need to measure them [34]: reference reactive power that needs to be provided by the
X wind turbine to support the grid voltage. This Q∗g can be set
ii3 = (Sx == 3) · ixg ,
to zero for unity, negative for leading, and positive for the
x=a,b,c
X lagging power factor. The vdc∗
is the reference net DC-link
ii2 = (Sx == 2) · ixg , (11) voltage and it can be set to 3.062 times grid phase voltage
x=a,b,c [1]. The proposed predictive control is very easy and intuitive
X
ii1 = (Sx == 1) · ixg . to understand. The implementation procedure is discussed as
x=a,b,c follows:
By substituting (11) in (10), the DC-link capacitor currents • Measure grid voltages, vag (k) and vbg (k), (vcg (k) =
can be expressed in terms of three-phase grid currents and −vag (k) − vbg (k)); grid currents iag (k) and ibg (k),
gains K1x , K2x , and K3x as follows: (icg (k) = −iag (k) − ibg (k)); and DC-link capacitor
voltages vc1 (k), vc2 (k), vc3 (k). In total 5 voltage sensors
X
ic3 = K3x · ixg ,
x=a,b,c and two current sensors are required for a balanced
X system.
ic2 = K2x · ixg , (12) • Obtain grid voltage angle, θg (k) using synchronous ref-
x=a,b,c
X erence frame phased lock loop (SRF-PLL) [14].
ic1 = K1x · ixg . • Convert grid voltages and currents from natural (abc)
x=a,b,c frame to synchronous (dq) frame, vdg (k), vqg (k), idg (k),
The gains K1x , K2x , and K3x are based on the inverter iqg (k). For easier analysis, the grid voltage orientation
switching states as demonstrated below: can be used by which the d-axis of the synchronous frame
is aligned with the grid voltage vector and as a result q-
K1x = sign (0 − Sx ), axis grid voltage becomes zero [1].
K2x = sign (1 − Sx ) · sign (Sx − 0), (13) • Obtain q-axis reference current i∗qg (k) from Q∗g (k). With
K3x = sign (2 − Sx ) · sign (Sx − 0) · sign (Sx − 1), grid voltage orientation, the grid reactive power can be
calculated as [45]:
where, sign is a Signum function whose output value corre- 
sponds to −1 or 0 or +1. 3 3
Qg = (vqg idg − vdg iqg ) = − vdg iqg , vqg = 0
The discrete-time model for DC-link capacitor voltages can 2 2
be obtained from (8) for j = 1, 2, 3, as [38]: (15)
from which the reference q-axis current can be calculated
Ts as:
vcj (k + 1) = vcj (k) + icj (k + 1). (14)
Cdcj Q∗g (k)
i∗qg (k) = . (16)
The performance of the model in [34] and the proposed −1.5 vdg (k)
model is same in terms of the DC-link capacitor voltages • Obtain d-axis reference current i∗dg (k) which represents
balancing, but the number of online calculations can be the active power of the system. Neglecting losses in
reduced with the presented method. The reason is that the inverter, the active power Pg on AC-side equals DC-
gains K1x , K2x , and K3x can be defined off-line along with power as follows [1]:
the switching states. Compared to [34], the DC-link capacitor
3
voltages modeling presented here can be understood more Pg = vdg idg = vdc idc . (17)
2
intuitively and can easily be extended to any level diode-
clamped inverter. During steady-state, the net DC-link voltage, vdc is main-
From (6), it can be noticed that the grid d and q-axis tained at its reference value vdc∗
by the PI controller.
currents are a function of inverter voltages vdi and vqi . These The idg (k) is generated by this DC-link PI controller
∗

inverter voltages (in natural frame) are related to the switching according to the operating conditions. In the complete
signals as described in (1). Similarly, the models in (12)-(14) wind energy system, the generator-side converter injects
demonstrates that the DC-link capacitor voltages are related to current idc according to the maximum energy harvested
the switching signals. From the modeling given in this section, from the wind. When the wind turbine operates below
it can be understood that the grid currents and capacitor cut-in wind speed, the idc supplied by the generator-side
voltages can be controlled by properly choosing a switching converter is zero. Similarly, when the turbine operates at
state. rated wind speed, the idc is rated. In order to simulate
such a variable-speed wind energy system, a variable DC-
current source is generated with the help of battery supply
III. M ODEL P REDICTIVE C ONTROL E and a resistor Rdc [1], [14]. The amount of active
The proposed model predictive control scheme is shown power injected into grid can be varied by changing the
in Fig. 3, where the wind turbine, generator and rectifier are value of idc (refer to (17)). When the battery voltage
represented by a battery in series with a small resistance that E > vdc , the idc (=(E − vdc )/Rdc ) becomes positive and
represents the power loss in the system [1]. The power flow power flows from the inverter to grid. For a given value
is considered from the inverter to the grid. The Q∗g is the of Rdc , by adjusting E value, the active power generation

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Four-level Inverter Rag Lag vag

idc iag
+
vc3 (k)
Rdc −
Rbg Lbg vbg
+ ibg
vc2 (k) n
−
E +
vc1 (k) Rcg Lcg vcg
− icg

Power flow 18 Power flow

Sjx (k) SRF-PLL
i∗dg (k + 1)
λdc
Cost Function iag (k) ibg (k) vag (k) vbg (k)
i∗dg (k + 1) Lagrange
λswc Minimization i∗qg (k + 1) Extrapolation abc/dq
i∗qg (k + 1) Transformation
Look-Up vc1 (k + 1) i∗qg (k) i∗dg (k) θg (k)
Table idg (k + 1) iqg (k + 1) vc2 (k + 1)
vc3 (k + 1) ÷ PI
−1.5vdg (k)
Sjx (k − 1)
Predictive + vc1 (k)
θg (k) Q∗g (k) idg (k) iqg (k) vdg (k) vqg (k)
Model
− + vc2 (k)
+
vc1 (k) + vc3 (k)
vc2 (k)
vc3 (k) vdc
∗
(k)
vdg (k) vqg (k) idg (k) iqg (k)

Figure 3. Proposed model predictive control algorithm for grid-tied four-level inverter.

can be adjusted. It should be noted that the generation of quadratic cost function for one-step prediction as follows:
reference dq-axis currents is similar to the classical VOC
g(k) = (i∗dg (k + 1) − idg (k + 1))2
given in [1], [14].
• Extrapolate reference currents from k th state to (k + 1) + (i∗qg (k + 1) − iqg (k + 1))2
state. The fourth-order lagrange extrapolation method is 2
X
considered as follows [26], [32], [46]: + λdc ∗ { ([vcj (k + 1) − vcj+1 (k + 1)]2 )
(19)
j=1
i∗dg (k + 1) = 4 i∗dg (k) − 6 i∗dg (k − 1) 2
+ [vc1 (k + 1) − vc3 (k + 1)] }
+ 4 i∗dg (k − 2) − i∗dg (k − 3), X
(18) + λswc ∗ swcx ,
i∗qg (k + 1) = 4 i∗qg (k) − 6 i∗qg (k − 1)
x=a,b,c
+ 4 i∗qg (k − 2) − i∗qg (k − 3).
where, λdc and λswc are weighting factors for the DC-
• Predict the future behavior of the d and q-axis grid link capacitor voltages balancing and switching frequency
currents (idg (k+1) and iqg (k+1)) and DC-link capacitor reduction respectively. The swcx is the number of semi-
voltages (vc1 (k +1), vc2 (k +1) and vc3 (k +1)) for all the conductor commutations involved in phase-x, which can
possible switching states of inverter using (6) and (14), be calculated as follows [33]:
• Estimate λswc based on extrapolated reference currents
i∗dg (k + 1), i∗qg (k + 1), swcx =| Sjx (k) − Sjx,opt (k) |, j = 1, 2, 3, x = a, b, c
• The predicted variables idg (k + 1), iqg (k + 1), vc1 (k + 1), (20)
vc2 (k + 1), vc3 (k + 1) which are function of switching where, Sjx (k) is the predicted gating signal of phase-
signals are used in the cost function. The gating signal x, and Sjx,opt (k) = Sjx (k − 1) is the optimal gating
Sjx (k) (j=1, 2, 3, x=a, b, c) which minimizes the cost signal of phase-x in the previous sample. When λswc > 0,
function is chosen and applied to the inverter directly. It the switching frequency reduction can be achieved. In a
should be noted that the gating signals are directly gener- similar manner, the DC-link capacitor voltages balancing
ated by the controller and any additional modulation stage can be accomplished with λdc > 0.
is not required, unlike in the classical control techniques
[1], [7], [11]–[14]. The control objectives of the proposed
IV. S IMULATION R ESULTS
method are: (i) regulation of active power for which the d-
axis current is controlled; (ii) regulation of reactive power To validate the proposed control scheme, simulations are
for which q-axis current is controlled; (iii) balancing carried out using MATLAB/Simulink software for different
of DC-link capacitor voltages; and (iv) minimization of operating conditions with the parameters as indicated in Table
switching frequency for which the number of switch II. One-step prediction horizon is considered in all the simula-
changes are penalized. All these goals are included into a tion tests. Unless otherwise stated, the cost function includes
all four variables as mentioned in (19).

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Table II
G RID -T IED F OUR - LEVEL I NVERTER AND C ONTROLLER PARAMETERS A1 corresponds to the pure active power injection into the grid,
which is the most common scenario in WECS [1]. The grid
current is maintained in phase with the grid voltage as shown
Simulation Experimental in Fig. 4a-(i). The active power component of grid current idg
Variable Description SI pu SI pu equals peak grid current and the reactive power component of
Sg Apparent Power (kVA) 4000 1.0 5 1.0
vg Grid L-L rms Voltage (V) 4000 – 208 – grid current iqg becomes zero as demonstrated in Fig. 4a-(i).
ig Grid rms Current (A) 577.35 1.0 13.88 1.0 With Qg = 0, the grid apparent power Sg becomes equals
fg Grid Frequency (Hz) 60 1.0 60 1.0 to the grid active power Pg as shown in Fig. 4a-(ii). The
Rg Filter Resistance (Ω) 0.042 0.0105 0.091 0.0105 net DC-link voltage vdc is maintained at its reference value
Lg Filter Inductance (mH) 2.1 0.2 4.6 0.2 with the perfect balancing of the capacitor voltages as shown
Cdc DC-link Capacitance (µF) 2596 3.914 1200 3.914
∗
vdc Ref DC-link Voltage (V) 7071 3.062 368.0 3.062 in Fig. 4a-(iii). The DC-link capacitor voltages deviation can
Rdc DC-link Resistance (Ω) 0.4624 0.1156 1 0.1156 be further reduced by increasing the λdc , but at an expense
Rx External Resistance (Ω) 46.24 11.56 100 11.56 of higher %eig and %THD. With λswc = 0, the switching
λdc Weighting Factor 0.5 – 0.25 – frequency is found to be 1338Hz. By using λswc = 1280
kP , kI DC-Link PI Parameters 1&200 – 1&200 – (obtained from look-up table), the switching frequency fsw
Ts Sampling Time (sec) 90µ – 90µ –
Base Values
is reduced to 798Hz. The reduced active power delivery with
vbg Base rms Voltage (V) 2309.4 1.0 120 1.0 leading and lagging power factors are analyzed in Case A2 and
ibg Base rms Current (A) 577.35 1.0 13.88 1.0 A3, respectively. The Case A4 corresponds to the operation of
Zbg Base Impedance (Ω) 4.0 1.0 8.6528 1.0 WECS below the cut-in wind speed, where the active power
Lbg Base Inductance (mH) 10.6 1.0 23.0 1.0 generated and delivered to the grid is zero.
Cbg Base Capacitance (µF) 663.15 1.0 306.56 1.0
The fault ride-through capability of the proposed inverter
is shown in Case A5. According to the grid codes [47], [48],
Table III the high power WECS should remain connected to the grid
S UMMARY OF S IMULATION S TEADY-S TATE A NALYSIS
during the grid faults and inject pure leading reactive power
to help the grid recover from the fault [3]. The detection of
Case i∗dg (pu) Q∗g (pu) %eig %THD %evc fsw (Hz) grid fault and generation of corresponding current references
A1 1.0 0 3.26 3.20 1.02 798 [49] is beyond the scope of this paper. For simplicity, the
A2 0.8 −0.6 3.68 3.88 1.41 799 current references are set by the user. The grid current leads
A3 0.8 0.6 3.09 2.92 1.39 804
the voltage by 90 degrees as shown in Fig. 4b-(i). The idg
A4 0 0 −− −− 0.03 798
A5 0 −1 2.18 2.07 0.96 804 becomes zero and the iqg becomes equal to the peak grid
A6 0 1 4.19 3.15 1.78 800 current. The active power is maintained at zero as shown in
Fig. 4b-(ii), and the grid apparent power becomes equal to the
absolute reactive power. The DC-link capacitor voltages are
well balanced as shown in Fig. 4b-(iii). This case demonstrates
A. Steady-State Analysis
the effectiveness of the proposed method to meet the grid
The steady-state analysis has been carried out for different codes. The injection of pure lagging reactive power, although
d and q-axis reference currents as summarized in Table III. rare in WECS, has been studied in A6. The inverter switching
The performance assessment parameters %eig , %THD, %evc frequency fsw variation is maintained in the range of 750-
and fsw are calculated for 3 fundamental cycles according to 850Hz during all the operating conditions while maintaining
the guidelines given in [26], [33], [36]. The absolute mean acceptable power quality and reference tracking.
reference tracking error %eig is computed with respect to the
peak base current. The absolute DC-link capacitor voltages B. Transient-State Analysis
deviation %evc is calculated with respect to the reference DC-
link voltage. The average switching frequency is computed To analyze the decoupled nature of the d and q-axis cur-
by measuring the number of switching changes in the gating rents, transient analysis has been carried out with trapezoidal
signals. The battery voltage E is adjusted to simulate the changes in Q∗g . The i∗dg is set to 0.8pu. As shown in Fig. 4c-(i),
variable DC-current source. As discussed in (17), by changing the grid power factor changes in the order of: unity-leading-
idc magnitude (with the help of E), the i∗dg magnitude can be unity-lagging. The grid reactive power tracks to its reference
varied proportionally. The relation between the battery voltage with fast dynamics and no overshoots. The decoupled control
E and required i∗dg is shown below: for active and reactive powers (thus for d and q-axis currents,
respectively) has been achieved as demonstrated in Fig. 4c-
E = vdc
∗
+ (i∗dg (pu) × ibg × Rdc ). (21) (ii). A good net DC-link voltage control and DC-link capacitor
voltages balancing has been achieved even during the transient
In order to obtain different i∗dg
values for a given Rdc , the
interval (refer to Fig. 4c-(iii)).
battery voltage E is adjusted. With the grid voltage orientation,
the vdg equals to the peak grid voltage and vqg becomes zero
during all operating conditions as shown in Fig. 4. The inverter C. DC-Link Dynamics
phase voltage (refer to Eq.(1)) is also shown where it contains In order to verify the proposed DC-link capacitor voltages
four-levels: 0, vdc /3, 2vdc /3 and vdc . The operating condition balancing algorithm, an external resistor Rx (11.46pu) is

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

vag idg vag iqg vag

+2kV vdg +1kA +2kV vdg +1kA +2kV
idg vdg +1kA
φg =0◦ vqg φg =90 lead◦
vqg vqg φ =36.8◦ lag
φg =36.8◦ lead g
0 0 0
0 0 0
i∗ag iqg iag idg φg =0 ◦
i∗ag
−2kV −1kA −2kV iag −1kA −2kV
iag iqg
−1kA
i∗ag

(i) (i) (i)

+4M +4M +4M

+2M Pg Sg +2M Pg +2M
Sg Pg Sg
0 0 0
−2M −2M Q∗g Qg −2M
Q∗g Qg
−4M −4M −4M
Q∗g Qg
vdc vdc vdc
2vdc /3 2vdc /3 2vdc /3
vdc /3 vdc /3 vdc /3
vaN 0 vaN 0 vaN 0
(ii) (ii) (ii)
8kV 8kV 8kV

6kV vdc
∗ vdc 6kV vdc
∗ vdc 6kV vdc
∗ vdc

4kV 4kV 4kV

2kV 2kV 2kV

vc1 vc2 vc3 vc1 vc2 vc3 vc1 vc2 vc3
0 0 0
(iii) (iii) (iii)
0 0.5T T 1.5T 2T 2.5T 3T 0 0.5T T 1.5T 2T 2.5T 3T 0 0.5T T 1.5T 2T 2.5T 3T
time (sec) time (sec) time (sec)
(a) (b) (c)
Figure 4. Simulation results for: (a) case-A1: i∗dg = 1.0pu and Q∗g = 0pu, (b) case-A5: i∗dg = 0pu and Q∗g = −1.0pu, and (c) transient-state: i∗dg = 0.8pu
and trapezoidal change in Q∗g . (Scale: k=103 , M=106 , T=1/60.)

8kV 8kV

6kV vdc vdc

∗
60A 6kV
vdc vdc
∗

iRx
4kV 40A 4kV

2kV 20A 2kV vc1

vc3 vc3
vc1 vc2
vc2
0 0 0
0 0.5T T 1.5T 2T 2.5T 3T 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
time (sec) time (sec)
(a) (b)
Figure 5. Simulation results for DC-link capacitor voltages balancing: (a) with a resistor across the DC-link capacitor−3, Cdc3 , and (b) with and without
the weighting factor λdc . (Scale: k=103 , T=1/60.)

deliberately connected across the DC-link capacitor Cdc3 at voltages balancing, the vc1 and vc3 gradually increases, while
time t=1/60 seconds. The d and q-axis reference currents are vc2 gradually decreases and becomes zero at 0.24 seconds.
same as in the Case A1. Even with the step-connection of the At the time t=0.5 seconds, the weighting factor has been
resistor, the controller takes action in a few sampling instants, changed to 0.5 and the DC-link capacitor voltages become
and thus the capacitor voltages continue to be balanced as balanced, with a fast recovery time of 0.06 seconds. These
shown in Fig. 5a. The current through the resistor iRx is results validate the proposed DC-link modeling and control.
found to be 11.33% of the base current. It is important to
note that the grid currents continue to track to their references
D. Switching Frequency Regulation
during this process. Another test has been conducted where
the weighting factor λdc is suddenly changed from 0.5 to The variable switching frequency nature of the predictive
zero at 0.2 seconds. With no penalty to the DC-link capacitor control is studied with respect to the i∗dg and Q∗g . The switching
frequency increases linearly with the d-axis grid current.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

2000 2000 2000

fsw (Hz)
fsw (Hz)

1500 1500 1500

λswc
1000 1000 1000
500 500 500
0 1 0 1 0 1
0 0 0 .8 .0 0 0 0 .8 .0 0 0 0 .8 .0
1 . .8 0 . 0 .4 .6 1 . .8 0 . 0 .4 .6 1 . .8 0 . 0 .4 .6
0 .6 − 0 2 0 .6 − 0 2 0 .6 − 0 2
0 .4 − −0 0. 0 .4 − −0 0. 0 .4 − −0 0.
0
id ∗ 0.2 0 − − 0 . 0 .6 .4 2 0
id ∗ 0.2 0 − − 0 . 0 .6 .4 2 0
id ∗ 0.2 0 − − 0 . 0 .6 .4 2
1. 8 1. 8 1. 8
g (p u 0 ∗ u)
(p g (p u 0 ∗ u)
(p g (p u 0 ∗ u)
(p
) (a) Qg ) (b) Qg ) (c) Qg

Figure 6. Simulation results for switching frequency variation with respect to i∗dg and Q∗g : (a) without λswc , and (b) with λswc ; and (c) λswc .

The fsw becomes greater with the negative reactive power V. E XPERIMENTAL I MPLEMENTATION AND VALIDATION
compared to the positive reactive power. Without the weight A. Experimental Setup
factor λswc , the switching frequency is found to vary in a
range of 1800-900Hz as shown in Fig. 6a. To overcome this The experimental setup for the grid-tied four-level inverter
issue, weighting factors as shown in Fig. 6c are designed with is shown in Fig. 8 with the parameters as indicated in Table
the knowledge obtained from the Fig. 6a and the guidelines II. The control algorithm has been implemented using a
given in [50]. Observation demonstrates that these values hold host PC running with MATLAB-Simulink 2006a software
a linear relation to the active and reactive powers. These values through Real-Time Interface (RTI). The dSPACE DS1103
are stored in a lookup table for online use and as a result controller is used to handle the control processes such as
the switching frequency has been maintained within the 750- grid current prediction, DC-link capacitor voltages balancing
850Hz range during all the operating conditions, as shown in and cost function minimization. The grid currents and DC-
Fig. 6b. It should be noted that the previous steady-state results link voltages are measured by LEM LA55-P and LV25-P
presented in the Table III are also obtained by including this transducers, respectively. Feedback from the sensors has been
online look-up table. sent to the controller through the CP1103 I/O connector. The
The %THD performance of four-level inverter with respect prototype has been built by Semikron SKM75GB123D dual-
to the switching frequency for proposed grid-tied and resistive pack IGBT modules. The IGBT gate drivers are based on four
load [34] cases are investigated and presented in Fig. 7. The SKHI22B dual cores which are powered with 0/15V supply.
case-A1 is considered for grid-tied application. For resistive An interface board with MC14504BCP and TLP521-4 is used
load, the controller is implemented in stationary reference between the dSPACE and SKHI22B gate drivers for TTL
frame (αβ) with the parameters given in Table II. The load to CMOS logic conversion and signal isolation, respectively.
resistance value is set as base impedance (Zbg ) of system. For The DC power supply (E) has been obtained by Xantrex
grid-tied case, the weighting factor λswc is changed from 0- XDC-600-20. The inverter is connected to the grid through
20000 and the switching frequency is reduced from 1338 to an isolation transformer and its impedance is added to the
105Hz and the THD is increased from 2.27% to 14%. For filter values.
resistive load case, the λswc is changed from 0-10000. The
fsw is reduced from 1333 to 102Hz with the increase in THD B. Delay Compensation
from 2.18% to 14%. The difference between both the cases is To verify the predictive control algorithm through MATLAB
found to be small during the entire switching frequency range. simulations [34], one-step prediction as shown in Fig. 9a can
It should be noted, for both the cases, a significant reduction be used. This is ideal case where the delay is zero between the
in switching frequency from 1330 to 600Hz has been achieved
with a slight increase in THD from 2.2% to 4.1%.
Host PC
4-Level Inverter
20 CP1103
Interface Board
16 Sensors
With Grid Connection
12
%THD

Oscilloscopes
8
DS1103
4 R&D controller 3-Phase
With Resistive Load Grid
0
1400 1200 1000 800 600 400 200 Isolation Transformer DC Power Supply
Switching Frequency, fsw
L Filter
Figure 7. Comparison of %THD for four-level inverter with grid-tied and
resistive load cases.
Figure 8. Photograph of experimental setup.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

1 1 1 Measurement
vc (k), vg (k), ig (k)

Estimate
θg (k), vdg (k), vqg (k)
idg (k), iqg (k)
64 64 64
k k+1 k+2 k k+1 k+2 Apply
(a) (b) Sjx,op (k), j = 1, 2, 3, x = a, b, c
Figure 9. Prediction of four-level inverter control variables: (a) without delay K1x,op (k), K2x,op (k), K3x,op (k)
compensation [34], and (b) with delay compensation [51], [52]. vdi,op (k), vqi,op (k)

Estimation of Variables at (k + 1) Instant

instant measurements are made and the instant new switching îdg (k + 1), îqg (k + 1), v̂c (k + 1)
state is applied [33]. The switching state which minimizes Using (22) and (23)
the error at (k + 1) instant is selected and applied at (k)
instant. In the experimental verification, the delay provided i=0
by the digital signal processor, gate drivers, and switching gop = ∞
devices is inevitable. The computational delay provided by the
digital signal processor is most significant, and this must be
i=i+1
compensated for. This delay can easily be compensated for by
calculating the cost function with modified one-step prediction
[51], [52] as shown in Fig. 9b. The optimal switching state Predictive Variables at (k + 2) Instant
from the previous iteration is used to estimate the variables at idg (k + 2), iqg (k + 2), vc (k + 2)
(k + 1) instant and all the possible switching states are used
Using (24) and (25)
to predict the variables at (k + 2) instant. The switching state
which minimizes the error at (k + 2) instant is selected and
applied at sampling instant (k + 1). With this approach, one Calculation of Cost Function
sampling period will be available for calculations. g(k + 1)
The control algorithm with this approach of control delay Using (26)
compensation is shown in Fig. 10. The measured grid and DC-
link signals along with the optimal gating signals are used by No
the predictive controller. The variables at (k + 1) instant are i = 64?
estimated using the optimal values obtained from the previous Yes
iteration. This is demonstrated as follows:
        Select iop
îdg (k + 1) idg (k) vdi,op (k) vdg (k)
=Φ + Γi + Γg , gop = min {gi }i=1...64
îqg (k + 1) iqg (k) vqi,op (k) vqg (k)
iop = i |gop
(22)
Ts X
v̂cj (k + 1) = vcj (k) + Kjx,op · ixg . (23) Figure 10. Predictive current control and DC-link capacitor voltages
Cdcj
x=a,b,c balancing algorithm with delay compensation.
The algorithm is initialized by setting the switching state
number i to 0 and optimal g value to ∞ and then the algorithm
enters the loop. The predictive variables at (k + 2) instant are instant are evaluated 64 times by a cost function g(k + 1).
calculated as follows: The switching state which produces minimal value of gop is
     
idg (k + 2) îdg (k + 1) vdi (k + 1) chosen and applied to the inverter directly. The cost function
=Φ + Γi
iqg (k + 2) îqg (k + 1) vqi (k + 1) g(k + 1) can be calculated by shifting the variable in (20)
  (24) one-step forward as follows:
vdg (k + 1)
+ Γg ,
vqg (k + 1) g(k + 1) = (i∗dg (k + 2) − idg (k + 2))2
+ (i∗qg (k + 2) − iqg (k + 2))2
Ts X
vcj (k + 2) = v̂cj (k + 1) + Kjx · ixg . (25) 2
Cdcj + λdc ∗ {
X
([vcj (k + 2) − vcj+1 (k + 2)]2 )
x=a,b,c
(26)
j=1
In (24), vdg (k + 1) and vqg (k + 1) are extrapolated grid 2
voltages and they can be obtained similar to (18). For stiff- + [vc1 (k + 2) − vc3 (k + 2)] }
X
grid, vdg (k) = vdg (k + 1) and vqg (k) = vqg (k + 1) and thus + λswc ∗ swcx .
no extrapolation is needed. The predicted variables at (k + 2) x=a,b,c

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

10

The reference currents at (k + 2) instant can be obtained by online performance assessment took 32µs. The authors in
Lagrange extrapolation as follows [27]: [54] suggested that, the weighting factor based switching
frequency reduction is more efficient compared to the selection
i∗dg (k + 2) = 10 i∗dg (k) − 20 i∗dg (k − 1)
of higher sampling times. With this recommendation, and
+ 15 i∗dg (k − 2) − i∗dg (k − 3), considering high power application, the sampling time (Ts )
(27)
i∗qg (k + 2) = 10 i∗qg (k) − 20 i∗qg (k − 1) of 90µs (11.11kHz sampling frequency) is selected.
+ 15 i∗qg (k − 2) − i∗qg (k − 3).
D. Experimental Results
C. Selection of Sampling Time The experimental results are obtained with the modified
For four-level inverter, 64 switching states are available and one-step prediction (Fig. 9b). Since the predictive controller
the proposed predictive control method incurs large number uses the SI parameters, the weighting factors used in the
of calculations to predict the future behavior of grid currents experiments are different from the simulations, but their impact
and DC-link capacitor voltages. The execution times for two- on the controller remains same. The steady-state analysis has
level inverter with 8 switching states [53] and three-level been carried out similar to the simulations and summarized in
inverter with 27 switching states [36] are reported as 7µs and Table. IV. Due to the modified one-step prediction, the number
52µs, respectively. The minimum execution time of 63µs is of calculations are increased, and as a results the switching
achieved for the proposed control system. It is observed that frequency is increased in all the cases [51]. For instance, for
the predictive control algorithm as shown in Fig. 10 took 31µs Case A1 with λswc = 0, the fsw through simulation studies is
only. The tasks such as measurement of signals, grid voltage 1338Hz and in the experimental implementation it is found to
orientation, detection of grid voltage angle, online estimation be 1736Hz. By using λswc = 1.9 (obtained through look-up
of weighting factors, calculation of reference currents, and table) in experiments, the switching frequency is reduced to

vag vag vag

idg iqg
iqg idg

iqg idg iag

φg =90◦ lead φg =36.8o lead φ =0oφ =36.8o lag
φg =0◦ iag g g
iag

(i) (i) (i)

Qg
Sg Pg Sg Pg Sg
Pg
Qg Qg

vaN vaN vaN

(ii) (ii) (ii)

vdc vdc vdc

vc1 vc2 vc3 vc1 vc2 vc3 vc1 vc2 vc3

(iii) (iii) (iii)

(a) (b) (c)
Figure 11. Experimental results during steady-state for: (a) case-A1: i∗dg = 1.0pu and Q∗g = 0pu, (b) case-A5: i∗dg = 0pu and Q∗g = −1.0pu, and
(c) transient-state: i∗dg = 0.8pu and trapezoidal change in Q∗g . Scope−(i): Ch1: phase-a grid voltage (50V/div), Ch2: d-axis grid current (20A/div), Ch3:
q-axis grid current (20A/div), Ch4: phase-a grid current (20A/div). Scope−(ii): Ch1: grid apparent power (2500VA/div), Ch2: grid active power (2500W/div),
Ch3: grid reactive power (2500VAR/div), Ch4: inverter phase voltage (200V/div). Scope−(iii): Ch1: DC-link capacitor−1 voltage (50V/div), Ch2: DC-link
capacitor−2 voltage (50V/div), Ch3: DC-link capacitor−3 voltage (50V/div), Ch4: Net DC-link voltage (100V/div). Time scale: 5ms/div.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

11

vdc
iRx

vc2
vc1
vc1 vc2 vc3 vc3

(a) (b)
Figure 12. Experimental results for DC-link capacitor voltages balancing: (a) with a resistor across the DC-link capacitor−3, Cdc3 , and (b) with and without
the weighting factor λdc . Scope−a: Ch1: DC-link capacitor−1 voltage (50V/div), Ch2: DC-link capacitor−2 voltage (50V/div), Ch3: DC-link capacitor−3
voltage (50V/div), Ch4: Current through Rx (200mA/div), Time scale: 5ms/div. Scope−b: Ch1: DC-link capacitor−1 voltage (50V/div), Ch2: DC-link
capacitor−2 voltage (50V/div), Ch3: DC-link capacitor−3 voltage (50V/div), Ch4: Net DC-link voltage (100V/div), Time scale: 80ms/div.

4000 2000 4
fsw (Hz)

fsw (Hz)

3000 1500 3

λswc
2000 1000 2
1000 500 1
0 0 0
0 1. 0 1. 0 1.
0 0 .8 0 0 0 .8 0 0 0 .8 0
1 . .8 0 . 0 .4 .6 1 . .8 0 . 0 .4 .6 1 . .8 0 . 0 .4 .6
0 .6 − 0 2 0 .6 − 0 2 0 .6 − 0 2
0 .4 − − 0 0 .2 0 .4 − − 0 0 .2 0 .4 − − 0 0 .2
0 2
id ∗ 0. 0 − − 0 . 0 .6 .4 0 2
id ∗ 0. 0 − − 0 . 0 .6 .4 0 2
id ∗ 0. 0 − − 0 . 0 .6 .4
1. 8 ) 1. 8 ) 1. 8 )
g
(p u 0 ∗ (p u
g
(p u 0 ∗ (p u
g (p u 0 ∗ (p u
) (a) Qg ) (b) Qg ) (c) Qg

Figure 13. Experimental results for switching frequency variation with respect to i∗dg and Q∗g : (a) without λswc , and (b) with λswc ; and (c) λswc .

804Hz. The higher current tracking error in the experiments balancing have been verified experimentally and presented
can be attributed to the delay provided by the sensors and in Fig. 12b. The divergence and convergence of capacitor
IGBT gate drivers and losses in the system. The results shown voltages are similar to the simulation results. The settling time
in Figs. 11a, 11b, and 11c, are similar to the simulation results of capacitor voltages is found to be same as of simulations.
with the perfect regulation of the DC-link voltages and grid The switching frequency variation with respect to the i∗dg
currents. and Q∗g is presented in Fig. 13a, where it is found to vary
Table IV
in the range of 1000-2580Hz. To deal with the lower power
S UMMARY OF E XPERIMENTAL S TEADY-S TATE A NALYSIS prototype, a scaling has been done to evaluate the weighting
factor λswc (refer to Fig. 13c). Similar to the simulations, the
switching frequency is maintained in the range of 750-850Hz
Case i∗dg (pu) Q∗g (pu) %eig %THD %evc fsw (Hz) as shown in Fig. 13b.
A1 1.0 0 4.83 4.16 0.58 804
A2 0.8 −0.6 4.55 4.41 0.71 796
A3 0.8 0.6 3.76 3.74 0.59 811 VI. C ONCLUSION
A4 0 0 −− −− 0.13 800 In this paper, a voltage-oriented, finite control set model
A5 0 −1 3.35 3.58 0.29 808 predictive control strategy is proposed for grid-tied four-level
A6 0 1 4.97 4.03 0.79 807
inverter for use in high power wind energy systems. The
future behavior of the grid currents and the DC-link capacitor
The DC-link dynamics, with the external resistor Rx across voltages for each of the 64 possible switching states are pre-
Cdc3 is shown in Fig. 12a. Similar to the simulations, the bal- dicted based on the discrete-time system model. The switching
ancing of the DC-link capacitor voltages have been achieved state which minimizes the cost function is selected and then
even with the step-connection of the resistor. During the applied directly to the inverter gating terminals. The simulation
step change, transients can be noticed by looking into the and experimental results are in a close relationship and thus
current waveform iRx . The DC-link voltage drops slightly and validate the proposed methodology. The major contributions
recovers back to normal within 1millisecond. Similar to the of this work are summarized as follows:
simulations, the current iRx is found to be 11.33% of the (i) The d and q-axis grid currents are tightly regulated
base current. The relation between the λdc and the capacitors during different active and reactive power conditions using

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

12

a simple and more intuitive approach. The internal current PI [14] R. Teodorescu, M. Liserre, and P. Rodriguez, Grid Converters for
controllers and modulation stage are eliminated and as a result, Photovoltaic and Wind Power Systems. Chichester, UK: Wiley-IEEE
Press, Jan. 2011.
fast dynamic response has been achieved. [15] M. Marchesoni and P. Tenca, “Diode-clamped multilevel converters:
(ii) An intuitive modeling and control for the DC-link a practicable way to balance DC-link voltages,” IEEE Trans. Ind.
capacitor voltages balancing is presented and verified. The Electron., vol. 49, no. 4, pp. 752–765, Aug. 2002.
[16] Z. Shu, X. He, Z. Wang, D. Qiu, and Y. Jing, “Voltage balancing
methodology presented in this paper can be easily extended to approaches for diode-clamped multilevel converters using auxiliary
any level diode-clamped inverter. The proposed methodology capacitor-based circuits,” IEEE Trans. Power Electron., vol. 28, no. 5,
can compensate the perturbations in the DC-link while the grid pp. 2111–2124, May. 2013.
[17] S. Busquets Monge, J. Bordonau, and J. Rocabert, “A virtual-vector
currents continue to track to their references. pulsewidth modulation for the four-level diode-clamped DC-AC con-
(iii) A lookup table based weighting factor design is pro- verter,” IEEE Trans. Power Electron., vol. 23, no. 4, pp. 1964–1972,
posed by which the switching frequency is regulated between Jul. 2008.
[18] F. Grigoletto and H. Pinheiro, “Generalised pulse width modulation ap-
750 and 850Hz during different active and reactive power proach for DC capacitor voltage balancing in diode-clamped multilevel
references and this has been achieved without any software or converters,” IET Pwr Electr., vol. 4, no. 1, pp. 89–100, Jan. 2011.
hardware reconfiguration. A significant reduction in switching [19] S. Busquets-Monge, S. Alepuz, J. Bordonau, and J. Peracaula, “Voltage
frequency is achieved while maintaining acceptable power balancing control of diode-clamped multilevel converters with passive
front-ends,” IEEE Trans. Power Electron., vol. 23, no. 4, pp. 1751–1758,
quality. This feature promotes the proposed methodology for Jul. 2008.
high power grid-tied applications. [20] M. Saeedifard, R. Iravani, and J. Pou, “A space vector modulation
approach for a back-to-back connected four-level converter,” in IEEE
ACKNOWLEDGMENT Power Electron. Spec. Conf. (PESC), Jun. 2007, pp. 2043–2049.
[21] S. Rivera, S. Kouro, P. Cortes, S. Alepuz, M. Malinowski, B. Wu,
The authors wish to thank the financial support from the and J. Rodriguez, “Generalized direct power control for grid connected
Natural Sciences and Engineering Research Council of Canada multilevel converters,” in IEEE Int. Conf. on Ind. Tech. (ICIT), Mar.
2010, pp. 1351–1358.
(NSERC) through Wind Energy Strategic Network (WESNet) [22] S. Kouro, B. La Rocca, P. Cortes, S. Alepuz, B. Wu, and J. Ro-
Project 3.1. driguez, “Predictive control based selective harmonic elimination with
low switching frequency for multilevel converters,” in IEEE Energy
R EFERENCES Conversion Congr. and Expo. (ECCE), Sep. 2009, pp. 3130–3136, San
Jose, CA, USA.
[1] B. Wu, Y. Lang, N. Zargari, and S. Kouro, Power Conversion and
[23] S. Kouro, P. Cortes, R. Vargas, U. Ammann, and J. Rodriguez, “Model
Control of Wind Energy Systems, 1st ed., ser. IEEE Press Series on
predictive control-A simple and powerful method to control power
Power Engineering. Hoboken, NJ: John Wiley & Sons, Inc., Jul. 2011. converters,” IEEE Trans. Ind. Electron., vol. 56, no. 6, pp. 1826–1838,
[2] Z. Chen, J. Guerrero, and F. Blaabjerg, “A review of the state of the art
Jun. 2009.
of power electronics for wind turbines,” IEEE Trans. Power Electron.,
[24] M. Preindl and S. Bolognani, “Model predictive direct speed control
vol. 24, no. 8, pp. 1859–1875, Aug. 2009.
with finite control set of PMSM drive systems,” IEEE Trans. Power
[3] J. Dai, D. Xu, B. Wu, and N. Zargari, “Unified DC-link current
Electron., vol. 28, no. 2, pp. 1007–1015, Feb. 2013.
control for low-voltage ride-through in current-source-converter-based
[25] D. Quevedo, R. Aguilera, M. Perez, P. Cortes, and R. Lizana, “Model
wind energy conversion systems,” IEEE Trans. Power Electron., vol. 26,
predictive control of an AFE rectifier with dynamic references,” IEEE
no. 1, pp. 288–297, 2011.
[4] Z. Xu, R. Li, H. Zhu, D. Xu, and C. Zhang, “Control of parallel multiple Trans. Power Electron., vol. 27, no. 7, pp. 3128–3136, Jul. 2012.
converters for direct-drive permanent-magnet wind power generation [26] V. Yaramasu, M. Rivera, B. Wu, and J. Rodriguez, “Model predictive
systems,” IEEE Trans. Power Electron., vol. 27, no. 3, pp. 1259–1270, current control of two-level four-leg inverters – Part I: Concept, algo-
Mar. 2012. rithm and simulation analysis,” IEEE Trans. Power Electron., vol. 28,
[5] M. Liserre, R. Cardenas, M. Molinas, and J. Rodriguez, “Overview of no. 7, pp. 3459–3468, Jul. 2013.
multi-MW wind turbines and wind parks,” IEEE Trans. Ind. Electron., [27] M. Rivera, V. Yaramasu, J. Rodriguez, and B. Wu, “Model predictive
vol. 58, no. 4, pp. 1081–1095, Apr. 2011. current control of two-level four-leg inverters – Part II: Experimental
[6] J. Li, S. Bhattacharya, and A. Huang, “A new nine-level active NPC implementation and validation,” IEEE Trans. Power Electron., vol. 28,
(ANPC) converter for grid connection of large wind turbines for dis- no. 7, pp. 3469–3478, Jul. 2013.
tributed generation,” IEEE Trans. Power Electron., vol. 26, no. 3, pp. [28] M. Rivera, C. Rojas, J. Rodriguez, P. Wheeler, B. Wu, and J. Espinoza,
961–972, Mar. 2011. “Predictive current control with input filter resonance mitigation for a
[7] V. Yaramasu and B. Wu, “Three-level boost converter based medium direct matrix converter,” IEEE Trans. Power Electron., vol. 26, no. 10,
voltage megawatt PMSG wind energy conversion systems,” in IEEE pp. 2794–2803, Oct. 2011.
Energy Conversion Congr. and Expo. (ECCE), Sep. 2011, pp. 561–567. [29] R. Vargas, U. Ammann, B. Hudoffsky, J. Rodriguez, and P. Wheeler,
[8] A. Faulstich, J. Stinke, and F. Wittwer, “Medium voltage converter for “Predictive torque control of an induction machine fed by a matrix con-
permanent magnet wind power generators up to 5MW,” in IEEE Eur. verter with reactive input power control,” IEEE Trans. Power Electron.,
Conf. on Power Electron. and Appl. (EPE), 2005, pp. 9–P.9. vol. 25, no. 6, pp. 1426–1438, Jun. 2010.
[9] S. Kouro, M. Malinowski, K. Gopakumar, J. Pou, L. Franquelo, B. Wu, [30] S. Davari, D. Khaburi, and R. Kennel, “An improved FCS-MPC al-
J. Rodriguez, M. Perez, and J. Leon, “Recent advances and industrial ap- gorithm for an induction motor with an imposed optimized weighting
plications of multilevel converters,” IEEE Trans. Ind. Electron., vol. 57, factor,” IEEE Trans. Power Electron., vol. 27, no. 3, pp. 1540–1551,
no. 8, pp. 2553–2580, Aug. 2010. Mar. 2012.
[10] B. Wu, High-power converters and AC drives, 1st ed., ser. Wiley-IEEE [31] D. Zhi, L. Xu, and B. Williams, “Model-based predictive direct power
Press. John Wiley & Sons, Inc., 2006. control of doubly fed induction generators,” IEEE Trans. Power Elec-
[11] J. Suul, A. Luna, P. Rodriguez, and T. Undeland, “Virtual-flux-based tron., vol. 25, no. 2, pp. 341–351, Feb. 2010.
voltage-sensor-less power control for unbalanced grid conditions,” IEEE [32] M. Rivera, V. Yaramasu, A. Llor, J. Rodriguez, B. Wu, and M. Fadel,
Trans. Power Electron., vol. 27, no. 9, pp. 4071–4087, Sep. 2012. “Digital predictive current control of a three-phase four-leg inverter,”
[12] J. Hu and Z. Zhu, “Investigation on switching patterns of direct power IEEE Trans. Ind. Electron., vol. 60, no. 11, pp. 4903–4912, Nov. 2013.
control strategies for grid-connected DC-AC converters based on power [33] J. Rodriguez and P. Cortes, Predictive Control of Power Converters and
variation rates,” IEEE Trans. Power Electron., vol. 26, no. 12, pp. 3582– Electrical Drives, 1st ed. Chichester, UK: IEEE Wiley press, Mar.
3598, Dec. 2011. 2012.
[13] A. Bouafia, J. Gaubert, and F. Krim, “Predictive direct power control of [34] P. Cortes, J. Rodriguez, S. Alepuz, S. Busquets-Monge, and J. Bordonau,
three-phase pulsewidth modulation (PWM) rectifier using space-vector “Finite-states model predictive control of a four-level diode-clamped
modulation (SVM),” IEEE Trans. Power Electron., vol. 25, no. 1, pp. inverter,” in IEEE Power Electron. Spec. Conf. (PESC), Jun. 2008, pp.
228–236, Jan. 2010. 2203–2208, Rhodes, Greece.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

13

[35] A. Rockhill, M. Liserre, R. Teodorescu, and P. Rodriguez, “Grid-filter [54] P. Cortes, L. Vattuone, and J. Rodriguez, “A comparative study of
design for a multimegawatt medium-voltage voltage-source inverter,” predictive current control for three-phase voltage source inverters based
IEEE Trans. Ind. Electron., vol. 58, no. 4, pp. 1205–1217, Apr. 2011. on switching frequency and current error,” in IEEE Eur. Conf. on Power
[36] R. Vargas, P. Cortes, U. Ammann, J. Rodriguez, and J. Pontt, “Predictive Electron. and Appl. (EPE), Sep. 2011, pp. 1–8, Birmingham, UK.
control of a three-phase neutral-point-clamped inverter,” IEEE Trans.
Ind. Electron., vol. 54, no. 5, pp. 2697–2705, Oct. 2007. Venkata Yaramasu (S’08) received the B.Tech de-
[37] J. Rodriguez, M. P. Kazmierkowski, J. R. Espinoza, P. Zanchetta, gree in electrical and electronics engineering from
H. Abu-Rub, H. A. Young, and C. A. Rojas, “State of the art of finite Jawaharlal Nehru Technological University, Hyder-
control set model predictive control in power electronics,” IEEE Trans. abad, India, in 2005 and M.E. degree in electrical
Ind. Informat., vol. 9, no. 2, pp. 1003–1016, May. 2013. engineering with specialization in power electronics
[38] V. Yaramasu and B. Wu, “Predictive power control of grid-connected from S. G. S. Institute of Technology and Science,
four-level inverters in stationary reference frame,” in IEEE Int. Conf. on Indore, India, in 2008. He is currently working
Circuits, Power and Computing Tech. (ICCPCT), Mar. 2013, pp. 636– towards the Ph.D. degree in electrical engineering
641. at Ryerson University, Toronto, ON, Canada. His
current research interests includes renewable energy
[39] T.-S. Lee and J.-H. Liu, “Modeling and control of a three-phase four-
system, high power converters, and predictive con-
switch PWM voltage-source rectifier in d-q synchronous frame,” IEEE
trol. Yaramasu received six best student paper awards, two first prizes in
Trans. Power Electron., vol. 26, no. 9, pp. 2476–2489, Sep. 2011.
national level technical quiz competitions during his undergraduate studies
[40] J. Pou, R. Pindado, and D. Boroyevich, “Voltage-balance limits in four-
in India. During his Ph.D. studies at Ryerson University, he received the
level diode-clamped converters with passive front ends,” IEEE Trans.
Best Poster Award at the Natural Sciences and Engineering Research Council
Ind. Electron., vol. 52, no. 1, pp. 190–196, Feb. 2005.
of Canada – Wind Energy Strategic Network Annual Meeting-2010, Best
[41] J. Hu, L. Shang, Y. He, and Z. Zhu, “Direct active and reactive power Teaching Assistant Award from the Faculty of Engineering, Architecture and
regulation of grid-connected DC/AC converters using sliding mode Science in 2010, Student Research Awards from the Toronto Hydro and
control approach,” IEEE Trans. Power Electron., vol. 26, no. 1, pp. Connect Canada in 2010, 2011 and 2012, Research Excellence Awards from
210–222, Jan. 2011. the Electrical and Computer Engineering Department in 2012 and 2013.
[42] J. Rodriguez, B. Wu, M. Rivera, A. Wilson, V. Yaramasu, and C. Rojas,
“Model predictive control of three-phase four-leg neutral-point-clamped
inverters,” in Int. Power Electron. Conf. (IPEC), Jun. 2010, pp. 3112–
3116, Sapporo, Japan.
[43] V. Yaramasu, B. Wu, M. Rivera, and J. Rodriguez, “Predictive current Bin Wu (S’89-M’92-SM’99-F’2008) received his
control and DC-link capacitor voltages balancing for four-leg NPC Ph.D. degree in electrical and computer engineering
inverters,” in IEEE Int. Symp. on Ind. Electron. (ISIE), 2013, pp. 1– from the University of Toronto, Canada, in 1993.
6. After being with Rockwell Automation Canada as
[44] M. Saeedifard, “Space vector modulation of multi-level and multi- a Senior Engineer, he joined Ryerson University,
module converters for high power applications,” Ph.D. dissertation, Toronto, Canada, where he is currently a Profes-
University of Toronto, Toronto, ON, 2008. sor and NSERC/Rockwell Industrial Research Chair
[45] M. Popat, B. Wu, and N. Zargari, “A novel decoupled interconnecting in Power Electronics and Electric Drives. Dr. Wu
method for current-source converter-based offshore wind farms,” IEEE has published more than 280 technical papers, au-
Trans. Power Electron., vol. 27, no. 10, pp. 4224–4233, Oct. 2012. thored/coauthored two Wiley-IEEE Press books, and
[46] O. Kukrer, “Discrete-time current control of voltage-fed three-phase holds more than 20 issued/pending patents in the
PWM inverters,” IEEE Trans. Power Electron., vol. 11, no. 2, pp. 260– area of power conversion, advanced controls, adjustable-speed drives and
269, Mar. 1996. renewable energy systems. Dr. Wu received the Gold Medal of the Governor
[47] E.ON Netz Gmbh, “Grid code - high and extra high voltage,” Apr. 2006, General of Canada, the Premiers Research Excellence Award, Ryerson Dis-
available at: http://www.thewindpower.net. tinguished Scholar Award, Ryerson Research Chair Award and the NSERC
[48] M. Tsili and S. Papathanassiou, “A review of grid code technical Synergy Award for Innovation. He is a fellow of Engineering Institute of
requirements for wind farms,” IET Renew. Power Gener., vol. 3, no. 3, Canada (EIC) and Canadian Academy of Engineering (CAE).
pp. 308–332, Sep. 2009.
[49] 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
back-to-back NPC converter-based wind power systems,” IEEE Trans.
Ind. Electron., vol. 60, no. 5, pp. 1787–1796, 2013. Jin Chen (M’12) received his Ph.D. degree in
[50] P. Cortes, S. Kouro, B. La Rocca, R. Vargas, J. Rodriguez, J. Leon, energy and power engineering from Huazhong Uni-
S. Vazquez, and L. Franquelo, “Guidelines for weighting factors design versity of Science and Technology, China, in 2004.
in model predictive control of power converters and drives,” in IEEE Int. After being with Wuhan University of Technology,
Conf. on Ind. Tech. (ICIT), Feb. 2009, pp. 1–7, Gippsland, Australia. he joined Shanghai Second Polytechnic University,
[51] P. Cortes, J. Rodriguez, C. Silva, and A. Flores, “Delay compensation in China, in 2007, where he is currently a professor.
model predictive current control of a three-phase inverter,” IEEE Trans. As a Visiting Professor in the Department of Electri-
Ind. Electron., vol. 59, no. 2, pp. 1323–1325, Feb. 2012. cal and Computer Engineering, Ryerson University,
[52] P. Cortes, J. Rodriguez, D. Quevedo, and C. Silva, “Predictive current Canada, from August 2012 to August 2013, he
control strategy with imposed load current spectrum,” IEEE Trans. worked in the area of power system and renewable
Power Electron., vol. 23, no. 2, pp. 612–618, Mar. 2008. energies in the Laboratory for Electric Driver Ap-
[53] J. Rodriguez, J. Pontt, C. A. Silva, P. Correa, P. Lezana, P. Cortes, and plications and Research (LEDAR). Dr. Jin Chen has published more than 90
U. Ammann, “Predictive current control of a voltage source inverter,” research papers, coauthored a book, and holds 15 issued patents in the area
IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 495–503, Feb. 2007. of measurements and control systems.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].