Indian Journal of Science and Technology
Vol:5 Issue:12 December 2012 ISSN:0974-6846
A Novel Control Strategy Study for DFIG-Based Wind Turbine
Reza Sedaghati
Young Researchers Club, Beyza Branch, Islamic Azad University, Beyza, Iran
reza_sedaghati@yahoo.com
Abstract
In recent years the use of renewable energy including wind energy has risen dramatically. Because of the increasing development of
wind power production, improvement of the control of wind turbines using classical or intelligent methods is necessary. In this paper,
in order to control the power of wind turbine equipped with DFIG, a novel intelligent controller based on the human mind’s emotional
learning is designed. The performance of proposed controller is confirmed by simulation results. Some outstanding properties of this
new controller are online implementation capability, structural simplicity and its robustness against any changes in wind speed and
system parameters variations.
Keywords: Double Fed Induction Generator(DFIG), Emotional learning, Wind turbine, Intelligent controller.
1. Introduction
2. Model of Wind Turbine Equipped With DFIG
One of the common controllers in industrial systems is the 2.1 Model of Rotor Aerodynamic
classical PI controller that despite its simple construction enjoys
Mechanical power of wind turbine that in fact is a percentage
an acceptable performance (especially in the areas of linear sys- of the total wind energy, is calculated as follows:
tems) [1]. In [2,3] two samples of the application of this controller
ρ
3
(1)
is given for controlling wind turbines. However, in systems such Pr = AC P (λ , β )V
2
as wind turbine, which enjoys non-linear and undefined factors,
3
in principle using a simple PI controller will not follow a desired P =Inρ which
AC P (λρ, βis)Vthe
air density, C P is power efficiency, λ is the
r
2
response. This is due to the fact that the control system parameters edge speed, is the step-angle, is the surface swept by the rotor
A
β
with wind speed change (and therefore changing the system work and finally
ωr R V is the wind speed. Parameter λ is also defined as folλ=
point) and system parameters change need to be adjusted again.
V
lowing:
This defect may be overcome in two ways: the use of classi(2)
ωR
cal linear control methods (adaptive /resistant) or the use of neu- λ = r
V
ral - fuzzy intelligent control methods (having adaptive-extension
power and the possibility of on – line implementation). In [4,5] the
In which R is the radius of the rotor and ω r is the rotor anguclassical control method of sliding mode (which is the oldest and lar speed. Due to the constant parameters ρ and A and also the
still most popular method of resistant nonlinear control) is used to wind speed is not at our disposal, from (1) it is implied that by
control the wind turbine system. Also in [6-8] the neural - fuzzy in- adjusting
C P (which is a function of two parameters λ and β ),
telligent methods are used to achieve better solutions in the design P can be controlled to a desired manner. It should be explained
r
of wind turbine control system.
that in wind speeds higher than rated speed , C P tuning is done by
Wind turbines equipped with double fed induction generator
parameter β and in wind speeds lower than rated speed (of course in
of coiling rotor due to their advantages (including: production caturbines with variable speed) C P tuning is done by parameter λ . In [11],
pacity in a range of wind speed, good performance, no need for causing numerical approximation methods, the relation between C P and
pacitor banks for producing required reactive power of generator
parameters λ and β is obtained as −following:
c5
and lower-cost of power electronic converters) have widely been
− c5
c2
λi
c
(3)
C
=
(
λ
,
β
)
c
−
c
β
−
c
e
CPP (λ , β ) = 1c1λi λ2i −3 c3 β −4 c 4 e +λi c 6+λ c 6 λ
used. Therefore, analysis of wind turbines equipped with double
fed induction generators and methods for controlling it, is one
−1
of the main analyses in producing wind power. In [9,10] for the
(4)
=
λi λ +01.08 β − 0β.035
3
+1
implementation of the proposed control strategies for controlling
wind turbines equipped with DFIG, PI controllers have been used.
In Figure 1, C P specification has been drawn based on λ and
In this paper, the proposed control strategy for controlling active
by different values of
​​ β .
C P specification has been drawn based on λ and by
and reactive power of wind turbine equipped with DFIG has been
implemented with the help of emotional intelligence controllers.
((
(
Research article
3741
) )
)
www.indjst.org
43
Vol:5
Issue:12
C
=
P (λ , β ) c1
( December
− c β −2012
c )e ISSN:0974-6846
+c λ
− c5
c2
3
λi
4
λi
Indian Journal of Science and Technology
6
Fig.1. Specification of power efficiency in terms of edge speed and
different values ​​of pitch-angle
[11].
−1
λi
(
− 0.035
1
(
)
)
3
− c5
λ + 0.08 β
cβ2 +1
λi
C
=
+ c6 λ
P (λ , β ) c1 λi − c 3 β − c 4 e
(
(
C
=
(λ , β ) c (
C
=
(λ , β ) c (
)
)
− c β − c )e
− c β − c−1)e
−1
controlled through the rotor Iqr flow, while the reactive power for
ψ s = ψ d can be controlled through the rotor Idr flow.
3. Designing Intelligent Controller Based on
Human Brain Mechanisms of Emotional Learning
c − 0.035
In recent
years,
hasλ been
values
of the
β development of computational models of
=
λiP (λ , βλC+) P01.08
C
=
cspecification
+ c 6 λdrawn based on λ and by different
1β λ − c 33 β − c 4 e
2
β +1
− c5
i
the function of parts of the brain that are responsible for emotional
+
c
λ
processing task is of interest to researchers. In [13], a new compuP
3
4
+ c 6 λ6
4
=
λPi λ +01.08 β − 0β3.035
tational model of the function of brain emotional processing unit,
3
+1
C P specification
has been drawn based on λ and by different values of β
−
1
including units: the amygdala, orbitofrontal cortex, thalamous and
.035−1
=
λi λ +011.08 β −00.β035
3
=
λi
finally the sensory cortex are presented (Figure 2). According to
dψ − 3 ++11
ψs
I s Rs + V s =λ +−0.08 β s −β jω
the above
model
CdtP specification has been drawn based on λ and by different
values
of and
β based on new theories the amygdala - orbitofrontal cortex system performs learning process in two steps.
C P dspecification
has
been
drawn
based
on
and
by
different
values
of
λ
ψ
At first,
the of
input
s
values
=
ψ s ILs Rs sI +s V+ sL=CmP−I rspecification
β βstimulation signals are evaluated and then the
− jωψ s has been drawn based on λ and by different
dt
assessment is used as the multiplier coefficients in response af2.2 The Generator Model
fected by stimulation. Amygdala is one of the primary structures
ψ r ψmodel in synchronous coordinates are as
Relations
for dDFIG
+V
−m Ij s(rω1 − ω r )ψ r
I r=
Rr ψ
r = Ls−I s + Ld
s
of the brain which exists relatively uniform in the form of large
+ V s =dt−
− jωψ s
I s Rs[12]:
follows
dt
scale structures among different species of animals. Amygdala
dψ s
ψrs −−j (jωωψ−sω )ψ
II sRRs ++VVrs= = −−ddψ
is a small network in temporal lobe, which has the task of emoILrs rRIrs r+L+V L
s s= +
(5)
ψ=
=
I−sLdtmdtI r− jωψ1 s r r
r ψ
s
sIm
dt
tional assessment of stimuli that these assessments, are used on:
moods, emotional reactions, attention signals and also long-term
=
ψ
ψ rs LLLrsIIIrss+++LLdLmψmIIrIs r
=
(6)
=
ψ
I rsRr + Vs r = − m r − j (ω1 − ω r )ψ r
memory. A number of intrinsic stimuli such as hunger, pain, some
dt
dψ r
smells etc., can stimulate the amygdala and the amygdala response
I R + V r = −dψ − j (ω − ω )ψ
ψs
I rr Rrr +LV rI = +− L dtIr s− j (ω1 1− ω r )rψ r r
(7)
ψ
=
to these stimulations
is used in learning. Research conducted by
r
r r
m
dt
scientists show
ψ s that animals that have suffered damage from the
ψ r Lr I r + Lm I s dψ q
=
(8)
amygdala
have
difficulty in learning and evidence of this is that
I rψ+s L=
I sd ;
ψ=
=0
mψ
s r=
q =ψ0r; ψ L
dtdψ
the learning is done in the amygdala. On the other hand, orbitofq
ψs =
ψd;
ψ q = 0; ψ s =
=0
the role of correcting responses and adverse res
dt system to the stator flux causes the rantal cortexψplays
Paralleling the coordinate
actions of amygdale. Experiments done on patients with damaged
constant, so the stator voltage amstator flux ψ s becomes almost
ψs
dψ q
ψ s cortex
show that these patients are not able to adapt
=
ψ sand
=
ψphase
ψ q =frequency
0; ψ
= fixed.
0broken into two dq axes and isorbitofrantal
d ; base
areis
Theplitude,
equation
ofs the
stator
as
follows:
dt
themselves to new conditions (in other words, previous learning
dψbase
The
equation
stator
is broken into two dq axes(9)
and is as follows:
q
d=
ψofψd sthe
ψ
d
;
ψ
ψ
=
ψ
=
0
;
=
0
q
s
d
prevent understanding; subsequently an appropriate response to
q
I d Rsψ+qV=d 0=
ψd
ψ d ; dt = 0
=
; ψ−s =
sψ
d
dt
dt
I d Rs + Vd =
−
new conditions).
The equation ofdtthe stator base is broken into two axes and is
The
equation
of the stator base is broken into two dq axes and is as
follows:
Fig.2.
Graphical description of the detailed computational model
follows:
I as
−ωrψ
q Rs + Vq =
q
(10)
I q Rs + Vq =
−ωdrψ
ψdq
of emotional learning system of the brain [13].
I d Requation
− the stator base is broken into two dq axes and is as follows:
The
of
s + Vd =
The equation of the
dt stator base is broken into two dq axes and is as follows:
dψ
I d Rs + Vd =
−dψ d d
IIdqRRss ++VVdq =
−−ωrdt
(11)
=
ψ
dtψq does not change very much dψ
SinceSince
the stator
fluxflux
and considering
have:
0 we
dψ does not change very much dψ d ≈ 0and considering R R=s 0=we
the stator
have:
d
s
≈ 0 and
dt
Since
the
stator
flux
does
not
change
very
much
ψ
I q Rs + Vq =
−ωrψ q
dt
d
Rs +=
Vqω=
ωrψ q
Vconsidering
0I q,dV
ψ−−0ω
V
=
d =
q 0, V−
r=
q we
Rq s =
rψ qhave:
Since
the stator flux
does not change very much dψ (12)
and considering Rs = 0 we have:
V = 0, V = −ω ψ ψ d
(
((
i
c2
1 c2λi
1 λi
)
))
− c5
− c5λi
λi
d
d
q
r
d
q
dt
≈0
Vd =
0, Vqstator
=
−ωrflux
ψ q ψ d does not change very much dψ
Since
the
and considering R = 0 we have:
≈ 0 considering R = s0 we have:
Since
theactive
stator and
flux reactive
not change
verybe
much
and
ψ d doespower
dψas
s
Stator
can now
stated
dt≈ follows:
0
d
d
dt
V =
0, V =
−ωrψ q
L I
03, Vqq =
−ω ψ
3 3
3 3 1ψ dLmmI qrqr ; ω1 = ω r
3Vdd =
d IV
d +IrV)qqI≈q ) ≈V IVq I= q = ω ω
Ps = Ps(V= d2I (dV+
ψ
21 d
L ; ω1 = ω r
q q
q2 q
2
2
2
Ls s
(13)
3 (V I − V I ) ≈ 33ω ψ I =33 ω ψ dL(mψI qr− L I )
(14)
= ωr
ψ1
dd Iqd + Vqq Id q3) ≈ V
1 q Idq d= 3 ω1ψ
d ;ω
ω1 dr
(Vd 2I q − Vq I d ) ≈ ω12ψ2 d I d = 2ω21 dLs(ψLds − Lω I dr )
=
Qs
2 3
2 3
23 Ls Lm I qr
Lm I qr ; ω1 = ω r
PThe
(Vd I d equation
+ Vq I q ) ≈clearly
= 3 ω1ψthat
3 Vq I q shows
s = 3above
control of the
Ps = 23(Vd I d + Vq I q ) ≈ 32Vq I q = 2ω31ψ d ψd d in
; ω1 = ω
L vector
r
In this paper, according to the model presented in [13] from
2
2
2
)
=
Qs flux,(Vactive
ω1ψ d I d =or absorbed
ω1 L(sψsdby
− Lthe
d I q − Vpower
q I d ) ≈ delivered
ω I drstator
stator
can be
2
2
2
Ls
human brain mechanisms of emotional learning, an intelligent
3
3
3 ψ
=
Q 3 (V I − V I ) ≈3 ω ψ I =3 ωψ1 d d (ψ d − Lω I dr )
www.indjst.org
=
Qss 2(VddI qq − VqqI dd) ≈ 2ω1ψ1 d dI d d= ω
2 1 L(ψ d − Lω I dr )
3742
Research article
2
2
2 Ls s
44
=
Q
3Pss =
Indian Journal of Science and Technology
Vol:5 Issue:12 December 2012 ISSN:0974-6846
control strategy has been presented for controlling blocks which
have the same performance with a self-regulatory PID controller.
Regardless of the details, schematic of brain emotional learning
system has been shown in Figure 3 that following in order to illustrate the proposed computational model of emotional learning,
the amygdala- orbitofrontal is used.
Fig.3. Block view of computational models of brain mechanisms of emotional learning [13].
t
SI k P ⋅ e + K I ⋅ ∫ e dt + K D ⋅ e
=
In which e is the tracking error of closed-loop system. It
should be explained that for various reasons including the noise of
t
measurement
SI k P ⋅ e devices,
dt + K Dworse
=
+ K I ⋅ ∫ eactually
⋅ e the operating system so at
0
best, PID controller will
not have better performance than PI controller. Therefore, the only two options P and PI will remain for SI
stimulation signal. EC Emotional signals in general, should indicate the desirability of the performance of the controller and closed
loop system. Therefore, without losing the whole, can be written
based
combination of the primary / secondary areas
ECon aaweighted
=
ec1 ⋅ e + a ec 2 ⋅ MO + other terms
of application (including tracking the desired output, trying to control the minimum or equivalent maximum efficiency, etc.,):
EC a ec1 ⋅ e + a ec 2 ⋅ MO + other terms
=
MO Computational model output (the emotional learning
amygdala-orbitofrontal system’s response to input stimuli and the
emotional rewards / Composition EC signal) is equal to:
(15)
MO = AO - OCO
In which AO and OCO are respectively output units of amygdala and orbitofrontal and are equal to:
=
AO
G ⋅ SI
AO = Gaa. SI
(16)
=
OCO Goc ⋅ SI
OCO = GOC . SI
(19)
0
(20)
From the above, the easiness of embedding the secondary
goals in intelligent structures (including the obvious advantages of
intelligent structures in comparison with the classical structures)
is clearly implied. In figure 4 block display of the proposed intelligent controller based on human brain a mechanism of emotional
learning has been shown.
Fig.4. The block display of proposed intelligent controller based
on the brain emotional learning.
In which Ga and GOC are respectively equivalent gain of
amygdala and orbitofrontal units. Learning low in the amygdala
InOCO
whichGG
and Goc
oca⋅ SI
and =
orbitofrontal
units, respectively, is:
=
AO
Ga ⋅ SI
∆Ga =k1 ⋅ max (0 , EC − AO) ≥ 0
(17)
∆Goc = k 2 ⋅ ( MO − EC )
In In
which
whichGakand
andGkoc are respectively learning rate inamygdala
1
2
and orbitofrontal units. Because of using max operator, unit gain
∆Ga =k1 ⋅ismax
(0 , EC −toAO
) ≥0
of amygdala
subject
increased
univocal changes. In other
∆
=
⋅
−
G
k
MO
EC
(
)
2
words, octhe desired
working conditions (which will be reflected in
EC ) gradually increase the
the large amounts of emotional signals EC
gain of the amygdala unit to its physical limits. However, for unknown reasons, if conditions are unfavorable in the future (with a
small amount of emotional signals EC ) the amygdala unit will not
be =
able
(Ga − Goc this
) ⋅ SIproblem
≡ G ( SI ,and
) ⋅ SIits answer and practiMOto diagnose
EC ,correct
EC
cally will respond the same as desired conditions. However, the
orbitofrontal unit gain is allowed to positive / negative change so
that the amygdala unit can properly carry out the reform against
the unfavorable responses. By combining (15) and (16) we have:
=
MO (Ga − Goc ) ⋅ SI ≡ G ( SI , EC , ) ⋅ SI
(18)
In other words, the output of amygdala-orbitofrontal system
in emotional learning is the product of a variable G gain (dependent on several factors, including EC emotional signals, SI stimulation input, etc.) in the SI stimulus input. Citing the proven values​​
of PID controller, most direct proposal for formulating SI stimulation signal, is a PID-like frame:
Research article
In Figure 5 diagram block of vector control (proposed) of wind
turbine equipped with DFIG has been shown. Also, in Figure 6 details related to the control subsystem related to Figure 5 has been
shown. Control strategy must be in the way that dual fed induction
generator can track be in the maximum absorbable power of the
wind turbine at any wind speed. Accordingly, the controller number 2 has been designed so that the active power produced by two
fed induction generator can track the desired active power at any
time through by injecting the desired voltage to the rotor Q component. The duty of controller No.1 is also production of desired active power (for controller No. 2) by tracking the reference speed of
the rotor. Meanwhile, and amount of exchanges of reactive power
between the DFIG and network is controlled by controlling the
voltage of d component of rotor and this has been performed using
the controller No. 3.
3743
www.indjst.org
45
Vol:5 Issue:12 December 2012 ISSN:0974-6846
Indian Journal of Science and Technology
Fig.5. Diagram block of vector control (proposed) of wind turbine
equipped with DFIG
Fig.8. Rotor speed
1.4
1.2
Refrence
Simulated
rotor speed (p.u)
1
0.8
0.6
0.4
0.2
0
Fig.6. Details related to the control subsystem related to Figure 5.
0
2
4
6
8
10
time (s)
12
14
16
18
Fig.9. Three-phase rotor voltage
1
0.8
0.6
voltage (p.u)
0.4
4. Simulation
0
-0.2
-0.4
-0.6
-0.8
-1
0
2
4
6
8
10
time (s)
14
16
18
14
16
18
12
Fig.10. Active power produced by the DFIG
Fig.7. Wind speed profile
15
2
1
active power (p.u)
All simulations were done using MATLAB software for 18
seconds and in all of them, rated wind speed (to activate pitchangle system) 12 m/s has been in order In Figure 7, wind speed
profile (based on the model presented in Part 2) has been shown
as a default. Figure 8 indicates the remarkable performance of the
proposed intelligent controller in comparison with the classical PI
controller for removing tuck and the responding rate. The simulation results of diagram block in Figure 7, for wind speed profile of
Figure 7, are also given in Figures 8 to 11, respectively. It should
be noted that all used controllers have been designed as intelligent
and based on brain emotional learning Figures 10 and 11 shows active and reactive power produced by the DFIG, respectively. Figures 8 and 9, respectively, show the rotor speed and rotor voltage.
0.2
0
-1
-2
13
wind speed (m/s)
-3
10
-4
2
4
6
8
10
time (s)
12
5
0
0
2
www.indjst.org
46
0
4
6
8
10
time (s)
12
14
16
18
3744
Research article
Indian Journal of Science and Technology
Vol:5 Issue:12 December 2012 ISSN:0974-6846
Wind Turbines, 17th ASME Wind Energy Symposium Proceedings, pp. 89–94.
Fig.11. Reactive power produced by the DFIG
1
4. Beltran, B., Ahmed-Ali, T. and Benbouzid, M. E., (2008) Sliding Mode Power Control of Variable-Speed Wind Energy
Conversion Systems, IEEE Trans. on Energy Conversion,
June.
0
reactive power (p.u)
-1
-2
5. Utkin, V. I., “Sliding Mode Control Design Principles and Applications to Electric drives, (1993) IEEE Trans. on Industrial
Electronics., Vol. 40, pp. 23–36.
-3
-4
-5
6. Lascu, C., Boldea, I. and Blaabjerg, F., (2004) Direct Torque
Control of Sensorless Induction Motor Drives: a SlidingMode Approach,” IEEE Trans. on Industrial Applications,
Vol. 40, pp.582–590.
-6
-7
-8
0
2
4
6
10
8
time (s)
12
14
16
18
7. Zhang, X.-Y., Cheng. J. and Wang, W.-Q., (2008) The Intelligent Control Method Study of Variable Speed Wind Turbine
Generator, ICSET.
5. Conclusion
Analysis of wind turbines equipped with double-feed induction generators and their control methods is one of the main analyses in wind power generation. Regarding the improved capabilities
of the intelligent controllers (include: high adaption and generalization ability, model-free capability, and also the ability of online
implementation), these controllers can be used as wind turbine
control systems to achieve more desirable responses and also to
facilitate designing process. In this paper, an appropriate control
strategy for active and reactive power control of wind turbines
equipped with DFIG has been presented and implementing this
control strategy has been done with the help of the new emotional
intelligent controllers. According to simulation results, this control
system meets the following objectives:
1. The desired quality from the perspective of transient and
persistent behavioral indicators in the same simple structure
(desired quality in terms of response rate, ripple response and
lasting tracking error),
2. The remarkable consistency against work point changes
(change in wind speed) and system parameters (by changing
the work point and system parameters, they do not need to
be redesigned but against these changes they are modified
automatically to maintain their optimal performance).
8. Jerbi, L., Krichen, L. and Ouali, A., (2009) A Fuzzy Logic
Supervisor for Active and Reactive Power Control of a Variable Speed Wind Energy Conversion System Associated to a
Flywheel Storage System, Electric Power Systems Research,
Vol. 79, pp. 919–925.
6. Acknowledgment
13. J. Moren, and Balkenius, (2000) A Computational Model of
Emotional Learning in the Amygdala, from animals to animals 6: Proc. of the 6th International Conference on the Simulation of Adaptive Behaviour, Cambridge, Mass., (the MIT
Press).
The authors acknowledge financial support from Young Researchers Club (YRC), Beyza Branch.
7. References
9. B. Boukhezzar and H. Siguerdidjane, (2009) Nonlinear control
with wind estimation of a DFIG variable speed wind turbine
for power capture optimization, Energy Conversion and Management, vol. 50, pp. 885–892.
10. L. Fan, H. Yin and Z. Miao, (2011) A novel control scheme
for DFIG-based wind energy systems under unbalanced grid
conditions, Electric Power Systems Research, vol. 81, pp.
254–262.
11. H. Nian and H. Xuh, (2011) Dynamic modeling and improved control of DFIG under distorted grid voltage conditions, IEEE Transaction on Power System, vol. 26, no. 1, pp.
163–175.
12. Ion. Boldea, (2006) A The Electric Generators Handbook
VARIABLE SPEED GENERATORS, J. CRC Press, the United State of America.
1. Astrom, K. J. and Hagglund, T., (1995) PID Controller: Theory, Design, and Tuning, 2nd edition.
2. Tapia, G., Tapia, A. and Ostolaza, J. X., (2006) Two Alternative Modeling Approaches for the Evaluation of Wind Farm
Active and Reactive Power Performances, IEEE Trans. on
Energy Conversion, Vol. 21, pp. 909-920.
3. Hand, M. M. and Balas, M. J., (1998) Systematic Approach for
PID Controller Design for Pitch-Regulated, Variable-Speed
Research article
3745
www.indjst.org
47