You are on page 1of 9

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/292971572

Voltage Sensorless Direct Model Predictive Control of 3L-NPC Back-to-Back


Power Converter PMSG Wind Turbine Systems with Fast Dynamics

Conference Paper · January 2016

CITATIONS READS

10 205

4 authors, including:

Zhenbin Zhang Christoph Michael Hackl


Shandong University Munich University of Applied Sciences
93 PUBLICATIONS   648 CITATIONS    181 PUBLICATIONS   1,054 CITATIONS   

SEE PROFILE SEE PROFILE

Mohamed Abdelrahem
Technische Universität München
54 PUBLICATIONS   277 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Peformance enhanced direct model predictive control for power electronics and electric drives View project

Modeling and control of Electric Submersible Pumps (ESP) for geothermal power systems View project

All content following this page was uploaded by Christoph Michael Hackl on 04 February 2016.

The user has requested enhancement of the downloaded file.


Voltage Sensorless Direct Model Predictive Control
of 3L-NPC Back-to-Back Power Converter PMSG
Wind Turbine Systems with Fast Dynamics
Zhenbin Zhang Christoph Hackl Mohamed Abdelrahem, Ralph Kennel
Student Member, IEEE Member, IEEE Student Member, Senior Member, IEEE
Institute for Electrical Drive Systems Munich School of Engineering Research Group Institute for Electrical Drive Systems
and Power Electronics “Control of Renewable Energy Systems (CRES)” and Power Electronics
Technische Universität München (TUM) Technische Universität München (TUM) Technische Universität München (TUM)
Munich, Germany Munich, Germany Munich, Germany
Email: james.cheung@tum.de Email: christoph.hackl@tum.de Email: mohamed.abdelrahem, ralph.kennel@tum.de

Abstract—Full power rating three-level neutral-point clamped power rating will require lower switching frequencies to reduce
(3L-NPC) back-to-back power converters seem promising for wind switching losses and converter/inverter topologies with more than
turbine systems with high power. Direct Model Predictive Control two voltage levels to meet grid codes and to guarantee a low
(DMPC) is an attractive technique, in particular, for multi-level con-
verters. Voltage sensorless control makes the voltage measurements total harmonic distortion [4]. In particular, the three-level neutral-
obsolete. This work proposes and verifies an Initial Bias Com- point (diode) clamped (3L-NPC) back-to-back converter seems
pensation based Virtual Flux (IBC-VF) voltage sensorless DMPC promising. It allows for more than two voltage levels, but the
scheme for 3L-NPC back-to-back power converter Permanent- required amount of components is drastically less than e.g. for
magnet Synchronous Generator (PMSG) wind turbine systems. five-level topologies. Permanent-magnet Synchronous Generators
With the proposed VF estimation scheme, no grid side voltage
measurement information is required; both generator and grid side (PMSGs) show nice properties like high efficiency, compact size,
controllers are realized using DMPC concepts incorporating both and low maintenance, hence PMSGs are a promising choice for
targeting and (nonlinear) constraint set into a single cost function offshore wind turbines systems.
with great flexibility. Compared to conventional filter based voltage
sensorless control strategies, the proposed one shows fast dynamics A simplified direct-drive PMSG based grid-tied 3L-NPC back-
and accurate estimation of the voltage (virtual flux), which leads to-back wind turbine system is shown in Fig. 1. The back-to-back
to overshoot-free power/current and DC-link voltage control. converter consists of generator/machine-side converter (MSC)
Index Terms—Back-to-Back Power Converter, Three-level NPC, and grid/net-side converter (NSC) which share a common DC-
PMSG Wind Turbine Systems, Direct Model Predictive Control, link. Control objectives for MSC and NSC are (see e.g. [5]):
Novel Virtual Flux Estimation, Voltage Sensorless Control.
i. Torque control: The underlying torque controller in the
MSC must be fast and accurate to assure either (a) maximum
N OMENCLATURE
power point tracking (MPPT) of the wind turbine system or
Vdc , Idc DC-link voltage [V], current [A] (b) nominal torque generation for wind speeds above nominal
speed. In addition, to achieve high efficiency and to reduce
C, L, R Cap. [F], Induc. [H], resist. [Ω]
loads on the mechanical components, a low torque ripple and
Ts , k Sampling interval [s], instant [1] a low total harmonic distortion (THD) must be guaranteed;
~x, ~x> Vector, transpose of the vector ii. Power control: The NSC assures net/grid-side active and
~xabc , ~xαβ , ~xdq Vector in abc, αβ, dq reference frame reactive power control (active power is controlled indirectly by
X ∗ , x̂ Reference, estimated value of x the DC-link voltage controller). Moreover, the underlying NSC
current controller(s) must guarantee a certain power quality to
xm , xn machine, net (grid) side quantities
fulfill grid codes; iii. Voltage balancing: For 3L-NPC back-
x∗y Convolution of x and y to-back converters, (at least) one side needs to assure voltage
~ P, Q
S, Appar. [VA], active [W], react. power [var] balancing in upper and lower DC-link capacitor (see Fig. 1).
~u, G~ Switching state, IGBT gate signal vectors It is essential (and sufficient) for maintaining a constant DC-
−1
L {x} Inverse Laplace transform of x link voltage and to allow for low-voltage ride through (LVRT)
capabilities.
I. I NTRODUCTION Control schemes, which achieve the control objectives men-
Installations of wind turbine systems have steadily increased tioned above, can be divided into four groups (see [3], [6], [7]):
over the last years. Currently, 7.5 MW systems are available (i) cascaded control schemes: MSC with field-oriented control
in the market and numerous research activities aim at 10-12 (FOC) and NSC with voltage-oriented control (VOC) – both
MW levels for offshore applications [1]–[4]. This increase in the with modulator; (ii) direct control with look-up table: MSC with
Fig. 1. Simplified electrical circuit and controller of a three-level NPC back-to-back converter PMSG wind turbine system with choke ((R)L-filter)
on the net/grid side (subscript n).

direct torque control (DTC) and NSC with direct power control contributions of this work are the following: (i) System modeling
(DPC); (iii) direct control with modulator: MSC with DTC and of a three level NPC back-to-back PMSG wind turbine system
NSC with DPC – both with linear controllers and modulators, using DMPC control scheme is developed in detail; (ii) The novel
and (iv) model predictive control schemes for MSC and NSC. Initial Bias Compensation Virtual Flux (IBC-VF) estimation
Voltage sensor based schemes are costly and vulnerable to method is incorporated into a DMPC controller for three level
noise and may complicate the system setup [1]. All control NPC back-to-back power converter PMSG wind turbine systems.
schemes which rely on a voltage sensor depend on the measure- No modulation is required and multiple (nonlinear) control
ment accuracy provided by the sensor hardware. By using virtual targets and constraints are included; (iii) The proposed method
flux estimation schemes, several publications extended Voltage is compared with the two conventional band-pass filter-based
Oriented Control (VOC), and DPC (with look-up table) methods schemes and the comparison results are analyzed.
such that a voltage sensor is not required anymore (see e.g. [1]).
Most of these VF estimation schemes rely either on a band- II. S YSTEM D ESCRIPTION AND M ODELING
pass or high-pass filter to extract the related component(s) of the Now, a grid-tied direct-drive 3L-NPC back-to-back converter
estimated VF. However, due to the inevitable transient time (time PMSG wind turbine system is modeled. Note that, all quantities
delay) of the filters, a relatively long delay occurs at the beginning ~xαβ in the αβ coordinate system, and quantities ~xdq in the dq
and during the transient phases of the estimation, yielding coordinate system can be derived by the corresponding quantities
inaccurate control during these phases. E.g. a huge overshoot ~xabc in the abc coordinate system invoking (power invariant)
in the estimation error will be observed and the control system Clarke- and Park-Transformation, i.e. for angle φ ∈ R [rad] [3],
might suffer from limitation problems. In [8] a novel Initial Bias q h i h i
αβ −1 −1
, ~xdq xαβ
1
Compensator based VF scheme is firstly proposed for a grid ~x[k] = 23 0 √23 − √23 ~xabc = −cos (φ) sin (φ)

2 2
[k] [k] sin (φ) cos (φ) [k] .
tied two level back-to-back power converter within a constant | {z } | {z }
=:Tc (Clarke trans.) =:Tp (φ) (Park trans.)
switching frequency predictive controller. Instead of using any
(1)
band-pass filters, a novel time domain (discrete) model based bias
For implementation, all models are derived in discrete-time. So,
compensator is proposed, which achieves accurate estimation
for some continuous signal x(·) and time instant t ≥ 0, sampling
within one sampling interval. However, due to the linear nature
with a sufficiently small sampling time Ts ( 1 [s]) and sampling
of the deadbeat like controller, no nonlinear constraints (like
instant k = 0, 1, 2, . . . gives x(t) ≈ x(kTs ) =: x[k] . Moreover,
currents, torque, switching frequency limitations and DC-link d x −x
applying the forward Euler method yields dt x(t) ≈ [k] Ts[k−1] .
balancing requirements) can be added to the controller design.
This work extended the linear/modulator controller based IBC- A. Turbine power and aerodynamical torque
VF estimation method to a nonlinear DMPC concept for 3L-NPC
The mechanical power extracted by the wind turbine from the
back-to-back power converter PMSG wind turbine systems. The
wind is given by the turbine power (see [5] or in great detail [9]):
proposed method achieves fast virtual flux/voltage estimation
with good accuracy and a dynamic response within one sampling 3
∀ k ≥ 0 : Pt[k] = 0.5ρA Vw[k] Cp (λ[k] , β[k] )
step. Moreover, due to the presence of the nonlinear DMPC
Rt 3 3
controller, multiple constraints can be included easily into a = 0.5ρACp (λ[k] , β[k] )( ) × ωm[k] ≥ 0, (2)
single cost function without requiring a complex modulator. Main λ [k]
>
where ρ [kg/m3 ] is the air density, A [m2 ] is the rotor area, Rt where P̂[k] , Q̂[k] is obtained by using the estimated virtual
[m] is the blade radius, Cp (λ, β) ≤ Cp,Betz = 16/27 ≈ 0.56 is flux as given in Equation (33) (see Sec. IV).
the power coefficient of the wind turbine, λ = RVt ωwm is the tip
speed ratio – depending on wind speed Vw [m/s], rotor radius
Rt [m] and generator/machine speed ωm [rad/s] – and β [◦ ] is
the pitch angle. For simplicity, it is assumed that the turbine D. 3L-NPC back-to-back converter and DC-link
power is transformed without losses to mechanical power in the
generator (otherwise introduce an adequate efficiency factor).
Hence, Considering the 3L-NPC back-to-back converter depicted in
Fig. 1, the gate signal of the upper-most IGBT in the first leg
Pt[k] a1
∀ k ≥ 0: Tt[k] = ≥0 (see [5] or [9]), (3) (phase a) of the machine side converter is denoted by Gm . In
ωm[k] general, for x ∈ {m, n}, y ∈ {a, b, c} and i ∈ {1, 2}, the gate
where Tt [Nm] is the turbine (aerodynamical) torque. signal for the upper IGBTs is introduced as Gyi x and, for the
lower IGBTs, the negated gate signal as Ḡyi x (complementary to
B. Permanent-magnet synchronous generator (PMSG) Gyi
x , see Fig. 1). So the switching state ux
y

An isotropic PMSG in rotor flux orientation (indicated by 


if (P), i.e. Gy1 y2
1
 x = 1 ∧ Gx = 1
superscript dq) is given by [3]: y y1 y2
" # un := 0 if (0), i.e. Gx = 0 ∧ Gx = 1 (7)
Ts Rs

1 − T ω
 y1 y2
−1 if (N), i.e. Gx = 0 ∧ Gx = 0
s e[k]

~idq Ls ~idq
m[k+1] =


m[k]
−Ts ωe[k] 1 − TL s Rs



for each phase (y ∈ {a, b, c}) can be defined. The switching
s


| {z } 
state vector has the following form ~un = (uan , ubn , ucn )> . Hence,


=:Am[k] 

" # ! 
Ts 
for DC-link voltages Vc1 and Vc2 (see Fig. 1), the line-to-line
Ls 0 dq 0 
+ Ts ~ vm[k] + Ts ψpm (4) voltages ~v ltl (on machine or net side) are given by
0 L − Ls ωe[k]  x
| {z s }

| {z }   a b
  
v − v 1 −1 0

=:Bm =:Hm[k] 
 x x
  ltl |~
u |+~u
ωe[k+1] = ωe[k] + ΘmTsNp Tt[k] − Np ψpm iqm[k] 


 ~vx[k] (~ux ) := vxb − vxc  = Vc1  0 1 −1 x[k] 2 x[k]
c a
vx − vx −1 0 1

| {z }  
=:Te[k]  
1 −1 0
|~u |−~ u
where Rm [Ω] and Lm [Vs/A] are stator resistance and in- − Vc2  0 1 −1 x[k] 2 x[k] (8)
ductance, resp., ~im = (im , im ) [A] is the stator current
dq d q > 2
−1 0 1
dq d q >
vector, ~vm = (vm , vm ) [V]2 is the output voltage vector
where
of the machine-side converter (control input), Np [1] is the >
|~ux | := |uax |, |ubx |, |ucx | (9)
pole pair number, ωe = Np ωm [rad/s] is the electrical angular
frequency of the stator signals, ψpm [Vs] is the flux linkage of is the vector of the absolute values of the elements of ~un . Clearly,
the permanent-magnet, Θm [kgm2 ] is the (overall) inertia, and the line-to-line voltages depend on the switching vector ~un . The
Te [Nm] is the electro-magnetic (generator) torque. converter terminal voltage vector in αβ frame can be computed
by
C. Net-side filter, active and reactive power " #
1 1
0 −
q
A RL-filter connected to an ideal (balanced) grid in stator fixed dq
~vx[k] (~ux ) = Tp (φ) 23 2 √3 2
~vxltl (~ux[k] ), (10)
orientation (indicated by superscript αβ) is given by (see [10], 0 3 0
[11]):
| {z }
αβ
=~ vx (~
ux )
~iαβ Rn ~αβ Ts αβ
− ~eαβ
 
n[k+1] = 1 − Ts Ln in[k] + Ln ~ vn[k] n[k] (5) where φ is either the electrical angle of the rotor flux, i.e. φe =
R
where Rn [Ω] and Ln [Vs/A] are filter resistance and inductance, ωe (τ ) dτR , or the electrical angle of the grid voltage vector,
~αβ α β > 2
resp., in = (in , in ) [A] is the current vector to the grid, i.e. φn = ωn (τ ) dτ .
~vnαβ = (vnα , vnβ )> [V]2 is the output voltage vector of the net- To achieve voltage balance, the difference voltage Vo :=
αβ α β > 2
side converter (control input), and ~en = (en , en ) [V] is the V c1 −Vc2 shall be zero for all time and it can be controlled either
(symmetrical) grid voltage vector (see Fig. 1). by the net side or by the generator side power converter. In this
Net-side power vector (active and reactive powers) are given work, it is assumed that the DC-link voltage balancing is achieved
by (See [11], [12]) by the generator side converter control, but for the completeness
! " # ! of the model description, the DC-link voltage balancing model
α β α α
P [k+1] T s e e e − v will still be introduced in the following paragraphs.
~[k+1] =
S = n[k] n[k] n[k] n[k]
β α β β
Q[k+1] Ln en[k] −en[k] en[k] − vn[k] Known from Fig. 1, Vo depends on the charging state of the
two DC-link capacitors C1 = C2 = C and will only change
!
Ts Rn
P[k] − Ln P̂[k] − ωn Ts Q̂[k]
+ (6) when currents Ixo are drawn from it (see Fig. 1), i.e., when
Q̂[k] − TsLR n
n
Q̂ [k] + ωn T s P̂ [k] ~ux contain zero elements. For a given phase current vector
~iabc a b c >
x := (ix , ix , ix ) , the dynamics of Vo are given by [3], [12] 1) Machine-side torque control and voltage balancing: The
generator/machine-side cost function is defined by
dVo dVc1 dVc2 1  p n p n

= − = (Im[k] + Im[k] ) − (In[k] + In[k] )
dt dt dt C m
JDMPC (~um ) = γTe |Te∗ − Te (~um )| + γidm |id∗ d
m − im (~
um )|
1  >~abc >~abc
 | {z }
|~um[k] | im[k] − |~un[k] | in[k] (11)
| {z }
= =:JTe =:Jid
C m

where |~ux | is as in (9). Applying the forward Euler approximation + γVo |Vo∗ − Vo (~um )| (15)
| {z }
yields the discrete voltage difference equation JVo =:JCSm

Ts 
|~um[k] |>~iabc un[k] |> ~iabc (12) with weighting factors γTe [1/Nm], γidm [1/A] and γVo [1/V]. The

Vo[k+1] = Vo[k] − m[k] − |~ n[k] .
C target set is represented by the sum of the sub-costs JTe and Jidm
Considering the DC-link current Idc[k] of a back-to-back con- for torque control with reference Te∗ and d-axis current control to
verter (see Fig. 1), the DC-link dynamics can be modeled by guarantee a constant flux (i.e. id∗m = 0 A for maximum torque per
Idc[k]
ampere). The constraint set JVo = JCSm shall assure voltage
Ts z }| { balancing with Vo∗ = 0 V. The predicted torque of Te (~um ),
Vdc[k+1] = Vdc[k] + Im[k] − In[k] (13) current of idm (~um ) and voltage difference V0 (~um ) are given by
C
equations (4) and (12), respectively. Due to the limited space of
where In[k] = (in[k] ) · ~un[k] and Im[k] = (~iabc
~abc > abc >
m[k] ) · ~ uabc
m[k] are this paper, the MPPT and torque reference generation parts are
DC-link components of the net (grid) and machine side currents, not introduced. Detailed discussions can be found in [2], [4],
respectively. [9] and the references therein.
2) Net-side power control and voltage balancing: Here, the
III. DMPC FOR 3L-NPC BACK - TO - BACK CONVERTER PMSG
grid/net-side cost function is defined by
WIND TURBINE SYSTEMS

DMPC exploits the finite number of switching states of power


n
JDMPC (~un ) = γP |P ∗ − P (~un )| + γQ |Q∗ − Q(~un )|
| {z } | {z }
converters and combines current (torque) or power control and =:JP =:JQ
modulation into one computational step. Besides fast dynamics + γVo |Vo∗ − Vo (~un )| (16)
of the closed-loop system, multiple nonlinear constraints can be | {z }
JVo =:JCSn
included into the controller design by an appropriately designed
cost function. In particular, for multi-level converters with more with weighting factors γP [1/W], γQ [1/Var] and γVo [1/V]. The
than eight switching states, DMPC will increase performance target set is represented by the sum of the sub-costs JP and
and ease controller design. Classical DMPC schemes (shown in JQ for active and reactive power control, resp. The constraint
Fig. 2) [10], [13] evaluate a given cost function set JVo = JCSn shall also assure voltage balancing. Note that,
m
since JCSm = JCSn , both converters may contribute to voltage
X
∗ p balancing. The predicted active and reactive power are given by
JDMPC (~ux ) = γT Sx,i |T Sx,i − T Sx,i (~ux )|
Equation (6).
i=1
| {z
=:JT Sx
} An overview of the general control structure is given by the
n lower part of Fig.1, where the ‘VF Estimation’ part is introduced
in the following section. Note that, the DC-link controller is not
X p

+ γCSx,j |CSx,j − CSx,j (~ux )| (14)
j=1 the scope of this paper, and due to the limited space it will not
| {z } be covered. For those who are interested to this part please refer
=:JCSx
to [5] and the references therein.
for each switching state ~ux of some admissible (finite) set U.
The cost function consists of two parts: JT Sx and JCSx which
represent sub-costs for the Target Set T Sx – such as reference IV. G RID S IDE VOLTAGE S ENSORLESS C ONTROL BASED ON
tracking of current, torque, and/or power – and the Constraint V IRTUAL F LUX
Set CSx – such as voltage balancing and limitations. Target
and constraint set may each include several commands i and j
The virtual flux (VF) concept allows to replace AC voltage
weighted by the factors γT Sx,i and γCSx,j , respectively. Predicted
sensors by a VF estimator. This has advantages like hardware
target/constraint variable and corresponding reference are denoted
p ∗ p ∗ cost reduction and system simplification. By treating the net/grid
by T Sx,i and T Sx,i and CSx,j and CSx,j , respectively. The
side line choke (RL-filter) and the grid voltage as the stator
problem of an adequate selection of the weighting factors will
circuit of an AC machine (see Fig. 1), the virtual flux can be
not be addressed in this paper and an “equal weighting” solution
introduced as [8]
proposed in [10] is simply adopted due to its effectiveness. An Z t
illustrative discussion of weighting factor design can be found ~ α β
>
~e (0).
ψen (t) := ψen (t), ψen (t) := ~en (τ ) dτ − ψ n
in [14]. In the following, the cost function design for machine 0
side torque control and grid side power control, both with DC-
link balancing is introduced. (17)
For implementation, the VF can be obtained by In practice Rn is very small and can be considered as 0.
~key is obtained by adding Ln~in (t)
Therefore, an estimate for ψ
Z t
(5) d~in (τ ) ~ to (23), i.e.
~
ψen (t) = (~vˆnαβ (τ ) + Ln + in (τ ) · Rn ) dτ =
0 dτ
Z t Z t ~ˆkey (t) = L−1 {Fbp (s)} ∗ ψ
ψ ~e (t)−
~vˆnαβ (τ ) · dτ + ~in Rn (τ ) · dτ + Ln~in (t) − Ln~in (0) . n

0 0 | {z } L−1 {Fbp (s)} ∗ {Ln~in (t)} + Ln~in (t) (24)


| {z }
f
~
ψ | {z }
~v
ψ ≈0
(18)
2) FBP-VF applies the band-pass filter (22) to filter the
where ~vˆnαβ (t) is calculated via the command switching sequence integration of the (approximated) converter side voltage ~vˆn (t)
and the measured DC-link voltage using Equations (8) to (10). and the current ~in (t) through Ln and Rn . One obtains the
Assuming that the grid is unknown but balanced (ideal), i.e., the following estimate ψ~ˆkey (t) of the virtual flux
grid voltage has the following form
> ~ˆkey (t) =(20)(18)
ψ ~e (t) + Ln~in (0)}
= L−1 {Fbp (s)} ∗ {ψ n
~en (t) = A cos(ωn t + θ0 ), sin(ωn t + θ0 ) (19) | {z }
ˆ
~
=ψ bias
with constant amplitude A, initial phase θ0 and constant angular
~e (t)}, (25)
≈ L−1 {Fbp (s)} ∗ {ψ
frequency ωn . Now the virtual flux may be written as follows n

Z t Z t ! Therefore, both FBP-VF and PBP-VF approximate the desired


~ cos(ωn τ + θ0 )
ψen (t) = ~en (τ ) · dτ = A· · dτ filtered estimate as follows
0 0 sin(ωn τ + θ0 )
! ! ψ~ˆkey (t) := L−1 {Fbp (s)} ∗ ψ
~e (t). (26)
n
A sin(ωn t + θ0 ) A − sin(θ0 )
= · + · . (20)
ωn − cos(ωn t + θ0 ) ωn cos(θ0 ) Remark IV.1. Known from (24) and (25), during steady state
| {z } | {z } since ~in (t) in (24) is a pure sinusoidal signal with frequency
=:ψ~key (t)=(ψ α (t), ψ β (t))> =: ~bias =(ψ α , ψ β )>
ψ
key key bias bias close to ωn (neglecting measurement and switching noise),
~ ~
Clearly, grid side voltage of ~en (t) can be directly calculated Fbp (t)∗{Ln in (t)}−Ln in (t) ≈ 0 holds true and, so, (24) equals
from the ‘key’ part of the VF given by (25). However, during transient phase the PBP filter method
> with (24) is not accurate. Here the filter may loose information
d ~ (20)

β α of Ln~in (t) being not sinusoidal or having a different frequency
~en (t) = ψ e (t) = ω n −ψkey (t), ψkey (t) . (21)
dt n than ωn . In contrast to that, the FBP filter method in (25) is not
Once ψ ~key is known the grid side voltage can be also obtained. affected. But for initial phase, due to the inevitable transient
Known from equation (20), the ‘Bias’ part of the VF is simply time of the filter dynamics, for both filter methods, the estimation
a constant. To “filter out” ψ ~e (t) using a high-pass or band pass is suffering from an initial delay time (> 1/ωn ) yielding initial
n
filters in the frequency domain is straight-forward in concept. estimation errors. Any controller, using these delayed estimates,
will be affected and will output non-ideal actuating signals and
(large) overshoots are to be expected e.g. in the power output
A. Frequency-domain Band-Pass Filter Based VF Estimation (for a more detailed analysis please see [8]).

A more detailed analysis for using the band-pass filter with


the following transfer function (for both α and β component) B. Initial Bias Compensation Based VF Estimation
α β
ψ̂key (s) ψ̂key (s) Kωn s To overcome the issues of the filter based methods, in [8],
Fbp (s) = α
= β
= 2 . (22)
ψen (s) ψen (s) s + Kωn s + ωn2 a time domain Initial Bias Compensation based VF estimation
method (IBC-VF) for a grid connected two-level converters with
can be found in [15]–[17]. It can be further divided into two
modulator-based (linear) predictive controller is introduced. In
filter methods: the Partial Band-Pass filter (PBP) based and the
this work, the similar idea is extended to a three-level back-to-
Full Band-Pass filter (FBP) based methods (see Fig. 2 A and
back PMSG wind turbine system using a cost function based
B).
direct model predictive (nonlinear) control scheme without a
1) PBP-VF applies the band-pass filter (22) only to filter the modulator.
integration of the (approximated) converter side voltage ~vˆnαβ as
In the discrete time domain, one can introduce the sampling
in (10), i.e.
period Ts and the period Tn := ω2πn of the grid fundamental
Z t
(20)(18)
n voltage. Assuming that Tn is a multiple of Ts , i.e., Tn = N Ts
L−1 {Fbp (s)} ∗ ~vn (τ ) dτ = L−1 {Fbp (s)} ∗ ψ ~e (t)
n for some fixed natural number N ≥ 1 (sampling instant) and
0
Z t o rearranging equation (20) in the discrete time domain, the basic
− Rn · ~in (τ ) dτ − Ln~in (t) + Ln~in (0) . (23) idea of IBC-VF can be derived by
0 | {z }
ˆ
~
=ψ bias ∀t ≥ N : ~ˆkey[k] = ψ
ψ ~ˆbias[k]
~e [k] − ψ (27)
n
A. Partial Band-Pass Filter Based VF B. Full Band-Pass Filter Based VF C. Initial Bias Compensation Based VF
Fig. 2. Three virtual flux estimation methods.

where ψ ~e [k] can be easily computed by Equation (18). The Therefore the active and reactive power can be estimated by
n

computation of ψ ~ˆbias[k] , is the key part for a fast and accurate ! " # !
P̂[k] eα
n[k] eβn[k] iα
n[k]
estimation. Starting from the first sampling interval, i.e., k = 1, = β · β
~ˆbias[k] can be calculated in two steps (depending on k): Q̂[k] en[k] −eα n[k] in[k]
ψ " β # !
a) For k ≥ N : For any l ≥ 0, assuming(19) holds (true in most α
−ψ̂key[k] ψ̂key[k] iα
n[k]
applications), the following can be observed,: = ωn α β · β (33)
ψ̂key[k] ψ̂key[k] in[k]
−1+l
NX −1+l
NX −1+l
NX
~e [i] (20)
ψ = ~key[i] +
ψ ~bias = N · ψ
ψ ~bias . The proposed direct model predictive control with VF estimation
n
i=0+l i=0+l i=0+l
scheme is introduced and its general overview is shown in Fig. 1.
| {z }
=0 V. R ESULTS AND A NALYSIS
~bias can be estimated after one period
So, the constant bias term ψ In this section the control performances of the proposed IBC-
Tn (i.e. after N samples) of the grid voltage by VF with DMPC for a three level NPC back-to-back power
N −1 converter PMSG wind turbine system are illustrated by simu-
ˆ 1 X ~
~ ~
ψbias = ψbias[N−1] = · ψe [i] . (28) lations and compared with two conventional VF based DMPC
N i=0 n schemes, namely, PBP-VF, and FBP-VF based DMPC schemes.
A grid-tied 3L-NPC back-to-back PMSG wind turbine system
b) For 1 ≤ k ≤ N : From (20) it is known that α and β compo-
(as described in Sec. II) is implemented in Matlab/Simulink.
nent of the virtual flux are given by ψeαn (t) = ωAn · [sin(ωn t +
Simulation and system data is collected in Tab. I.
θ0 ) − sin θ0 ] and ψeβn (t) = ωAn · [− cos(ωn t + θ0 ) + cos θ0 ],
respectively. Considering the case ωn t → 0, one may rewrite A. Generator side/DC-link control performance comparison
the equations above as follows The generator side control performance for all the three
α
! schemes is illustrated in Fig. 3. It can be observed that during the
ψen (t) A sin(ωn t + θ0 ) − sin θ0
 
lim = lim = initial phases, i.e., t ∈ [0, 0.016]s, IBC-VF based DMPC scheme
ωn t→0 ψ β (t) ωn t→0 ωn − cos(ωn t + θ0 ) + cos θ0
en
  obviously outperforms both the two conventional schemes in
Aωn t sin(ωn t+θ 0 )−sin θ0
ωn t
Aωn t cos(θ0 ) terms of the DC-link voltage control. The reason can be found
lim − cos(ωn t+θ0 )+cos θ0 = lim
ωn t→0 ωn ωn t ωn t→0 ωn sin(θ0 ) from the grid side inaccurate voltage/power estimation, which
(29) are involved to fulfill the DC-link voltage control.
Now, for a small sampling time Ts  1s (Ts = 50µs in this
B. Net side/DC-link voltage balancing control performance
work) and ωn t → ωn Ts in (29), the virtual flux is already
comparison
estimated after one sampling interval by
DC-link balancing and grid side control performances are
> > (29)
shown in Fig. 4. Both the DC-link balancing and grid side
 
α β α β
ψ̂en [1] , ψ̂en [1] = lim ψ̂en (t), ψ̂en (t) ≈
ωn t→ωn Ts power and current are affected by the use of PBP- and FPB-
 > VF based DMPC schemes (see Fig. 4 (a) and (b)). On the
ATs cos(θ0 ), sin(θ0 ) . (30)
contrary, the IBC-VF base DMPC yields fast dynamics for the
Hence, for the initial estimation phase, the following estimate DC-link voltage balancing and active and reactive power control
of the bias component can be used: (especially during the initial and transient phase, see Fig. 4
> (a), (b) and (c) for t ∈ [0, 0.16]s and t ∈ [0.07, 0.08]s). The
ˆ ˆ (20),(30)
 β
~ ~ ψ̂e [1] ψ̂eαn [1]
ψbias[k] = ψbias[1] = − Ts ·ωn , Ts ·ωn
n . (31) (huge) over- and undershoot of the conventional methods may
potentially lead to system damage if the system protection is
The overall structure of IBC-VF is illustrated in Fig. 2C. not well designed.
After obtaining ψ ~ˆKey the estimate ~eˆn[k] can be calculated by
invoking (21), i.e. in discrete time as: C. Estimation performance comparison
h i> In Fig. 5, the estimation performances of the proposed IBC-
~eˆn[k] = ωn −ψ̂key[k] β α
ψ̂key[k] . (32) VF based method and the conventional filter based methods
are compared. For all system variables (including grid side
Parameter [unit] Value Parameter [unit] Value
Air density ρ [kg/m3 ] 1.225 DC-Link cap. C1 = C2 = C [F] 1000 × 10−12
Turbine radius Rt [m] 0.8 PMSG stator inductance Ld =Lq =Ls [Wb] 19 × 10−3
Maximum power coefficient Cp [1] 0.48 PMSG stator resistance Rs [Ω] 1.3
Tip speed ratio λ [1] 8.4 PMSG flux linkage ψpm [Vs] 4.26 × 10−2
Pitch angle β [◦ ] 0 Sampling frequency fs = 1/Ts [kHz] 20
1
Grid-side phase voltages k~en k [V] 250 γTe [1] Np ψpm
Grid-side voltage frequency ωn [rad/s] 100π γidm [1] 1
Grid-side resistance Rn [Ω] 1.56 × 10−3 γVo [1] 0.001
Grid-side inductance Ln [Wb] 16 × 10−3 γP [1] = γQ [1] 1

TABLE I. Implementation and system data.

1 1 1
iαmβ [p.u.]

iαmβ [p.u.]

iαmβ [p.u.]
0 0 0

−1 iαm iβm −1 iαm iβm −1 iαm iβm

1 Te∗ Te 1 Te∗ Te 1 Te∗ Te


Te [p.u.]

Te [p.u.]

Te [p.u.]
0 0 0

−1 −1 −1

Vdc∗ Vdc 1.1 Vdc∗ Vdc


1.02 1.02
Vdc [p.u.]

Vdc [p.u.]

Vdc [p.u.]
1
0.98 1
1
0.96 Vdc∗ Vdc
0.94 0.98
0 0.05 0.1 0 0.05 0.1 0 0.05 0.1
Time [s] Time [s] Time [s]
(a) PBP based DMPC control scheme. (b) FBP based DMPC control scheme. (c) IBC-VF based DMPC control scheme.
Fig. 3. Performance comparison for Generator side/DC-link control. For all three sub-figures, from up to down are: generator stator current,
generator torque, and DC-link voltage, respectively.

active/reactive power, virtual flux, and grid side voltage), very


performance of the proposed IBF-VF DMPC strategy is com-
acceptable performances can observed for the proposed IBC-VF pared with two conventional band-pass filter based VF methods.
based DMPC scheme. All performed simulation scenarios highlight the effectiveness
VI. C ONCLUSION and the performance improvement of the proposed IBC-VF
based DMPC control scheme compared to the classical methods.
In this work, a voltage sensorless direct model predictive Further work will focus on the experimental verification of the
control method for three-level back-to-back PMSG wind turbine proposed method and its extension to unbalanced grids.
systems with Initial Bias Compensation Virtual Flux (IBC- Acknowledgment: This work is supported by DFG founding
VF) estimation is proposed and illustrated by simulations. The (No.: KE817/32-1). The first author (who is the corresponding
following have been realized: 1) A three-level NPC back-to- author of this work) would like to express his gratefulness to
back converter PMSG wind turbine system is systematically Chinese Scholarship Council (CSC) for the scholarship support.
modeled and the required discrete-time equations for direct
model predictive direct torque/power control are developed in R EFERENCES
detail. 2) With the proposed control scheme, no modulator is [1] P. Antoniewicz and M. P. Kazmierkowski, “Virtual-Flux-Based Predictive
required and both the targeting and nonlinear constraint set are Direct Power Control of AC/DC Converters With Online Inductance
Estimation,” Industrial Electronics, IEEE Transactions on, vol. 55, no. 12,
directly included into the cost functions. 3) A time domain pp. 4381–4390, 2008.
IBC-VF virtual flux estimation scheme with fast dynamics and [2] M. Liserre, R. Cardenas, M. Molinas, and J. Rodriguez, “Overview of
good transient accuracy is introduced and included into the Multi-MW Wind Turbines and Wind Parks,” Ind. Electron. IEEE Trans.,
vol. 58, no. 4, pp. 1081–1095, 2011.
direct model predictive control controller realizing an overall [3] Z. Zhang and R. Kennel, “Direct Model Predictive Control of Three-Level
voltage sensorless control method. 4) The effectiveness and NPC Back-to-Back Power Converter PMSG Wind Turbine Systems Under
1.5 P∗ P Q∗ Q 2 P∗ P Q∗ Q 0.5 P∗ P Q∗ Q
P /Q [p.u.]

P /Q [p.u.]

P /Q [p.u.]
1 1
0.5 0
0 0
−0.5 −1 −0.5
1 1 1 ean i an
ean /ian [p.u.]

ean /ian [p.u.]

ean /ian [p.u.]


0
0
−1 0
−1
−2
−3 ean i an −2 ean i an
−1
1 1
Vdc1 Vdc2 Vdc1 Vdc2
Vdc12 [p.u.]

Vdc12 [p.u.]

Vdc12 [p.u.]
1
0.9 0.9
0.9
0.8 0.8
Vdc1 Vdc2 0.8
0 0.05 0.1 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Time [s] Time [s] Time [s]

(a) PBP based DMPC control scheme. (b) FBP based DMPC control scheme. (c) IBC-VF based DMPC control scheme.
Fig. 4. Performance comparison for Net side/DC-link voltage balancing control. For all three sub-figures, from up to down are: grid side
active/reactive power, grid side phase current v.s. phase voltage and DC-link capacitor voltages, respectively
2
1 P̂ P Q̂ Q P̂ P Q̂ Q 0.5 P̂ P Q̂ Q
P /Q [p.u.]

P /Q [p.u.]

P /Q [p.u.]
0.5 1
0
0 0
−0.5 −1 −0.5
1 1 1
key [p.u.]

key [p.u.]

key [p.u.]
0 0 0
α β α β α β
ψ̂key ψ̂key ψ̂key ψ̂key ψ̂key ψ̂key
ψ

ψ
ˆ~

ˆ~

ˆ~
−1 −1 −1
1 1 1
~ˆen [p.u.]

~ˆen [p.u.]

~ˆen [p.u.]
0 ê αn eαn êβn eβn 0 ê αn eαn êβn eβn 0 ê αn eαn êβn eβn

−1 −1 −1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Time [s] Time [s] Time [s]

(a) PBP based DMPC control scheme. (b) FBP based DMPC control scheme. (c) FBP-VF based DMPC control scheme.
Fig. 5. Performance comparison for state estimation. For all three sub-figures, from up to down are: real and estimated active and reactive power,
estimated virtual flux, and real and estimated grid side voltage, respectively.

Unbalanced Grid,” in Predictive Control of Electrical Drives and Power [11] Z. Zhang, H. Xu, M. Xue, Z. Chen, T. Sun, R. Kennel, and C. Hackl,
Electronics (PRECEDE 2015), Valparaiso, Chile., 2015. “Predictive Control with Novel Virtual FluxEstimation for Back-to-Back
[4] F. Blaabjerg, Y. Yang, and K. Ma, “Power electronics – Key technology Power Converters,” Industrial Electronics, IEEE Transactions on, vol. PP,
for renewable energy systems – Status and future,” in Proceedings of the no. 99, p. 1, 2014.
3rd International Conference on Electric Power and Energy Conversion [12] Z. Zhang and R. Kennel, “Novel Ripple Reduced Direct Model Predictive
Systems, 2013, pp. 1–6. Control of Three-level NPC Active Front End With Reduced Computational
[5] Z. Zhang, C. Hackl, F. Wang, Z. Chen, and R. Kennel, “Encoderless model Effort,” in Predictive Control of Electrical Drives and Power Electronics
predictive control of back-to-back converter direct-drive permanent-magnet (PRECEDE 2015), Valparaiso, Chile., 2015.
synchronous generator wind turbine systems,” in Proceedings of 15th [13] ——, “Fpga based direct model predictive power and current control of 3l
European Conference on Power Electronics and Applications, 2013, pp. npc active front ends,” in PCIM - 2016, Nurnburg, May 2016.
1–10. [14] C. M. Hackl, F. Larcher, A. Dötlinger, and R. M. Kennel, “Is multiple-
[6] T. Geyer, “A comparison of control and modulation schemes for medium- objective model-predictive control “optimal”?” in Proceedings of the 2013
voltage drives: Emerging predictive control concepts versus Field Oriented IEEE International Symposium on Sensorless Control for Electrical Drives
Control,” in Energy Conversion Congress and Exposition (ECCE), 2010 and Predictive Control of Electrical Drives and Power Electronics, 2013,
IEEE, 2010, pp. 2836–2843. pp. 1–8.
[7] Z. Zhang and R. Kennel, “Fully fpga based direct model predictive [15] J. A. Suul, A. Luna, P. Rodriguez, T. Undeland, and P. Rodriguez, “Virtual-
power control for grid-tied afes with improved performance,” in Industrial Flux-Based Voltage-Sensor-Less Power Control for Unbalanced Grid
Electronics Society, IECON 2015 - 41th Annual Conference of the IEEE, Conditions,” Power Electronics, IEEE Transactions on, vol. 27, no. 9,
Nov 2015. pp. 4071–4087, 2012.
[8] Z. Zhang, H. Xu, M. Xue, Z. Chen, T. Sun, R. Kennel, and C. Hackl, [16] J. Gonzalez Norniella, J. M. Cano, G. A. Orcajo, C. H. Rojas, J. F.
“Predictive control with novel virtual-flux estimation for back-to-back power Pedrayes, M. F. Cabanas, and M. G. Melero, “Improving the Dynamics of
converters,” Industrial Electronics, IEEE Transactions on, vol. 62, no. 5, Virtual-Flux-Based Control of Three-Phase Active Rectifiers,” Industrial
pp. 2823–2834, May 2015. Electronics, IEEE Transactions on, vol. 61, no. 1, pp. 177–187, 2014.
[9] C. Dirscherl, C. Hackl, and K. Schechner, “Modellierung und Regelung [17] Y. Han, G. Tan, H. Li, and X. Wu, “Study of multi-level rectifier in high
von modernen Windkraftanlagen: Eine Einführung (available at the authors power system based on a novel virtual flux observer,” in Power Electron.
upon request),” in Elektrische Antriebe – Regelung von Antriebssystemen, Drive Syst. 2009. PEDS 2009. Int. Conf., 2009, pp. 989–992.
D. Schröder, Ed. Springer-Verlag, 2015, ch. 24, pp. 1540–1614.
[10] Z. Zhang, F. Wang, T. Sun, J. Rodriguez, and R. Kennel, “FPGA Based
Experimental Investigation of a Quasi-Centralized Model Predictive Control
for Back-to-Back Converters,” Power Electron. IEEE Trans., vol. PP, no. pp,
p. 99, 2015,to be published.

View publication stats

You might also like