Energy and Power E ngineering, 2013, 5, 404-408
doi:10.4236/epe.2013.54B078 Published Online July 2013 (http://www.scirp .org/journal/epe)
Optimum Setting Strategy for WTGS by Using an
Adaptive Neuro-Fuzzy Inference System
Yang Hu, Jizhen Liu, Zhongwei Lin
State Key Laboratory for Alternate Electrical Power System with Renewab le Energ y So urces,
North China Electric Power University, Beijing, P. R. China
Received March, 2013
ABSTRACT
With the popularization of wind energy, the further reduction of power generation cost became the critical problem. As
to improve the efficiency of control for variable speed Wind Turbine Generation System (WTGS), the data-driven
Adaptive Neuro -Fuzzy Inference System (ANFIS) was used to establish a sensorless wind speed estimator. Moreover,
based on the Supervisory Control and Data Acquisition (SCADA) System, the optimum setting strategy for the maxi-
mum energy capture was proposed for the practical operation process. Finally, the simulation was executed which sug-
gested the effectiveness of the appro aches.
Keywords: Wind Energy; Data Processing; Adaptive; Takagi-Sugeno (T-S) Fuzzy; Neuro-N etwork
1. Introduction
Nowadays, wind energy is growing rapidly [1, 2]. The
crucial problem of popularizing wind energy has become
the further reduction of power generation cost. Thus, the
higher efficiency and more optimal operation for wind
power generation are required. In order to improve the
efficiency of WTGS, the notion of maximum energy
capture is introduced. And with the emergenc e of varia-
ble speed variable pitch technique, the operation optimi-
zation can be deepened.
In the practice, many anemometers have to be used to
measure the wind speed which can derive the optimum
setting values of rotor speed and generator power. And
not only the single WTGS but also the wind field needs
many anemometers to provide adequate information.
However, due to the varied environment, the measure-
ment ma y pr o vid e i nac cur a te si gna l t o the s ys tem and t he
traditional o ptimum setting strateg y may ha ve so me drif t
to the initial setting after a period of operation. Besides,
the installation and maintenance of the anemometers in-
crease the cost and reduce the reliability of the whole
s ys t em.
Recent ly, se veral ki nds o f sen sorle ss max i m u m energy
capture method were proposed in the literatures. Bhow-
mik et al. [3] used the power coefficient polynomial to
estimate wind speed by solving the polynomial roots
online with an iterative algorithm. Because the poly-
nomial was seventh order, the calculation was complex
and time-consuming. Tan [4] and Simoes [5] et al. ap-
plied a two-dimensional (2D) look-up table of power
coefficient and power mapping method to estimate the
wind speed, but the technique needed huge memory
space and suboptimum solution was often caused by the
inhere nt slo w se archin g mechanism. H. Li et al. [6 ] used
the Artificial Neuro Network (ANN) to establish a sen-
sorless wind speed estimator but the neuro-network is
easy to be over-fitting or fall into the local minima. V.
Calderaro et al. [7,8] combined the advantages of T-S
fuzzy system [9], Genetic Algorithms (GA) and Fuzzy
C-Means clustering (FCM). Then, an adaptive optimum
setting strategy was realized. However, the GA was un-
stableness and also time- consuming to train the parame-
ters of T-S fuzzy sys t em.
In [10], the ANFIS was firstly proposed by Jang J-SR.
In [11], a kind of ANFIS was applied on the data-mod-
eling for thermal processes. In the paper, in order to im-
prove the training efficiency and accuracy, the ANFIS
was ad o p te d whic h full y c o m b ine d t he a d va nta g es o f T-S
fuzzy system and neuro-network and is very useful for
data-driven modeling.
In section 2, the system analysis of WTGS is carried
out and the profile mapping between rotor speed, wind
speed and mechanical power is discussed. In section 3,
based on the characteristic data from the wind tunnel test,
an adaptive sensorless wind speed estimator is firstly
established by ANFIS. Then, using the wind speed esti-
mator and the measured data source in the SCADA sys-
tem for wind power generation process, the optimum
setting strategy based on measured value is given. In

Y. HU ET AL.
405
section 4, the simulatio n is exec uted to valid ate the e ffec-
tiveness of the approaches. And in section 5, we con-
clude the paper.
2. System Analysis
Generally speaking, the variable speed variable pitch
WTGS include two control level, the wind turbine con-
trol level and Doubly-Fed Induction Generator (DFIG)
control level [12]. From Figure 1, we can see that in or-
der to realize the maximum energy capture, the optimum
setti ng value s of r otor speed
r
ω
and turbi ne mecha nica l
p ower m P are needed besides their measured values.
Usually, the mechanical power extracted from the wind
energy can be represented as
( )
23
0.5 ,
mP
PRC V
ρπλ β
=
where
ρ
is the air density, R is the rotor radius,
( )
,
P
C λβ
is the power coefficient,
r
RV λω =
is the
tip-speed-ratio and
β
is the pitch angle. When
λ
and
β
get the optimum values,
( )
,
P
C λβ
reaches the
maximum value. Then, the wind turbine has the most
efficiency to extract wind energy.
Here, we take the operating region below rated wind
speed for example. The main operating mode is the vari-
able speed fixed pitch operation and the main control
task is the maximum energy capture. Thus, the pitch an-
gle is fixed at zero degree to maintain the maximum
power coefficient
( )
,
P
C
λβ
. Consequently, the nonli-
near profile mapping between
λ
,
β
and
( )
,
P
C λβ
can be simplified to the one between
r
ω
,
V
and
( )
,
P
C λβ
. There are many ways to demonstrate the non-
linear profile mapping such as fitted nonlinear function
and look-up tabl e. U sin g the nonlinear functio n i n [1 3] , a
schematic of the profile curve can be shown in Figure 2.
3. Algorithms Application
In this section, we introduce the ANFIS to establish the
sensorless wind speed estimator. While considering the
the possible drift of the optimum power coefficient curve
to the initial setting, a no vel opti mum settin g mecha nism
is proposed based on the SCADA system.
Rotor side
converter controlGrid side converter control
DFIG control level
θ
meas
r
ω
Optimum
Setting Value
Cross
coupling
Power ControlSpeed Control
Wind turbine control level
opt
r
ω
opt
m
P
PWM
Aerodynamic
System Drive TrainGeneratorConverter
Wind turbine
m
T
r
ω
g
T
g
ω
,PQ
meas
m
P
Figure 1 . Wind T urbine Generation System.
3.1. Wind Speed Estimation
From the analysis in section 2, we know that a two di-
mensional inverse mapping between
r
ω
,
m
P
and
V
needs to be established. Using the characteristic data of
wind tunnel test, the data-driven modeling approach, ANFIS,
is introduced. The modeling mechanism is shown in
Figure 3.
The fuzzy sys tem mainl y includes the Ma mdani fuzz y
system and the Takagi-Sugeno fuzzy system. Because
the neuro-net wor k usually de als with the numerical data,
choosing the T-S fuzzy system which has numerical
outputs is more convenient. Thus, we choose T-S fuzzy
system and a kind of neuro-network to approximate the
inverse mapping. For establishing the whole T-S fuzzy
system, it usually includes the identification of premise
parts and consequent parts. The BP neuro-network [14,
15] is used to identify the premise parameters of the T-S
fuzzy system. However, it is a kind of globally approx-
imating network which is easily to fall into the local mi-
nimums. Then, we adopt sub-clustering method [16] to
partition the input space of the premise variables by
which we try to compensate the disadvantages of BP
neuro-network. After the determination of the number
and shape of membership functions, the combination of
BP neuro-network and Least Square (LS) algorithm [17,
18] is used to identify the premise structure parameters
and the consequent parameters respectively.
0246810
0
0.5
1
1.5
2
0
0.5
1
1.5
Wind speed V (m/s)
R otor speed ω
r(rad/s)
Mechani cal power
P
m (MW)
Figure 2 . Wind turbine mechanical power curve.
Characteristic Data
ANFIS
r
ω
V
m
P
V
m
P
r
ω
Figure 3 . ANFIS for wind s peed estimation.

Y. HU ET AL.
406
The diagram of the algorithms is shown in Figure 4.
The first layer is the input la yer. And each node in the
second layer computes the membership degree of each
input value. The third layer and the forth layer complete
the fuzzy inference together. The third layer mainly deals
with the premise parts of the T-S fuzzy rules. Then the
consequent parts are tackled by the fourth layer. The fifth
layer is the output la yer which gives the nu merical val ues.
It is noted that the premise parameters are given by the
BP neuro -netwo rk a lgor it hm a nd the co nseq uent p a r ame-
ters are de termined by the LS algo rithm. At last, t he sen-
sorle ss wind speed e st imator is estab l i shed.
3.2. Optimum Setting Strategy
For the WTGS contro l, usually, we just co ncern the con-
trol method more. However, in the industrial process, the
correct setting values also matter a lot. As to realize the
maximum energy capture in the operation process of
WTGS, we need to provide the optimum setting values
of rotor speed
r
ω
and mechanical po wer
m
P
. In ge ner-
al case, we use the measured wind speed values to esti-
mate the optimum
m
P
. And the optimum
P
C
is used to
set the optimum
r
ω
. However, in the practice, the effi-
ciency of the wind energy conversion process may be
changed and the optimum
P
C
may have some drift with
time and varied environment. Thus, the setting values
determined by the initial status of WTGS need to be up-
dated according to the current op er ating status of WTGS.
Based on the SCADA system, we can establish the
profile mapping between meas
r
ω
,
meas
m
P
and
V
. Then,
using the estimated wind speed
ˆ
V
, we can search out
the primarily optimum output power opt
m
P and rotor
speed
opt
r
ω
corresponding to the current status, meas
r
ω
and meas
m
P.
The data acquisition process can be executed as fol-
low:
1) Regulate the rotor speed until it can be stable at a
fixed value. And then, store up the measured value of
()
meas
r
i ω
into t he SCADA system.
2) Measure the corresponding turbine power
( )
,
meas
m
P ij. Meanwhile, estimate the wind speed
( )
Vj
using the wind speed estimator and keep in storage.

1
x
n
x




y
Figur e 4. T rain mechanism of ANFIS.
3) Update the rotor speed to the next fixed value
( 1)
meas
ri
ω +.
4) Repeat step 2) and 3) until data of most operation
points has bee n co lle c te d in the SCADA system.
5) With the collected data, a profile mapping can be
established which ha s the same shape with Figure 2.
It is noted that the profile mapping is discrete. Then,
the curve fitting and other approaches can be used to
establish one with high accuracy. Combining the wind
speed estimator and the established profile mapping, we
can search out the primarily optimum setting values of
opt
m
P
and opt
r
ω
. However, the data are acquired in the
closed loop and the controller can’t accurately tracking
the setting values, so some compensation needs to be
given according to the performance precision of the con-
troller. The process can be shown in Figure 5.
Through the optimum setting strategy proposed above,
we can get the op timu m setti ng values,
opt
m
P
and opt
r
ω
.
4. Simulation
Taking the 1.5 MW DFIG-based variable speed variable
pitch WTGS for example, we mainly execute the ap-
proaches for fixed pitch variable speed operation mode.
And for other operation modes, the processes are very
the same. Using the data source in the Blade software,
the characteristic data of the wind turbine for some kind
of WTGS can be gotten by which we establish the wind
speed estimator using ANFIS. Then, using the data
source in the SCADA system, the profile mapping be-
t ween meas
r
ω
,
meas
m
P
and
V
can be established and the
optimum setting value of
r
ω
can be given for the fixed
pitch variable speed operation mode through searching.
The DFIG-based variable speed WTGS has the fol-
lowing parameters:
Rated power 1.5
m
P MW=; turbine radiu s
40Rm=
;
Rated wind speed
11.5 /V ms=
;
Optimum tip-speed-ratio
6.8
opt
λ =
.
After being trained, the estimated wind speed and the
error are shown in Figure 6 and Fig ure 7. T he optimum
rotor speed compared with the measured rotor speed is
shown in Figure 8. The optimum tip-speed-ratio and the
practical one is shown in Figure 9.
From Figure 6 and Figure 7, we can find that the es-
timated wind speed approximates accurately to the actual
wind speed which shows the effectiveness of the ANFIS
approach. Fro m Figure 8 and Figure 9, we can find that
Wind speed
estimator Profile mapping
Compensation to
controller performance
ˆ
V
,
opt opt
mr
P ω
,
opt opt
mr
P ω

,
opt opt
mr
P ω
Figure 5 . Optimum setting strategy.

Y. HU ET AL.
407
the actual rotor speed tracks closely to the optimum rotor
speed which shows the availability of the optimum set-
ting strategy.
050 100 150200
5
5.2
5.4
5.6
5.8
6
6.2
6.4
6.6
6.8
7
Time t(s)
Wind speed V /(m/s)
E st i m ated wi nd speed
A ct ual wi nd speed
Figure 6 . Compa r ison of estimated and act ual wind speed.
050 100 150 200
-0.25
-0. 2
-0.15
-0. 1
-0.05
0
0.05
0.1
0.15
0.2
0.25
Time t /(s)
Wi nd sp eed error
V
error
/(m/s)
Figure 7 . Error o f estimated and actua l wind speed.
050100 150200
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Time t /(s)
Rot or speed
ω
r
/(rad/s)
E s ti m ated opti m um rot or speed
A c tual rotor speed
Figure 8 . Comparison of opt im um and actual rot or speed.
050100 150200
0
2
4
6
8
10
12
Time t/(s)
Tip-speed-ratio λ/(rad/s)
E st i m ated ti p-speed-rat i o
OP ti m um t i p-speed-rat i o
Figure 9. Co mparison of optimum and actual tip-spee d-rati o.
5. Conclusions
As to further reduce the cost of the wind power genera-
tion, a kind of sensorle ss wind speed estimator is pro-
posed based on the ANFIS. Combining the wind speed
estimation and the special data-acquisition mechanism in
the SCADA system, a kind of optimum setting strategy is
established. According to the simulation, the results show
the effectiveness of the approaches. Especially, the ap-
proaches can not only be the optimum setting strategy
but also be the scheduling setting strategy. For the mo d-
ern wind power generation, the scheduling order form the
grid side needs to be considered. Based on the optimum
setting strategy, the way to give the setting values cor-
responding to the scheduling order can also be estab-
lished whic h will be studied in future .
6. Acknowled gements
This work is partially supported by the National Basic
Research Program of China (973 Program) (Grant No.
2012CB215203), the National Natural Science Founda-
tion of China (No. 51036002, No. 61203043) and the
Fundamental Research Funds for the Central Universi-
ties.
REFERENCES
[1] “Technology Roadmap: Wind Energy,” International
Energy Agency, 2009.
[2] “World Energy Outlook 2012,” International Energy
Agen cy, 2012.
http://www.iea.org/publications/freepublications/publicati
on/Wind_Roadmap.pdf
[3] S. Bhowmik, R. Spee, and J. H. R. Enslin, “Performance
Optimization for Doubly Fed Wind Power Generation
http://www.iea.org/publications/freepublications/publicati
on/English.pdf

Y. HU ET AL.
408
Systems,” IEEE Transactions on Industry Applications,
Vol. 35, N o. 4, 19 99, pp. 949-958.
[4] K. Tan and S. Islam, “Optimal Control Strategies in
Energy Conversion of PMSG Wind Turbine System
without Mechanical Sensors,” IEEE Transactions on
Energ y Conversion, Vol. 19, No. 2, 2004, pp. 392-399.
doi:10.1109/28.777205
[5] M. G. Simoes, B. K. Bose and R. J. Sp iegel, “Fuzzy Log-
ic Based Intelligent Control of a Variable Speed Cage
Machine Wind Generation Sys t e m, ” IEEE Transactions
on Pow er Electronics, Vol. 12, No. 1, 1997, pp. 87 -95.
doi:10.1109/TEC.2004.827038
[6] H. Li, K. L. Shi and P. G. McLaren, “Neural-Ne two r k -
Based Sensorless Maximum Wind Energy Capture with
Compensated Power Coefficient,” IEEE Transactions on
Industry Applications, Vol. 41, No. 6, 2005, pp. 1548-
1556.
doi:10.1109/63.554173
[7] V. Calderaro, V.Galdi, A.Piccolo and P.Siano, “A Fuzzy
Controller for Maximum Energy Extraction from Varia-
ble Speed Wind Power Generation Systems,” Electric
Power Systems Research, Vol. 78, No. 6, 2008, pp.
1109-1118.
doi:10.1109/TIA.2005.858282
[8] V. Galdi, A. Piccolo and P.Siano, “Exploiting Maximum
Energy from Variable Speed Wind Power Generation
Systems by Using an Adaptive Takagi-Sugeno-Kang
fuzzy Model,” Energy Conversion and Management, Vol.
50, No. 2, 20 09, pp. 413-421.
doi:10.1016/j.epsr.2007.09.004
[9] T. Takagi and M. Sugeno, “Fuzzy Identification of Sys-
tems and Its Applications to Modeling and Control,”
IEEE Transactions on Systems, Man and Cybernetics,
Vol. 15, N o. 1, 19 85, pp. 116-132.
doi:10.1016/j.enconman.2008.09.004
[10] J. J-SR, “ANFIS Adaptive-network-based Fuzzy Infe-
rence Systems,” IEEE Transactions on Systems, Man and
Cybernetics, Vol. 23, No. 3, 1993, pp. 665-85.
doi:10.1109/TSMC.1985.6313399
[11] X. N. Yu, F. Z. Cheng, L. L. Zhu and Y. J. Wang, “AN-
FIS Modeling Based on T-S Model and Its Application
for Thermal Process ,” Proceedings of the CSEE, Vol. 26,
No. 15, 20 06 , pp. 78-82.
doi:10.1109/21.256541
[12] A. D. Hansen, P. Sørensen, F. Iov and F. Blaabjerg,
“Control of Variable Speed Wind Turbines with Doub-
ly-fed Induction Generat ors,” Wind Engineering, Vol. 4,
No. 28, 20 04 , pp. 411-432.
doi:10.1260/0309524042886441
[13] S. Heier and R. Waddington, “Grid Integration of Wind
Energy Conversion Sys t e m, ” 2nd Edition, Wiley, West
Sussex, 2006.
[14] S. S. Haykin, “Neural networks: A Comprehensive
Foundation,” 2nd Edition, Prentice Hall, USA, 1999.
[15] M. T. Hagan, H. B. Demuth and M. Beale, “Neural Net-
work Design,” 1st Edition, Thomson Learning, Boston,
1996.
[16] R. Y. Ronald and P. F. Dimitar, “Approximate Clustering
via the Mountain Method,” IEEE Transactions on Sys-
tems, Man and Cybernetics, Vol. 24, No. 8, 1994,
pp.1279-1284.
[17] J. Wolberg, “Data Analysis Using the Method of Least
Squares: Extracting the Most Information from Experi-
ments,” 1st Edition, Springer, Germany, 2005.
doi:10.1109/21.299710
[18] T. Strutz, “Data Fitting and Uncertainty (a practical in-
troduction to Weighted Least Squares and Beyond),” 1st
Edition, Vieweg Teubner, Germany, 2010.