Professional Documents
Culture Documents
Clase14-Expo-Wind Generator Power Model
Clase14-Expo-Wind Generator Power Model
4, DECEMBER 2014
flux measurement or estimation [4]–[6]. On the other hand, the accurate average modeling approach was presented, while a
slow outer-loop controllers provide the current references for stability analysis was provided for the fifth-order model of the
the fast inner-loop controllers of both the RSC and the GSC wound rotor induction generator (WRIG) that unfortunately is
devices. At the RSC control, the two controlled inputs often valid for a constant input vector limited by unknown bounds.
use either the rotor angular speed for maximum power point Finally, in [16], [22], and [23], the control of both the RSC
tracking (MPPT) operation or the active power generation and and the GSC is considered but either no stability analysis is
the stator reactive power generation [4]–[6], [8], respectively. provided [16] or linearization techniques are used [22], [23]
At the GSC, the other two controlled inputs are used to keep with complicated controllers based on pole placement that
the dc-link voltage constant and to auxiliary support either the requires the knowledge of almost all the system parameters.
total reactive power generation of the DFIG or a local ac grid In this paper, a new model-based direct power PI controller
voltage [4]–[6], [8]. design is developed and analyzed for DFIG wind power
Direct power control (DPC) of DFIGs has been intro- systems. A suitable complete eighth-order nonlinear model
duced as an alternative to the conventional vector control [4]. of the DFIG is extracted based on the conventional current-
The main advantage of DPC is the avoidance of the cascaded flux model of the WRIG and the current model of the GSC.
controller structure, since the inner current control loops are This model is developed on the synchronously rotating dq
not required, while only single active and reactive power loops reference frame using a common voltage-oriented control
are used. As a result, only measurements of the appropriate (VOC) alignment for both the rotor and grid sides. Active
active and reactive power components are required. Other and reactive powers of both the induction generator stator and
advantages of DPC over vector control are referred to fast the GSC output connected to the grid are directly appeared
dynamic response, robustness to parameter variation, and grid in the proposed model state vector. These innovations make
disturbances [9]. possible the individual control of each power state (active and
Especially, DPC of simple GSCs was introduced in [10] reactive powers on stator or grid sides) to be implemented
and hysteresis control was used. The virtual-flux concept by simple PI output feedback controllers. The proposed DPC
was utilized in [11] with hysteresis controllers and in [12] schemes are independent from the system parameters, while
with proportional (P) controllers and decoupling terms that their application does not require the introduction of decou-
require the knowledge of system parameters, while in both pling networks, cancelation of nonlinear terms and the mea-
cases the grid filter resistances were neglected; in this case, surement or estimation of any extra states—such as fluxes—
the virtual-flux magnitude had to be estimated for reference or noncontrolled variables. Furthermore, the introduced PI
frame orientation. None of the above methods provided a controllers allow for a rigorous nonlinear stability analysis of
stability analysis of the system studied. In [13], a model- the entire 12th-order closed-loop system, based on advanced
based DPC for GSCs was introduced that uses P controllers Lyapunov methods recently developed in [24] and [25]. Unlike
but neglects the grid-filter resistance and requires grid-filter the authors’ published work for the RSC DPC [20], [21] and
reactance estimation through an adaptive law to prove stability the GSC DPC [14], in this paper, the complete DFIG system
of the closed-loop system. In addition, in [14], P controllers is studied and, to the best of the authors’ knowledge, it is the
with an appropriate extra term were introduced and stability first time that input-to-state stability (ISS) and in consequence
was proven for the GSC closed-loop system, while stability convergence to the equilibrium is concluded. Moreover, extra
was not proven when PI controllers were introduced to achieve PI controllers are introduced and analyzed for MPPT operation
zero steady-state error. and dc-link voltage regulation. Finally, simulations are carried
DPC of DFIGs was introduced in [7] and hysteresis con- out to verify the effectiveness of the overall modeling and
trollers were used, while stator-flux estimation was used and controller design in the face of reactive power reference signal
the knowledge of the stator resistance was required. In [15], or wind speed variations.
P controllers with decoupling terms were used that again
require the knowledge of system parameters, while stator flux II. S YSTEM M ODELING
estimation and orientation were required. In both [16] and [17], A. Wind Turbine Modeling
the RSC is considered and although the system is of fifth order,
only the two power components were used in the analysis. At the wind turbine mechanical rotating part of the system,
Many DPC methods neglect the small stator resistance while the mechanical power captured from the wind by the turbine
the stator flux is considered constant [18], to avoid variations is given by [26]
in the stator flux and coupling between the stator power 1
Pm = ρπ RT2 C p (λ, β) v3wind (1)
components. Only when the stator resistance is omitted, the 2
resistive voltage drop on the stator windings has no effect where ρ is the air density, RT is the radius of the turbine
on the stator flux magnitude; however, this holds true for blades, vwind is the wind speed, and C p (λ, β) is the power
almost constant active power and small variations of the stator coefficient of the wind turbine given by
reactive power [18]. A sliding-mode control method applied
only on the RSC was presented in [19]; this paper includes a 116 − 12.5
C p (λ, β) = 0.22 − 0.4β − 5 e λi
stability analysis with only the two stator power components λi
considered and the stator flux required to be estimated. In 1 1 0.035
= − 3 (2)
[20] and [21], only the RSC control was considered and an λi λ + 0.08β β +1
766 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 4, DECEMBER 2014
with β being the blade pitch angle and λ the tip-speed ratio rotor resistances, and L s and L r are the stator and rotor induc-
defined as tances while L m represents the mutual inductance. Parameter
ωrot RT ωs is the grid angular frequency, while p is the number of
λ= . (3) pole pairs. In addition, Udgrid and Uqgrid represent the stator
vwind
voltage components, which in this paper coincide with the
Due to the existence of a gearbox with gear ratio n g , the
grid voltage components, while Vdr and Vqr indicate the rotor
rotational speed of the wind turbine ωrot is related to the rotor
voltage components. Moreover, J represents the total inertia
speed ωr as
at the rotor of the WRIG and b is the friction factor, while
ωr = n g ωrot (4) Te represents the electromagnetic torque at the machine rotor.
Due to the adopted motor sign convention in the DFIG
while the torque at the generator’s rotor is given by
modeling, the mechanical torque is fed with a negative sign
Pm in the rotor dynamics equation for operation as generator.
Tm = . (5)
ωr The dot over a symbol stands for the time derivative of the
The power coefficient expression (2) has been shown to have corresponding variable.
a unique maximum, C p , for λopt ∼
opt In addition, the dynamic model of the GSC with states and
= 6.325 and βopt = 0 [26].
For the optimal tip-speed ratio λopt , (3) and (4) provide the inputs in the dq synchronously rotating frame is written as [6]
optimal rotational speed of the generator as L f I˙d f = −R f Id f + ωs L f Iq f − Vd f + Udgrid
opt λopt n g L f I˙q f = −ωs L f Id f − R f Iq f − Vq f + Uqgrid (11)
ωr = vwind (6)
RT
where Id f and Iq f are the GSC current d- and q-axis compo-
which corresponds to the maximum wind power extrac- nents, R f and L f are the grid filter resistance and inductance,
tion Pmmax . Then, the optimum torque at the generator’s rotor while Vd f and Vq f are the voltage components at the ac output
can be determined as of the GSC.
opt
opt 1 ρπ RT5 C p opt The dynamic model of the dc link with states and inputs in
Tm = ωr . (7) the dq synchronously rotating frame is written as [23]
2 λ3opt
3
From (6) and (7), it is evident that forcing the wind turbine C Vdc V̇dc = Vd f Id f + Vq f Iq f −Vdc Idc,RSC
2
system to rotate with the optimal rotational speed, the optimum 3 3
torque at the generator’s rotor acts as the external input of ≡ Vd f Id f +Vq f Iq f − Vdr Idr +Vqr Iqr (12)
2 2
the mechanical part of the system. This is accomplished by
where Vdc represents the dc-link voltage, C is the dc-link
applying MPPT control techniques on the power electronic
capacitance, and Idc,RSC the RSC dc current.
devices, i.e., by determining in a suitable speed controller
opt Thus, the complete system dynamic model is given by
the rotor speed reference ωrref to be equal to ωr as given
(8)–(12) as an eighth-order system with variables the stator
by (6).
and rotor currents and fluxes, the GSC currents, and the
dc-link voltage.
B. DFIG System Modeling in the dq Reference Frame To proceed with the analysis and the proposed DPC design,
At the electromechanical generator system, the dynamic the initial model (8)–(12) is transformed into a state-space
grid grid
model of WRIG with states and inputs in the dq synchronously form with x = [Vdc Ps PGSC Q s Q GSC ωr λds λqs ]T state
rotating frame can be written (using motor sign convention) vector, where Ps and Q s represent the stator active and reactive
as [27] grid grid
power components, while PGSC and Q GSC represent the active
Udgrid = λ̇ds − ωs λqs + Rs Ids and reactive powers exchanged between the GSC and the grid,
grid grid
respectively. In this way, as Ps , Q s , PGSC , and Q GSC are
Uqgrid = λ̇qs + ωs λds + Rs Iqs
system states, then it becomes evident that a DPC scheme
Vdr = λ̇dr − (ωs − pωr )λqr + Rr Idr could be easily implemented by simple output feedback
Vqr = λ̇qr + (ωs − pωr )λdr + Rr Iqr PI controllers, as it is explained in the following section.
J ω̇r + bωr = Te − Tm (8) Hence, to proceed with the model transformation, we recall
grid grid
the well-known expressions for Ps , Q s and PGSC , Q GSC as
with
given by
3 Lm
Te = p (λqs Idr − λds Iqr ) (9) 3
2 Ls Ps = (Udgrid Ids + Uqgrid Iqs )
2
and 3
Q s = (Uqgrid Ids − Udgrid Iqs ) (13)
λds = L s Ids + L m Idr λdr = L r Idr + L m Ids 2
and
λqs = L s Iqs + L m Iqr λqr = L r Iqr + L m Iqs (10)
grid 3
where Ids , Iqs and λds , λqs are the stator current and flux PGSC = (Udgrid Id f + Uqgrid Iq f )
2
d- and q-axis components, Idr , Iqr and λdr , λqr are the rotor grid 3
Q GSC = (Uqgrid Id f − Udgrid Iq f ). (14)
current and flux components, Rs and Rr are the stator and 2
BOURDOULIS AND ALEXANDRIDIS: DPC OF DFIG WIND SYSTEMS 767
The signs used in (13) and (14) indicate power flow toward Hence, by regulating individually the stator and GSC power
the machine and toward the GSC device. Also note that components, the total active and reactive powers generated by
in (13) and (14), Udgrid and Uqgrid are constant in the dq the DFIG are regulated since it is
synchronously rotating frame. grid
Then, the resulting system takes the form of an open-loop
tot
PDFIG = Ps + PGSC
grid
nonlinear system as follows: Q tot
DFIG = Q s + Q GSC . (19)
ẋ f 1 (x 1 , x 2 , m(u))
ẋ = 1 = (15)
ẋ 2 f 2 (x 2 , u, d) B. Closed-Loop System
Controllers (17) and (18) are developed on a common
where vector [x 1 x 2 ]T results from the separation of vector x
grid grid grid VOC alignment in accordance with Assumption 3.1.
into x 1 = Vdc and x 2 = [Ps PGSC Q s Q GSC ωr λds λqs ]T ,
Incorporating these controllers into the open-loop system, the
while f1 (·) and f 2 (·) are given in Appendix A. It is noted that
whole model is described in VOC alignment instead of the
u stands for the controlled input, i.e., u = [Vdr Vqr Vd f Vq f ]T ,
standard used field orientation. This constitutes a significant
and d stands for the uncontrolled open-loop inputs
innovation since no need for flux measurement or estimation is
2
T required. However, in this case, stability has to be analytically
3 U grid 1
d = α1 L r Ugrid2
Tm . (16) proven since the well-known advantage of the field-oriented
2 Lf J technique on proving stability is omitted. To this end, we
It is significant to observe that (15) appears to have a proceed with the entire system and controller closed-loop
cascaded form since the dc-link dynamics x 1 depend from the representation in the following form that permits a rigorous
rest system states but none of the last seven states x 2 of (15) stability analysis in voltage orientation:
depend on the dc-link voltage. V̇dc = α̃m
T
m (Vdc , u)
2
III. C ONTROLLER D ESIGN AND C LOSED -L OOP Ṗs = − α1 (Rs L r + Rr L s )+α1 L m Ugrid k P Pr Ps
3
S YSTEM A NALYSIS
− ωslip Q s − pα1 L r Ugrid λds ωr + α1 Rr Ugrid λqs
A. Direct Active and Reactive Power Controllers’ Design
In our study, the VOC alignment is adopted for both the − α1 αqr z qr
RSC and the GSC; therefore, without loss of generality, grid Rf 3 Ugrid grid grid
alignment on Uqgrid [21] is considered. ṖGSC = − + k P P f PGSC − ωs Q GSC − αq f z q f
Lf 2 Lf
Assumption 3.1: VOC alignment conditions:
Udgrid = 0 and Uqgrid = Ugrid = constant. 2
Q̇ s = ωslip Ps − α1 (Rs L r + Rr L s )+α1 L m Ugrid k P Qr Q s
As can be seen from the open-loop model equations, 3
through this voltage orientation, each of the controlled inputs, + pα1 L r Ugrid λqs ωr + α1 Rr Ugrid λds − α1 αdr z dr
grid
i.e., Vqr , Vq f , Vdr , and Vd f corresponds to Ps , PGSC , Q s , and
grid grid grid Rf 3 Ugrid grid
Q GSC , respectively. As a result, a simple output feedback PI Q̇ GSC = ωs PGSC − + k P Q f Q GSC − αd f z d f
controller can be used in each controlled input to separately Lf 2 Lf
regulate each power state at the desired reference. Therefore, b 1
the following output feedback PI controllers are proposed for ω̇r = pα2 L r Ugrid λds Ps − pα2 L r Ugrid λqs Q s − ωr − Tm
J J
the RSC:
t λ̇ds = −α3 Rr Ugrid Q s + ωs λqs
Vdr = k P Qr Q s + k I Qr Q s − Q ref
s dτ λ̇qs = −α3 Rr Ugrid Ps − ωs λds
0t
Lr ż dr = αdr Q s − Q ref
s
Vqr = k P Pr Ps + k I Pr Ps − Psref dτ + Ugrid (17)
0 L m ż qr = αqr Ps − Psref
while for the GSC, the proposed output feedback PI controllers
grid grid,ref
are ż d f = αd f Q GSC − Q GSC
t
grid grid,ref
grid
Vd f = k P Q f Q GSC + k I Q f
grid grid,ref
Q GSC − Q GSC dτ ż q f = αq f PGSC − PGSC (20)
0
t
grid grid grid,ref with z = [z dr z qr z d f z q f ]T representing the controllers’
Vq f = k P P f PGSC + k I P f PGSC − PGSC dτ + Ugrid
0 state vector while, instead of (17) and (18), the controllers’
(18) descriptions are equivalently given in state space as follows:
αdr
where the gains k P Qr , k I Qr , k P Pr , k I Pr and k P Q f , k I Q f , Vdr = k P Qr Q s + z dr
k P P f , k I P f are positive scalars. The last terms on the right- L m Ugrid
hand part of Vqr and Vq f are added for stability reasons with αqr Lr
Vqr = k P Pr Ps + z qr + Ugrid
the value of Ugrid constant and known. L m Ugrid Lm
768 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 4, DECEMBER 2014
Using storage function H for the 11th-order unforced sys- Now, coming back to the forced original 11th-order closed-
tem and making use of Remark 3.2 and [28, Th. 8.4], one can loop system (21b) and (21c), it is noted that this is globally
easily conclude that states x d → 0. Lipschitz in u cl , since it is linearly dependent on the external
Considering Proposition 3.2 and Remark 3.1, this implies input vector, and therefore satisfies the following inequality:
that also the continuously differentiable right-hand side of f˜2 (x 2 , z, u cl1 ) f˜2 (x 2 , z, u cl2 )
≤ L u cl1 − u cl2
(26) is bounded and as a result ẋ du = [ẋ d ẋ u ]T is bounded. f˜3 (x 2 , u cl1 ) − f˜3 (x 2 , u cl2 )
Moreover, ẍ du is bounded since through (26) it can be easily
(28)
shown that it is equal to a sum of terms that contain either
linearly the bounded derivatives ẋ du or their products with for some Lipschitz positive constant L. Moreover, the deriv-
their bounded states x du . As a result, it is proven that ẋ du is ative of H̃ with respect to the last 11 equations of (20) or
uniformly continuous1 [29, pp. 125–126]. equivalently (21) is calculated as
Uniform continuity of ẋ d clearly ensures that x d = 0 also
∂ H̃ f˜2 (x 2 , z, 0)
implies that ẋ d = 0. Under these conditions, AS to the origin H̃˙ =
for the 11th-order unforced closed-loop system is concluded ∂x f˜3 (x 2 , 0)
since the required rank condition for uniform observability, as ∂ H̃ f˜2 (x 2 , z, u cl ) f˜2 (x 2 , z, 0)
+ −
established in [24], holds true ∂x f˜3 (x 2 , u cl ) f˜3 (x 2 , 0)
−1
D − M12 M22 Ju and using (27) and (28), the following inequality holds:
rank −1 −1 −1
Ju − DM22
H̃˙ ≤ −c3 x du 2 + c4 x du L u cl .
M12 M22 Ju M22 Ju
(29)
D
= rank −1 =6 = n−r
−DM22 Ju To use the first term on the right-hand side of inequality (29)
to dominate the second term for large x du , inequality (29)
with M12 = 05×6 in the case studied. can be rewritten as
Furthermore, in the sense of the proof of [24, Proposition 1],
the latter rank condition permits us to apply [28, Th. 8.5], H̃˙ ≤ −c (1 − θ) x 2 − c θ x 2 + c L x u
3 du 3 du 4 du cl
with storage function V (t, x) = H that additionally satisfies where 0 < θ < 1. Then
W1 (x du ) ≤ H ≤ W2 (x du ) where the radially unbounded c4 L u cl
functions W1 and W2 are given by W1 (x du ) = k1 x du 2 and H̃˙ ≤ −c3 (1 − θ ) x du 2 , ∀ x du ≥
c3 θ
W2 (x du ) = k2 x du 2 , with
for all (x du , u cl ). Hence, the conditions of [28, Th. 4.19] that
1 1 1 1
k1 = min , , , proves ISS are satisfied, with the functions determined there
2 2α1 2α2 2α3 given as α1 (r ) = c1r 2 , α2 (r ) = c2r 2 , and ρ(r ) = cc43 Lθ r .
1 1 1 1
k2 = max , , , Therefore, the following proposition is established.
2 2α1 2α2 2α3 Proposition 3.3: The 11th-order closed-loop system (21b)
and x du 2 = x du
T x . These additional conditions guarantee and (21c) is ISS with input gain
du
that the origin of the 11th-order unforced closed-loop system −1 c2 c4 L
is not only an AS but also a GES equilibrium point. γ (r ) = α1 ◦ α2 ◦ ρ = r.
c1 c3 θ
Nevertheless, as GES for the 11th-order unforced closed-
loop system has been established, using (converse Lyapunov Finally, let us now assume the unique equilibrium
theorems) [28, Th. 4.14], a suitable Lyapunov function H̃ ∗
x du
exists for the same system that satisfies the following three T
grid,ref grid,ref ∗ ∗ ∗ ∗ ∗ ∗ ∗
inequalities: = Psref PGSC Q refs Q GSC ω λ λ z z z z
r ds qs dr qr d f q f
c1 x du 2 ≤ H̃ (x du ) ≤ c2 x du 2 (27a) (30)
∂ H̃ f˜2 (x 2 , z, 0) which can be obtained under the assumption that all the ele-
≤ −c3 x du 2 (27b)
∂x f˜3 (x 2 , 0) 11×1 ments of the closed-loop input vector u cl are constant, with the
power references known and the mechanical torque unknown.
∂ H̃
Considering Propositions 3.1 and 3.3, which prove passivity
≤ c4 x du (27c)
∂x and state boundedness of (21b) and (21c), [25, Lemma 3]
and [25, Th. 4] are satisfied, resulting in convergence to the
for some positive constants c1 , c2 , c3 , and c4 .
unique equilibrium.
Note that H̃ always exists although it is not known how
For this equilibrium, it should be now proven that also the
to construct it, with main characteristic the “fully damped”
12th state x 1 = Vdc converges to its equilibrium. To this end,
derivative as shown by (27b) instead of the well-known
considering the cascaded form of (21), the following dynamic
initially used H with the “partially damped” derivative, as
expression is acquired for Vdc state:
Remark 3.2 describes.
1 grid,ref 1 grid,ref
1 In a similar way, the boundedness of states x
du and the form of (26), as C V̇dc = P mq f + Q m d f − Idc,RSC
pointed out in Remark 3.1, establishes the globally Lipschitz property in xdu , Ugrid GSC Ugrid GSC
uniformly in t, of (21b) and (21c). (31)
770 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 2, NO. 4, DECEMBER 2014
Fig. 2. Overall control block diagram and the DFIG wind system.
with the RSC dc current shown in Fig. 2. where ωn2 = k I dc m ∗q f /Ugrid and 2ζ ωn = k Pdc m ∗q f /Ugrid .
Since all states of x du are bounded and limiters are incor- Applying the 2% criterion [31], the desired
settling time Ts is
porated for all duty ratios to keep their values bounded by linked with ζ and ωn through Ts = 4 (ζ ωn ), leading to the
absolute one for linear modulation [30], it is concluded that following gain expressions:
the dc-link voltage dynamics are bounded. Then, the Vdc state
8CUgrid 16CUgrid
can also converge to its desired value under the independent k Pdc = and k Pdc = ∗ 2 2 .
stable action of a slower outer voltage controller. The analysis m ∗q f Ts m q f ζ TS
of such a controller is given in the next section.
To avoid the gain dependence on m ∗q f , which may vary in
a wide operation range of the DFIG, the division of the above
D. DC-Link Voltage Controller acquired gain expressions is used
The following voltage controller is introduced: k Pdc ζ 2 Ts
t = . (35)
grid,ref ref k I dc 2
PGSC = −k Pdc Vdc + k I dc Vdc − Vdc dτ . (32)
0 Expression (33) can be rewritten using (35) as
Recalling (31) that describes the dc-voltage dynamics and 1
using (32), we note that it is common practice in the analysis Vdc CUgrid
= . (36)
of the GSC to consider the dc current on the opposite side as a ref 1 ζ 2 Ts
Vdc γ s2 + CUgrid 2 s + 1
CUgrid
disturbance input [6]. The reactive power output, as appeared
in the second term of (31), is usually controlled to take small It is clear from (36) that the
closed-loop
performance may
values, to minimize the rating of the GSC, or even becomes vary with parameter γ = 1/ k I dc m ∗q f , which means that this
zero when reactive neutral operation is desired (when power controller design should be accompanied by a generalized root
factor equals to one). As a result, the last two bounded terms locus of the closed-loop system with varying parameter γ ,
of (31) can be considered as disturbances and the following so that stable performance is accomplished for the whole
closed-loop transfer function is derived: expected operation range of the DFIG system, as further
m ∗q f clarified in the example of Fig. 4. As m ∗q f is expected to remain
Vdc CUgrid k I dc in a predetermined range of values, the root locus analysis
ref
= m ∗q f m ∗q f
(33)
Vdc provides an accepted value range for the dominant pole of (36),
s2 + CUgrid k Pdc s + CUgrid k I dc which should be associated with the desired settling time Ts .
where Assumption 3.1 and the following standard Then, through the expression of parameter γ, an appropriate
assumption [6] have been adopted. value for the integral gain k I dc can be selected, while the value
Assumption 3.2: Duty ratio m q f is considered to be constant of k Pdc is determined from (35) for the desired settling time
and equal to its steady-state value m ∗q f . and damping factor.
Transfer function (33) allows the controller designer to
reach a desired closed-loop transfer function with ζ damping
E. MPPT Operation Controller
factor, ωn natural frequency and unity static gain as follows:
Although Psref can be arbitrarily determined, in values less
Vdc ωn2 grid,ref
than Pmmax − PGSC , an outer speed controller can be used to
= (34)
ref
Vdc s 2 + 2ζ ωn s + ωn2 achieve MPPT operation. Such a solution can be implemented
BOURDOULIS AND ALEXANDRIDIS: DPC OF DFIG WIND SYSTEMS 771
TABLE I
2-MW DFIG W IND S YSTEM PARAMETERS
Fig. 3. Simplified block diagram of the rotor dynamics and the proposed
MPPT controller.
TABLE II
C ONTROLLER G AINS
V. C ONCLUSION
A novel nonlinear model of the DFIG, suitable for DPC
was extracted and, based on its form, suitable simple and
parameter-free PI power controllers were proposed. A common
grid-VOC alignment method was used for both the converters,
a fact that simplifies the real-time implementation of the
proposed control scheme, while it was shown that a near stator-
flux oriented operation was achieved for the induction genera-
Fig. 7. Stator, GSC, and total DFIG active power responses. tor. A rigorous nonlinear stability analysis based on advanced
Lyapunov methods was conducted and suitable MPPT and
dc-link voltage controllers were designed. The proposed
system modeling, controller design, and analysis were
successfully evaluated through simulation results under hard
conditions, i.e., wind speed and reactive power reference rapid
changes.
A PPENDIX
A. Open-Loop Nonlinear Model of the DFIG
The duty ratios at the RSC and the GSC can be defined as
Vqr Vd f Vq f
m dr = Vdr
Vdc , m qr = Vdc , md f = Vdc , mq f = Vdc
.
Using (10), (12)–(14), and the above duty-ratio expres-
sions, the following expression is acquired for the dc-link
dynamics:
Fig. 8. Stator, GSC, and total DFIG reactive power responses. f 1 (x 1 , x 2 , m(u)) = amT (x 2 ) m (x 1 , u)
with the fourth-order vector am (x 2 )
⎡ ⎤
U U
− 2L3m λds + LLms Udgrid Ps + Uqgrid Qs
⎢
2 2
⎢ grid grid ⎥
⎥
L s Uqgrid Udgrid
⎢ −
1 ⎢ 2L m
3
λ + P − Q ⎥
qs Lm U 2 s U 2 s
⎥
am (x 2 ) = ⎢ grid grid
⎥
C⎢ U dgrid grid U
PGSC + U 2 Q GSC
qgrid grid ⎥
⎢ 2
Ugrid ⎥
⎣ Uqgrid grid
grid
Udgrid grid
⎦
U 2 PGSC − U 2 Q GSC
grid grid
and
⎡ ⎤
m dr
⎢ m qr ⎥ 1
m (x 1 , u) = ⎢ ⎥
⎣ m d f ⎦ = Vdc u
mq f
T
where u = Vdr Vqr Vd f Vq f .
Also, using (8)–(11) and (13)–(14) then expression
Fig. 9. (a) DC-link voltage. (b) RSC and GSC duty-ratio responses. f 2 (·) is a seventh-order column vector with row elements
f 2i (i = 1, 2, . . . , 7) provided
at the top of the next page,
where σ = (L r L s − L 2m ) (L r L s ) is the leakage factor, ωslip =
power changes, while due to the absence of decoupling terms ωs − pωr is the slip angular frequency, α1 = 3/(2σ L r L s ),
in the power controllers insignificant disturbances are present α2 = 1/(J L r Ugrid 2 ), α = 2R /(3R U 2 ), and U 2
3 s r grid grid =
when reactive power changes occur. Clearly the analysis in Udgrid + Uqgrid , with Ugrid being equal to the grid peak phase
2 2
2
f21 = − α1 (Rs L r + Rr L s ) Ps − ωslip Q s + α1 Rr Udgrid λds + α1 Rr Uqgrid λqs − pα1 L r Uqgrid λds − Udgrid λqs ωr
3
2
−α1 L m Udgrid Vdr − α1 L m Uqgrid Vqr + α1 L r Ugrid
2
R f grid grid 3 Udgrid 3 Uqgrid 3 Ugrid
f 22 = − PGSC − ωs Q GSC − Vd f − Vq f +
Lf 2 Lf 2 Lf 2 Lf
2
f23 = ωslip Ps − α1 (Rs L r + Rr L s ) Q s + α1 Rr Uqgrid λds − α1 Rr Udgrid λqs + pα1 L r Udgrid λds + Uqgrid λqs ωr
3
−α1 L m Uqgrid Vdr + α1 L m Udgrid Vqr
grid R f grid 3 Uqgrid 3 Udgrid
f 24 = ωs PGSC − Q GSC − Vd f + Vq f
Lf 2 Lf 2 Lf
b 1
f25 = pα2 L r Uqgrid λds − Udgrid λqs Ps − pα2 L r Udgrid λds + Uqgrid λqs Q s − ωr − Tm
J J
f26 = α3 Rr Udgrid Ps − α3 Rr Uqgrid Q s + ωs λqs
f27 = −α3 Rr Uqgrid Ps + α3 Rr Udgrid Q s − ωs λds
B. Matrix Expressions for the Closed-Loop System Analysis under the assumptions
that Ju is uniformly
bounded for all
Jd (x du ) − Rd D
The 11th-order closed-loop system matrices are t ≥ 0 and matrix is locally Lipschitz in
provided here −D T Ju
the sense that provides a Lipschitz constant.
M11 05×6 1 1 1
M = , M11 = diag , 1, , 1,
06×5 M22 α1 α α2
1 R EFERENCES
1 1 Jd (x du ) D
M22 = diag , , 1, 1, 1, 1 , J (x du ) =
α3 α3 −D T Ju
⎡ ω ⎤ [1] R. Teodorescu, M. Liserre, and P. Rodriguez, Grid Converters for
0 α3s Photovoltaic and Wind Power Systems. West Sussex, U.K.: Wiley,
Rd 05×6 0
R= , Ju = − ωα3s 0 2×4 ⎦,
⎣ 2011.
06×5 06×6 [2] F. Blaabjerg and K. Ma, “Future on power electronics for wind turbine
04×2 04×4 systems,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 1, no. 3,
Jd (x du ) pp. 139–152, Sep. 2013.
⎡ ω ⎤ [3] H. Polinder, J. A. Ferreira, B. B. Jensen, A. B. Abrahamsen, K. Atallah,
0 0 − αslip1
0 − pL r Ugrid λds and R. A. McMahon, “Trends in wind turbine generator systems,” IEEE
⎢ 0 0 0 −ωs 0 ⎥ J. Emerg. Sel. Topics Power Electron., vol. 1, no. 3, pp. 174–185,
⎢ ωslip ⎥
=⎢ ⎢ pL r Ugrid λqs ⎥
Sep. 2013.
α1 0 0 0 ⎥, [4] G. Abad, G. Iwanski, J. López, L. Marroyo, and M. Rodríguez, Doubly
⎣ 0 ωs 0 0 0 ⎦ Fed Induction Machine: Modeling and Control for Wind Energy Gener-
pL r Ugrid λds 0 − pL r Ugrid λqs 0 0 ation Applications. Hoboken, NJ, USA: Wiley, 2011.
[5] A. Yazdani and R. Iravani, Voltage-Sourced Converters in Power Sys-
2 Rf tems: Modeling, Control, and Applications. Hoboken, NJ, USA: Wiley,
Rd = diag (Rs L r + Rr L s ) + L m Ugrid k P Pr , 2010.
3 Lf
[6] R. Pena, J. C. Clare, and G. M. Asher, “Doubly fed induction generator
3 Ugrid 2 using back-to-back PWM converters and its application to variable speed
+ k P P f , (Rs L r + Rr L s ) wind-energy generation,” IEE Proc.-Elect. Power Appl., vol. 143, no. 3,
2 Lf 3
pp. 231–241, May 1996.
Rf 3 Ugrid 2 [7] L. Xu and P. Cartwright, “Direct active and reactive power control
+ L m Ugrid k P Qr , + k P Q f , bL r Ugrid of DFIG for wind energy generation,” IEEE Trans. Energy Convers.,
Lf 2 Lf
⎡ ⎤ vol. 21, no. 3, pp. 750–758, Sep. 2006.
0 Rr Ugrid 0 −αqr 0 0 [8] W. Qiao, W. Zhou, J. M. Aller, and R. G. Harley, “Wind speed estimation
⎢ 0 0 0 0 0 −αq f ⎥ based sensorless output maximization control for a wind turbine driving
⎢ ⎥
⎢
D = ⎢ Rr Ugrid −αdr 0 0 ⎥
a DFIG,” IEEE Trans. Power Electron., vol. 23, no. 3, pp. 1156–1169,
0 0 ⎥. May 2008.
⎣ 0 0 0 0 −αd f 0 ⎦ [9] H. Nian, Y. Song, P. Zhou, and Y. He, “Improved direct power control
0 0 0 0 0 0 of a wind turbine driven doubly fed induction generator during transient
grid voltage unbalance,” IEEE Trans. Energy Convers., vol. 26, no. 3,
pp. 976–986, Sep. 2011.
C. Some Nonlinear Stability Prerequisites [10] T. Noguchi, H. Tomiki, S. Kondo, and I. Takahashi, “Direct power
control of PWM converter without power-source voltage sensors,” IEEE
For the reader convenience, [24, Proposition 1] is provided Trans. Ind. Appl., vol. 34, no. 3, pp. 473–479, May/Jun. 1998.
here. [11] M. Malinowski, M. P. Kazmierkowski, S. Hansen, F. Blaabjerg, and
Proposition 1 [24]: Hamiltonian system (26) is uniformly G. D. Marques, “Virtual-flux-based direct power control of three-phase
PWM rectifiers,” IEEE Trans. Ind. Appl., vol. 37, no. 4, pp. 1019–1027,
asymptotically stable if Jul./Aug. 2001.
−1
D − M12 M22 Ju [12] D. Zhi, L. Xu, and B. W. Williams, “Improved direct power control
rank −1 −1 −1 ≥ n −r of grid-connected DC/AC converters,” IEEE Trans. Power Electron.,
M12 M22 Ju M22 Ju − DM22 Ju vol. 24, no. 5, pp. 1280–1292, May 2009.
BOURDOULIS AND ALEXANDRIDIS: DPC OF DFIG WIND SYSTEMS 775
[13] S. Vazquez, J. A. Sanchez, J. M. Carrasco, J. I. Leon, and E. Gal- [29] J.-J. E. Slotine and W. Li, Applied Nonlinear Control.
van, “A model-based direct power control for three-phase power con- Upper Saddle River, NJ, USA: Prentice-Hall, 1991.
verters,” IEEE Trans. Ind. Electron., vol. 55, no. 4, pp. 1647–1657, [30] B. K. Bose, Modern Power Electronics and AC Drives.
Apr. 2008. Upper Saddle River, NJ, USA: Prentice-Hall, 2002.
[14] M. K. Bourdoulis and A. T. Alexandridis, “Direct power flow modeling [31] J. J. D’Azzo and C. H. Houpis, Linear Control System Analysis and
and simple controller design for ac/dc voltage-source converters,” in Design: Conventional and Modern, 2nd ed. New York, NY, USA:
Proc. 39th Annu. Conf. IEEE IECON, Nov. 2013, pp. 637–642. McGraw-Hill, 1981.
[15] D. Zhi and L. Xu, “Direct power control of DFIG with constant
switching frequency and improved transient performance,” IEEE Trans.
Energy Convers., vol. 22, no. 1, pp. 110–118, Mar. 2007.
[16] P. Zhou, Y. He, and D. Sun, “Improved direct power control of a DFIG-
based wind turbine during network unbalance,” IEEE Trans. Power
Electron., vol. 24, no. 11, pp. 2465–2474, Nov. 2009. Michael K. Bourdoulis (S’07) was born in Patras,
[17] D. Zhi, L. Xu, and B. W. Williams, “Model-based predictive direct Greece, in 1986. He received the Diploma degree
power control of doubly fed induction generators,” IEEE Trans. Power in electrical and computer engineering and the B.A.
Electron., vol. 25, no. 2, pp. 341–351, Feb. 2010. degree in economics from the University of Patras,
[18] N. Amiri, S. M. Madani, T. A. Lipo, and H. A. Zarchi, “An improved Patras, in 2008 and 2014, respectively, where he
direct decoupled power control of doubly fed induction machine without is currently working towards the completion of the
rotor position sensor and with robustness to parameter variation,” IEEE Ph.D. degree in electrical and computer engineering
Trans. Energy Convers., vol. 27, no. 4, pp. 873–884, Dec. 2012. at the Department of Electrical and Computer Engi-
[19] J. Hu, H. Nian, B. Hu, Y. He, and Z. Q. Zhu, “Direct active and reactive neering.
power regulation of DFIG using sliding-mode control approach,” IEEE His current research interests include nonlinear
Trans. Energy Convers., vol. 25, no. 4, pp. 1028–1039, Dec. 2010. modeling, control, and stability analysis of power
[20] M. K. Bourdoulis and A. T. Alexandridis, “Rotor-side PI controller converter, and electric machine systems with an emphasis on renewable energy
design of DFIG wind turbines based on direct power flow modeling,” systems.
in Proc. 2nd ICRERA, Oct. 2013, pp. 155–160. Mr. Bourdoulis is a Member of the National Technical Chamber of Greece.
[21] M. K. Bourdoulis and A. T. Alexandridis, “An alternative modeling and
controller design guaranteeing power stability for DFIG wind systems,”
in Proc. 52nd IEEE CDC, Dec. 2013, pp. 7461–7466.
[22] E. Rezaei, A. Tabesh, and M. Ebrahimi, “Dynamic model and control
of DFIG wind energy systems based on power transfer matrix,” IEEE
Trans. Power Delivery, vol. 27, no. 3, pp. 1485–1493, Jul. 2012. Antonio T. Alexandridis (M’88) received the
[23] A. Balogun, O. Ojo, and F. Okafor, “Decoupled direct control of natural Diploma degree in electrical engineering from the
and power variables of doubly fed induction generator for extended wind Department of Electrical Engineering, University of
speed range using feedback linearization,” IEEE J. Emerg. Sel. Topics Patras, Patras, Greece, in 1981, and the Ph.D. degree
Power Electron., vol. 1, no. 4, pp. 226–237, Dec. 2013. from the Department of Electrical and Computer
[24] E. A. Androulidakis and A. T. Alexandridis, “On the stability of unforced Engineering, West Virginia University, Morgantown,
nonautonomous underdamped dissipative Hamiltonian systems,” in Proc. WV, USA, in 1987.
18th Int. Conf. MMAR, Aug. 2013, pp. 11–15. He joined the Department of Electrical and Com-
[25] G. C. Konstantopoulos and A. T. Alexandridis, “Stability and conver- puter Engineering at the University of Patras in 1988,
gence analysis for a class of nonlinear passive systems,” in Proc. 50th where he is currently a Professor and the Head of the
IEEE CDC-ECC, Dec. 2011, pp. 1753–1758. Power Systems Division. During 1998, he joined the
[26] L. Yang, Z. Xu, J. Ostergaard, Z. Y. Dong, and K. P. Wong, “Advanced Control Engineering Research Centre at the City University London, London,
control strategy of DFIG wind turbines for power system fault ride UK, as a Visiting Researcher. He has authored more than 130 international
through,” IEEE Trans. Power Syst., vol. 27, no. 2, pp. 713–722, journal and conference papers. His research interests include control theory,
May 2012. nonlinear dynamics, optimal control, eigenstructure assignment, passivity,
[27] P. C. Krause, O. Wasynczuk, and S. D. Sudhoff, Analysis of Electric and advanced control applications on power systems and drive systems. His
Machinery and Drive Systems, 2nd ed. Piscataway, NJ, USA: Wiley, current interests include renewable power generation control and stability
2002. (wind generators, PV systems, and microgrids).
[28] H. K. Khalil, Nonlinear Systems, 3rd ed. Upper Saddle River, NJ, USA: Prof. Alexandridis is a Member of the National Technical Chamber of
Prentice-Hall, 2002. Greece.