Energy 149 (2018) 684e696

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Effect investigation of yaw on wind turbine performance based on

SCADA data
Juchuan Dai a, *, Xin Yang a, Wei Hu a, Li Wen a, Yayi Tan b
a
School of Mechanical Engineering, Hunan University of Science and Technology, Xiangtan, 411201, China
b
XEMC Windpower Co., Ltd., Xiangtan, 411000, China

a r t i c l e i n f o a b s t r a c t

Article history: Due to the time-varying wind direction, yaw operation of wind turbines is a common state. Under yaw,
Received 31 May 2017 the aerodynamic behavior of wind turbines is complicated, and also causes complex energy capture
Received in revised form performance. To clarify some vague knowledge, a detailed investigation of yaw effect based on SCADA
25 January 2018
data is carried out. Firstly, a yaw coefficient definition and its specific calculation method are presented.
Accepted 12 February 2018
Available online 14 February 2018
Furthermore, to analyze the energy capture mechanism, the loss factor of energy capture (power coef-
ficient) is subdivided into aerodynamic loss factor and inertia loss factor, which means both the effects of
aerodynamic characteristic and mechanical inertia on energy capture are considered. According to this
Keywords:
Wind turbines
understanding, power coefficient is expanded as a function of four factors. Then, a single-valued pro-
Yaw operation cessing method is employed to investigate the relationship between wind speed and output power.
SCADA data Subsequently, relationship between yaw coefficient and output power is investigated. Comparative in-
Power coefficient vestigations are carried out by two different methods, that is, least square fitting (LSF) method and kernel
density estimation (KDE) method. Also, characteristics of power coefficient and rotor torque under yaw
are investigated. Effect laws of yaw coefficient on wind turbine power, power coefficient, and rotor
torque are obtained. Finally, the relationship between the internal control mechanism and external
output of wind turbines is discussed.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction these ask for further understanding of the operating mechanism

Renewable energy has great potential to enhance the diversity
of the energy supply market and to ensure long-term security [1].
Especially, wind power is an abundant, renewable, and clean en-
ergy source has been the world's fastest-growing source of elec-
tricity [2,3]. The total installed capacity of wind power has reached
486 GW in 2016 based on world wind energy association (WWEA)
data; all wind turbines installed can provide about 5% of the world's
electricity demand. As a development trend of wind turbines, the
unit capacity is increasing, for example, Vestas has launched V164-
8 MW wind turbine in 2014. On the other hand, with available land
wind farm dwindling, the future wind farm will have to expand
from onshore to offshore [4]. Subsequently, various challenges are
more severe in an offshore wind farm. Therefore, how to further
improve the efficiency of wind energy capture, and improve the
reliability of wind turbines has become the industry focus. All of
* Corresponding author.
E-mail address: daijuchuan@163.com (J. Dai).
and performance evaluation on wind turbines in a complex service
environment. However, in the process of energy converting, multi-
physical processes are involved, energy capture, conversion, and
loss mechanism are complex, and understanding of many phe-
nomena need numerical simulation or experimental analysis such
as dynamic stall [5], wake effect [6,7]. Furthermore, due to the
random nature of the wind resource, the rotor aerodynamic effects
become even more challenging. The design of systems for con-
verting wind energy into electric energy, energy capture mecha-
nism and efficiency of wind turbines become of prime concern.
Support vector machine (SVM) [8], fuzzy logic control (FLC) [9],
sliding mode control (SMC) [10] and some other advanced methods
such as perturbation-observation-method [11] have been
employed to realize maximum power point tracking (MPPT).
For a wind turbine with fixed mechanical structure and control
strategy, the effective wind speed and wind direction acting on the
wind rotor is the basic effect factor on energy capture and aero-
dynamic load. Yaw will, therefore, lead to a change in output power
and aerodynamic load [12,13]. Yaw means that wind direction is not

https://doi.org/10.1016/j.energy.2018.02.059
0360-5442/© 2018 Elsevier Ltd. All rights reserved.
 J. Dai et al. / Energy 149 (2018) 684e696
parallel to the axis of the rotor for horizontal axis wind turbines. So,
in the actual operation of wind turbines, yaw effect should not be
ignored. Since yaw affects the energy capture, large wind turbines
are equipped with yaw mechanism, which has the ability to adjust
the wind rotor axis to wind direction (active control of yaw [14]). An
accurate understanding of the correct yaw-alignment of the turbine
relative to the incoming wind vector is crucial [15]. However, the
yaw mechanism will run only when yaw angle is greater than a
predetermined critical value because frequent yaw movement will
not only increase the burden of yaw mechanism (affecting life) but
also increase vibration amplitude (affecting stability). The yaw state
of wind turbines is quite a common occurrence, investigation on
wind turbine yaw is particularly necessary. In fact, much research
on yaw has been carried out in the past years. In the field of yaw
aerodynamics, the current trend is towards the detailed under-
standing of flow field characteristic [16], some useful methods have
also been employed, for example, stereoscopic particle image
velocimetry (SPIV) [17]. Investigations have shown that wake flow
may deflect in the near-wake region and the wake trajectory
downwind may change [18e20], wake intensity becomes slightly
smaller [21], trailing vortex circulation varies as a function of rotor
phase angle and yaw angle [22], aerodynamics will be more com-
plex in oscillating yaw [23]. As a result, changing the yaw angle will
have a significant impact on the unsteady aerodynamic loads [24].
In addition, experimental investigation for yawed conditions has
clearly shown how vorticity is swept away from the blade [25], also
shown that operating of the upstream turbine in yaw condition has
an effect on the downstream turbine [26], downwind wind turbines
have better yaw stability than upwind wind turbines [27]. To
investigate the structural dynamic behavior of wind turbines in
yaw, University of Utah has developed a software named as Yaw-
Dyn, it can be used to analyze both fixed and free-yaw behavior
[28]; vibration response to yaw motion based on testing signals is
also included [29]. An investigation based on blade beam model
shows aero-elastic damping under yaw can be reduced to
approximately 33% in the worst case [30]. In the field of yaw con-
trol, CFD simulation has shown that yaw setting optimization can
increase the energy production in a wind plant [31]; a study on the
Princess Amalia Wind Park showed that significant increases in
annual energy production by employing wake steering with yaw
control [32]. On the basis of the traditional yaw control based on
wind direction measurement, yaw control methods based on the
output power [33] and rotor speed [34] are also presented;
advanced wind LIDAR technology [35,36] and some advanced
control theories have been introduced into the application of yaw
control, such as FLC [37] and k-NN(k-nearest neighbor classifier)
[38]. Besides theoretical and experimental analyzing, SCADA data
investigating is a new trend to analyze wind turbine operation
mechanism, for example, SCADA data recorded for six months has
been used to investigate wake impacts on wind turbine perfor-
mance and yaw alignment [39].
Although much research on wind turbine yaw has been carried
out from different angles, some problems are still not clear in
existing research due to the yaw complexity, and need to be further
investigated, including (1) wind energy capture mechanism under
yaw, (2) energy capture characteristics under the joint action of yaw
and internal control, (3) quantitative influence of yaw on wind
turbine performance using field data. For example, though power
production in yawed flow is investigated in Ref. [20], the function
description between yaw angle and power production needs to be
further extended. Meanwhile, the coupling influence investigation
of external yaw and internal control strategy on the state parameter
is also expected. In view of this, the purpose of this paper is to do an
investigation on yaw effect of horizontal-axis direct drive wind
turbines based on SCADA data. The remaining part is organized as
685
follows. In Section 2, the yaw angle and yaw coefficient definition
are introduced which is the basis of investigating the yaw effect. In
Section 3, energy capture mechanism under yaw is analyzed. The
effects of not only aerodynamic characteristics but also mechanical
inertia of the wind rotor on energy capture are considered. For this
reason, the loss factor of energy capture is subdivided into aero-
dynamic loss factor and inertia loss factor. In Section 4, the specific
calculation method of yaw coefficient is presented; then in Section
5, the detailed investigation is carried out including relationship
investigation between wind speed, yaw coefficient and output
power, and power coefficient investigation under yaw. Torque
characteristic investigation under yaw is presented in Section 6 and
the conclusions are given in Section 7.
2. Yaw angle and yaw coefficient
Fig. 1 shows the yaw operation state of a Horizontal Axis Wind
Turbine (HAWT). In Fig. 1(a), the yaw anglegis the angle difference
between the wind direction and wind rotor axis. If yaw exists, air
flow will appear skewed after flowing through the wind rotor.
Wake skew angle is usually defined as the difference between rotor
plane and incoming flow [40] (Fig. 1(b)). Assuming fv is wind di-
rection angle which indicates the angle between wind direction
and the datum direction, fs is rotor axis angle which indicates the
angle between rotor axis and the datum direction, the relationships
between yaw angle, wake skew angle, wind direction angle, and
rotor axis angle can be written as [40]
8
< g ¼ fv  fs
>
> sing  atanðc=2Þ (1)
: tan c ¼
cos g  a
where, a is axial induction factor; c is wake skew angle.
Though wake skew angle c is difficult to determine, yaw angle
data is directly available in SCADA data, which provides an
analytical approach for wind turbine performance under yaw. The
original data format of SCADA system in a wind farm is shown in
Table 1. The sampling frequency of the operating parameters is
1 Hz, and the yaw angle is given by the mean of five seconds. The
total number of sampling points is 256753 for this investigation.
The mean environment temperature is 26.5 Cduring the sampling
period.
For the convenience of the subsequent analysis, a yaw coeffi-
cient expression is proposed, which is defined as the ratio of the
current yaw angle to the yaw angle base, i.e.
gðKTÞ
x¼ (2)
gb
where, x is the yaw coefficient;gðKTÞ is the current yaw angle, Tis
sample period, K is sampling number;gb is the yaw angle base
determined according to the special yaw control characteristic.
3. Energy capture mechanism under yaw
Under yaw, the effective value of inflow wind speed acting on
the wind blade will change, which will affect the energy capture
efficiency of the wind rotor. To analyze the energy capture mech-
anism under yaw, an energy flow structure is first given in Fig. 2 to
explain a general energy flow scheme existing in a wind turbine.
The inflow wind is the system input and output power of the
generator is defined as the system output. Air kinetic energy is
captured by wind rotor and converted to mechanical energy; its
value is embodied in the product of rotor torque and rotor speed.
686 J. Dai et al. / Energy 149 (2018) 684e696

(a) (b)
Fig. 1. Yaw operation of wind turbines.

Table 1
Original style of SCADA data.

Sample Record time mm/dd Rotor speed Yaw angle (5s mean) Nacelle vibration Wind speed Wind direction Output power $$$ Environment
number hh:mm (rpm) (deg) (m/s2) (m/s) (deg) (kW) temperature ð CÞ

1 08/02 14:20 12.78 2.1 0.00 6.1 280 654 $$$ 32.7
2 08/02 14:20 12.78 2.0 0.01 6.5 280 645 $$$ 32.7
3 08/02 14:20 12.80 4.3 0.00 6.0 281 646 $$$ 32.7
4 08/02 14:20 12.81 3.0 0.02 6.4 280 642 $$$ 32.7
$$$ $$$ $$$ $$$ $$$ $$$ $$$ $$$ $$$ $$$
60 08/02 14:20 14.07 3.7 0.03 6.9 288 860 $$$ 32.7
… … … … … … … … … …
256753 08/05 18:39 13.06 3.7 0.01 7.8 120 687 $$$ 26.3

Wind rotor Transmission shaft Generator

Input Output
(mechanical energy) (mechanical energy) (electrical energy)
(inflow wind) (electrical energy)
(thermal energy) (thermal energy) (thermal energy)

MF EF EF EF
EF Torque Torque Voltage
Rotor Rotor Current
speed speed

IF IF IF

Wind speed EF=energy flow

Wind direction MF=material flow
IF=information flow
IF Information detection, processing and control system

Fig. 2. Energy flow scheme arising in a wind turbine.

Then, the mechanical energy of the wind rotor is transferred to the
generator rotor through the transmission shaft. Inside the gener-
ator, mechanical energy is converted into electrical energy based on
the electromechanical coupling effect and the electromagnetic
coupling effect. Since the process of converting mechanical energy
into electrical energy is the physical process of the closed coil
cutting rotor magnetic field, the change rate of generator rotor
speed determines the change speed of output power of the
generator. Furthermore, due to the mechanical inertia of the wind
rotor, it has a certain time delay for electric energy to response the
change of wind and rotor speed. From the control point of view, the
process of converting wind energy into electrical energy can be
considered as an inertial link.
In wind power industry, power coefficientCp is used to describe
the coefficient of wind energy capture. It can be defined as the ratio
of the actual power produced by the wind rotor divided by the wind
power flowing into the wind rotor at specific wind speed. It is
usually considered that effect factors of power coefficient include
tip speed ratio l and blade pitch angleb, that is
Cp ¼ f ðl; bÞ (3)
 J. Dai et al. / Energy 149 (2018) 684e696 687

For example, a relationship betweenCp , l and bis given as
Cp ðl; bÞ ¼ 0:22ð116=li  0:4b  5Þe12:5=li in [41,42], where
l ¼ uR=v (uis rotor speed, R is the radius of the wind rotor, v is wind
speed), li ¼ 1=ððl þ 0:08bÞ1  0:035=ðb3 þ 1ÞÞ.
However, the effect of yaw on power coefficient is ignored in the
above expressions. In addition, the effect of the change rate of the
input wind energy on power coefficient is also ignored. In this
paper, both the two factors are considered to analyze power coef-
ficient and extend the understanding of power coefficient. Then, Eq.
(3) can be rewritten as
Cp ¼ f ðl; b; g; uv Þ
where, uv is the change rate of the input wind energy which may be
caused by wind speed change or wind direction change.
Furthermore, based on Eq. (4), power expression can be written
as
P ¼ 0:5Cp ðl; b; g; uv ÞrpR2 v3
where, P is power produced by the wind rotor;r is air density.
Obviously, the value of the coefficientCp is less than 1. The well-
known Betz's law indicates that the maximum power coefficient is
16/27 (0.593) [43]. The factor 16/27 (0.593) is also known as Betz's
coefficient. Betz's law mainly concerns the aerodynamic charac-
teristics of the wind rotor, and is deduced from the principle of
mass and momentum conservation of the airflow flowing through
the idealized “actuator disk”. In fact, there are two reasons causing
the incomplete capture of the air kinetic energy, namely, (1) aero-
dynamic characteristics of the wind rotor, (2) frequency response
characteristics of energy variation due to the wind rotor inertia.
When the energy change frequency inputting the wind rotor is too
high, the response output is close to zero.
Here, the symbol Ca is used to indicate the aerodynamic loss
factor of power capture, CJ is the inertia loss factor. The corre-
sponding expressions are
8
< Ca ¼ f ðl; b; gÞ
C ¼ f ðuv ; tÞ
: J
Cp ¼ Ca $CJ
where, tis inertia time constant of wind rotor.
According to the characteristic of inertia link, the expression of
CJ can be also written as
1 gation. Considering the randomness of the wind and the uncer-
CJ ¼ f ðuv ; tÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 þ ðuv tÞ2
Based on Eqs. (5)e(7), the power expression can be further
obtained as
filter out the high-frequency component of yaw angle rate through
a kind of filtering algorithm. The actual time-varying curve of yaw
angle in Fig. 3 extracting from SCADA data has also shown that
high-frequency component may be included.
A discrete two-order low pass filter is employed to process Eq.
(2) and filter out the high-frequency component of yaw angle rate.
Its state equation is written as

xðn þ 1Þ ¼ AxðnÞ þ BuðnÞ
(9)
yðnÞ ¼ CxðnÞ þ DuðnÞ
where, u is filter input; y is filter output; x is state variation; A, B, C,
(4)
D are matrix coefficients.
According to the above analysis, the calculation diagram of yaw
coefficient is shown in Fig. 4. The yaw angle base gb needed for the
following analysis is selected according to yaw control character-
istic of the turbine (for example, 20 ).
The key for implementing the filtering algorithm is to find the
low-pass cut-off frequency of the filter. Through the analysis of the
(5)
inertia loss factorCJ , it is known that the inertia time constant of the
wind rotor is the first parameter needed to be determined.
In Fig. 4, matrix coefficient A, B, C, D depend on the filter cut-off
frequencyuc. If uc ¼ 1=t ¼ 0.1 (the inertia time constant of
 
0:9991 0:0003946
the wind rotor is 10s), they are A ¼ ;
0:0009996 1
   7

; C ¼ 1:973  10
0:9996
B ¼ ; D ¼ 9:865  108
0:0004998 0:0003948
5. Energy capture characteristic investigation under yaw
5.1. Turbine parameters and wind characteristics
Parameters of the investigated wind turbine are shown in
Table 2. The turbine has three blades and is installed in a mountain
wind farm in South China. It is a direct drive type wind turbine
which means its wind rotor is connected directly to the generator
rotor and full power converter is also used. Its rated power is
(6) 2000 kW, and the rated wind speed is 10.5 m/s. Variable-speed and
variable-pitch control are used in this turbine. Its rated rotor speed
is 17 rpm but it can operate in other rotor speed due to the variable-
speed function. The turbine adopts MPPT control below the rated
wind speed and pitch control above the rated wind speed.
Wind is the source of energy and the external input of the wind
turbine. The understanding of the wind is the basis of the investi-
(7)
tainty caused by the sensor test error, finding the wind probability
distribution is helpful to the investigation. Since the frequency
distribution characteristic of wind speed in the wind farm is not
known, the non-parametric Kernel Density Estimation (KDE)

1
P ¼ 0:5Cp rpR2 v3 ¼ 0:5 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiCa ðl; b; gÞrpR2 v3 (8) 40
1 þ ðuv tÞ2
20
Yaw angle /deg

0
4. Calculation method of yaw coefficient

According to Eq. (7), if wind energy frequency inputting the -20

wind rotor changes too fast, there will be no energy response
output from the wind rotor. On the other hand, because of the error -40
of the sensing test, there are some high-frequency noise signals in 0 200 400 600 800 1000
time t/s
the wind speed data. So, when analyzing the effect of yaw on en-
ergy capture characteristics of wind turbines, it would be better to Fig. 3. Yaw angle time-varying curve.
688 J. Dai et al. / Energy 149 (2018) 684e696

B11 C11 1
1 Z -1 1
B21 C21 B
yaw angle yaw coefficient
(kT) x(n 1) Ax(n) Bu (n) A11 A12
A21 A22
y (n) Cx(n) Du (n)

Fig. 4. Calculation diagram of yaw coefficient.

Table 2
Wind turbine parameter.

Parameter Values Parameter Values Parameter Values Parameter Values

Rated power (kW) 2000 Blade number 3 Cut-in wind speed (m/s) 3 Control model Variable-speed and variable-pitch
Rated wind speed (m/s) 10.5 Rotor diameter 95.9 Cut-out wind speed (m/s) 25 Rated rotor speed (rpm) 17

method [44e46] is employed to estimate its distribution which 5.2. Relationship between wind speed and output power
does not rely on a priori knowledge. KDE method is a very useful
method and will also be used in the following Section 5.2 and From system analysis point of view, external wind is regarded as
Section 5.3. The expression of KDE method is the system input; output power of the generator is regarded as the
system output. Then the problem is to analyze the relationship
between system input and system output. Since the change law of
8   internal variables, which include rotor torque and rotor speed, can
>
> 1 X n
x  Xi
> f
> h ðxÞ ¼ K be determined by the internal mechanism for the wind turbine
>
> nh i¼1 h
>
> with fixed mechanical structure and control strategy, the effects of
>
<
KðxÞ  0 internal variables on system output are not considered in the
(10)
>
> analysis process.
>
> Zþ∞
>
> In order to obtain a more precise investigation result, it would
>
> KðxÞdx ¼ 1
>
: be better to divide the yaw coefficient range of [0,1] into several
∞ sub-ranges. A typical value is 3, namely, [0, 0.3], [0.3, 0.6] and [0.6,
1]. Subsequently, the scattered points of the wind and output po-
where, X i ði ¼ 1; 2; …; nÞ is the wind speed sample extracted from wer of the generator corresponding to the three regions of yaw
the SCADA system; fh ðxÞis kernel density function; K ($) is kernel coefficient are given in Fig. 6 (a), (b) and (c) by extracting SCADA
function; h is a parameter called bandwidth. data. The wind speed data in Fig. 6 has been filtered using the
To implement Eq. (10), Gaussian function is employed which is similar method in Fig. 4 to eliminate the high-frequency compo-
one of the most common choices for function K($) and its expres- nent. It can be seen from the figure that these scattered points are
pffiffiffiffiffi
sion is KðuÞ ¼ e0:5u = 2p .
2
mainly distributed in low wind speed region. This is because the
In the sampling period of the employed SCADA data, wind speed wind turbine mainly runs below the rated wind speed, which is
probability distribution and wind direction distribution are shown dependent on wind resource characteristics in the selected wind
in Fig. 5. Since low wind speed dominates in this mountain wind farm. In the three yaw coefficient regions, the region with the
farm, the wind speed with the maximum probability is 4.2 m/s in smallest yaw coefficient has the largest number of scattered points
Fig. 5(a). Wind direction in Fig. 5(b) is shown by using a wind rose. (in the test time period), the region with the largest yaw coefficient
There are two main wind directions flowing through the turbine. has the minimum number of scattered points, which also shows

90 8000
0.25 120 60
Frequency histogram 6000
0.2 KDE
150 4000 30
2000
0.15
f(x)

180 0
0.1

0.05 210 330

0 240 300
0 2 4 6 8 10 12 14 16 270
(a) Wind speed v/(m/s) (b)

Fig. 5. Wind speed and wind direction distribution.

 J. Dai et al. / Energy 149 (2018) 684e696 689

2500 2500

2000 2000

Power P/kW
Power P/kW

1500 1500

1000 1000

500 500

00 2 4 6 8 10 12 14 0
0 2 4 6 8 10 12 14
(a) Wind speed v/(m/s) (0< <0.3) (b) Wind speed v/(m/s) (0.3 <0.6)

2500 2200

2000 1800
0 0.3
Power P/kW

0.3 0.6

Power P/kW
1500 1400
0.6 1
1000
1000
600
500
200
00 2 4 6 8 10 12 14 3 4 5 6 7 8 9 10 11 12 13 14
(c) Wind speed v/(m/s) (0.6< 1) (d) Wind speed v/(m/s)

Fig. 6. Relationship between wind and power under yaw.

that the wind turbine mainly runs in the small yaw coefficient
region.
However, it is difficult to determine the effect law of yaw on
output power directly from scattered points (Fig. 6 (a), (b) and (c))
because each wind speed value corresponds to a number of output
power values. Therefore, multi-values of output power corre-
sponding to a wind speed value should be processed using a single-
valued processing method. Firstly, discrete wind speed points are
selected from the start wind speed (3 m/s) to the maximum wind
speed (14 m/s) in the test period according to the interval of 0.5 m/s.
So, for a specific discrete wind speedvk , the corresponding output
power set isPk ¼ ½Pk1 ;Pk2 ;…;Pkn . It should be pointed out that zero
power values have been eliminated in the output power set
because it means the wind turbine is in a shutdown state or the
sensing test system is in error state. Then, for a wind speed vk , the
power value with the maximum probability distribution can be
obtained by using nonparametric kernel density estimation (KDE)
method illustrated in Eq. (10). Then, this power value is selected to
be the final power value corresponding to the wind speedvk , that is
 role on output power in this region. In more detail, when the inflow
fh ðPkm Þjv¼vk ¼ max fh ðPki Þjv¼vk
where, i ¼ 1; 2; ,,,n; 1 < m < n.
Eq. (11) means that the output power value Pkm has the
maximum probability corresponding to wind speedvk .
The relationship curve of wind and output power after single
valued processing is shown in Fig. 6(d). As illustrated in the figure,
below the rated wind speed, the change of yaw coefficient has a
large effect on the output power. When yaw coefficientxis in the
region of [0.6, 1], the output power is smaller than that whenxis in
the regions of [0, 0.3] and [0.3, 0.6]. Though the output power
whenxis in the region of [0, 0.3] is closer to that whenxis in the
region of [0.3, 0.6], in most cases, the output power of the generator
is gradually decreasing with the increase of the yaw coefficient. For
example, output powers in yaw coefficient regions of [0, 0.3], [0.3,
0.6] and [0.6, 1] are 550 kW, 511 kW and 345 kW corresponding to
0.03
frequency histogram
0.025 kernel density estimation curve
0.02
0.015
0.01
0.005
00 50 100 150 200
Output power P/kW
Fig. 7. Kernel density estimation of output power.
the wind speed of 6 m/s, respectively; 1180 kW, 1120 kW and
1038 kW corresponding to the wind speed of 8 m/s. Nevertheless,
above the rated wind speed, the change of yaw coefficient has little
effect on output power since pitch control has a stronger leading
(11) wind speed is below the rated wind speed, it means that the wind
energy cannot provide enough energy to the wind turbine. At the
same time, yaw will affect the effective wind speed acting on the
wind rotor. So, the change of yaw coefficient has a large effect on
the output power. On the contrary, when the inflow wind speed is
above the rated wind speed, it means that the wind energy can
provide enough energy. For this scenario, even yaw will affect the
effective wind speed acting on the wind rotor, the wind turbine can
capture energy near the rated power by employing pitch control
which is based on power feedback.
To give a demonstration, Fig. 7 shows the frequency histogram
and kernel density estimation curve of output power when the yaw
coefficient is in the range of [0, 0.3] corresponding to the wind
speed is 3 m/s.
690 J. Dai et al. / Energy 149 (2018) 684e696

5.3. Relationship between yaw coefficient and output power undetermined polynomials;vk is the selected wind speed value.
Based on the least squares method, substituting yaw coefficients
By extracting SCADA data, the scatter relationship of yaw coef- and the corresponding power values (under different wind speed in
ficient and output power of the generator can be directly given as Fig. 8 into Eq. (12), the specific values of parameters a0 ; …; an can be
illustrated in Fig. 8. It is difficult to reflect directly the effect char- obtained as
acteristic of yaw on output power from scattered point distribution
because each yaw coefficient value corresponds to a number of l 
X 2
output power values. Therefore, multi-values of output power Fða0 ; a1 ; …an Þ ¼ Pi  f ðxi Þjv¼vk (13)
corresponding to a yaw coefficient value should be processed using i¼1
a single-valued processing method. In the following section, two
kinds of single-valued processing methods are employed to solve where, l is the number of SCADA data points in the figure.
this problem. One is the least square fitting method which uses a Though the higher order will have a higher accuracy in Eq. (12),
polynomial to fit the functional relationship of yaw coefficient and a first-degree polynomial has the advantage to reflect the overall

output power based on global data in each figure. The other is KDE trend more directly. Thus, the expression of fp ðxÞ
under
v¼vk
method which seeks the output power value with the maximum
different wind speed conditions is simplified into Eq. (14) and the
probability corresponding to each special yaw coefficient value.
corresponding parameter values ofav andbv are in Table 3.

5.3.1. Investigation using least square fitting (LSF) method fp ðxÞ

¼ av x þ bv (14)
v¼vk
Under a certain wind speed, a generalized polynomial regres-
sion method to describe the relationship between yaw coefficient
where, av andbv are undetermined coefficient.
and output power of the generator can be written as
By combining the scattered point trend and the fitting line trend

in Fig. 8, it can be seen that if the wind speed keeps constant but

P ¼ fp ðxÞ
¼ a0 þ a1 x1 þ a2 x2 þ … þ an xn (12) yaw coefficient increases, the corresponding output power de-
v¼vk
creases below the rated wind speed. But above the rated wind
where,a0 ; …; an are undetermined coefficients; n is the order of speed, yaw coefficient has less effect on output power. In other

250 500 800

P 14 75 v = 3m/s P 23 160 v = 4m/s P 86 340 v = 5m/s
200 400 600
Power P/kW
Power P/kW

Power P/kW

150 300
400
100 200
100 200
50
0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(a) Yaw coefficient (b) Yaw coefficient (c) Yaw coefficient

1200 1500 2500

P 160 590 v = 6m/s P 250 960 v = 7m/s P 370 1300 v = 8m/s
1000 2000
Power P/kW
Power P/kW

Power P/kW

800 1000
1500
600
1000
400 500
200 500
0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(d) Yaw coefficient (e) Yaw coefficient (f) Yaw coefficient

2500 2500 2500 P 160 2200

P 310 1900 v = 9m/s P 250 2100 v = 10m/s v = 11m/s
2000 2000 2000
Power P/kW

Power P/kW
Power P/kW

1500 1500 1500

1000 1000 1000
500 500 500
0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(g) Yaw coefficient (h) Yaw coefficient (i) Yaw coefficient

2500 2500 2500

2000 2000 2000
Power P/kW
Power P/kW
Power P/kW

v = 12m/s P 26 2100 v = 13m/s P 5.5 2100 v = 14m/s

1500 P 1.4 2100 1500 1500

1000 1000 1000

500 500 500

0 0.2 0.4 0.6 0.8 1 1.2 0 0

0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(j) Yaw coefficient (k) Yaw coefficient (l) Yaw coefficient

Fig. 8. Data fitting analysis of yaw coefficient and output power.

 J. Dai et al. / Energy 149 (2018) 684e696 691

Table 3
Parameter values ofav and.bv

v 3 m/s 4 m/s 5 m/s 6 m/s 7 m/s 8 m/s 9 m/s 10 m/s 11 m/s 12 m/s 13 m/s 14 m/s

av 14 23 86 160 250 370 310 250 160 1.4 26 5.5
bv 75 160 340 590 960 1300 1900 2100 220 2100 2100 2100

words, when inflow wind speed is close to the rated wind speed
(10.5 m/s) or above the rated wind speed, the fitting line gradually
becomes flat (similar to the horizontal line). This shows that the
effect of yaw on output power has become weaker because pitch
control has a stronger effect on output power than yaw in this
region.
5.3.2. Investigation using KDE method
For a certain wind speed vk and a yaw coefficientxj , the corre-
sponding output power set is assumed to be Pðk;jÞ ¼ ½Pðk;jÞ1 ;Pðk;jÞ2 ;…;
Pðk;jÞn . Then by employing KDE (kernel density estimation) method,
a power value Pðk;jÞ is selected from Pðk;jÞ to be the final value, which
has the maximum probability corresponding to yaw coefficient xj
under wind speedvk , that is

 
 k0 u3
v1 ¼ þ vd ¼ k1 u
fh Pðk;jÞm
¼ max fh Pðk;jÞi
(15) 2rSv2d
ðv¼vk ;x¼xj Þ ðv¼vk ;x¼xj Þ
KDE curves, which reflect the relationship between yaw coeffi-
cient and output power of the generator under different wind
speed, are shown in Fig. 9.
When the wind speed is 3 m/s and 4 m/s (Fig. 9(a) and (b)), the
yaw effect on the output power of the generator has no significant
upward or downward trend, but fluctuates along the horizontal
line. In the wind speed range of [5 m/s, 8 m/s], though fluctuations
and discontinuities appear in KDE curves, the general trend of KDE
curves is tending to decline with the increase of yaw coefficient.
When wind speed is 9 m/s, KDE curve shows an irregular change.
Correspondingly, the scattered points in Fig. 9(g) are quite sparse.
When wind speed is 10 m/s or above 10 m/s, the KDE curve be-
comes flat.
5.4. Power coefficient under yaw
As illustrated in Fig. 8, yaw has little effect on the output power
of the generator above the rated wind speed. Using the above
analysis results in Section 4.4, equation set (16) can be obtained. For
a wind turbine with fixed mechanical structure and control strat-
egy, parameters av and bv in equation set (16) are both relative to
inflow wind speed and will change with the change of inflow wind
speed. Obviously, the first equation in this equation set is obtained
based on the physical mechanism of the wind turbine, the second
equation is deduced from SCADA data fitting. The two have
different angles to describe power characteristics. Meanwhile, the
relationship between yaw coefficient and power coefficient is
constructed.
 5.5. Discussion of the energy capture characteristic
P ¼ 0:5Cp rpR2 v3
P ¼ av x þ bv
From equation set (16), Eq. (17) is deduced as
av x þ bv 3
Cp ¼ v (17)
0:5rpR2
It should be pointed out that since the value of inflow wind
speed in the SCADA system is sampled from the anemometer fixed
on the nacelle, its value is smaller than the actual wind speed. In
more detail, the energy captured by a wind turbine is extracted
from the kinetic energy of the air, so that when the flow wind
passes through the wind rotor, the kinetic energy of the air will
decrease. That is to say, the wind speed will decrease after passing
the wind rotor. The anemometer is installed behind the wind rotor,
so the measured wind speed will be smaller than that in front of the
wind rotor. If this value is used to calculate the power coefficientCp ,
the results may be larger than 0.593, which is the theoretical limit
value given by Betz in 1919. For example, when wind speed is 3 m/s
and yaw coefficient is zero, power coefficient calculated from Eq.
(17) is 0.64 (air densityr ¼ 1:2kg=m3 ). To solve this issue, an
approximate correction equation is given in Ref. [47] illustrated in
Eq. (18), which has given the relation function between the
measured wind speed and the actual wind speed in MPPT region.
(18)
where, v1 is the upstream wind speed in front of the wind rotor;vd is
the wind speed flowing through the wind rotor, which can be
replaced by the nacelle measurement wind speed; S is sweep area
of the wind rotor; k0 andk1 are undetermined fitting coefficients
introduced in Ref. [47].
Similarly, according to the method presented in Ref. [47], and
then selecting SCADA data in MPPT region (wind speed: [3 m/s,
7 m/s]), Eq. (17) is rewritten as
av x þ bv
Cp ¼ ðavÞ3 (19)
0:5rpR2
where, a is correction coefficient of wind speed; here a value 1.125
is adopted.
Fig. 10(a) shows the effect relationship lines of yaw coefficient
on power coefficient under different wind speed conditions based
on LSF method in Section 4.4.1; Fig. 10(b) is the corresponding two-
dimensional curved surface. In MPPT region, yaw coefficient has an
obvious effect on power coefficient. The greater the yaw coefficient,
the smaller the power coefficient under the same wind speed. For
example, when wind speed is 5 m/s and yaw coefficient is zero, the
corresponding power coefficient is 0.44; while yaw coefficient in-
creases to 1.2, power coefficient decreases to 0.31, which shows a
30% decreasing amplitude. Furthermore, under different wind
speed the effect characteristics of yaw on power coefficient are
different, namely, the slopes of the relationship lines in Fig. 10(a)
are different.
(16)
To quantify the effect of yaw on output power under a certain
wind speed, a definition of yaw effect is given as
Px¼0  Px¼m
Gv¼vk ¼  100% (20)
Px¼0
where, Gv¼vk is the defined yaw effect index which reflects the
relative variation of output power when the yaw coefficient in-
creases from zero to m under a certain wind speed; m is equal to 1.2
692 J. Dai et al. / Energy 149 (2018) 684e696

250 500 800

v = 3m/s v = 4m/s v = 5m/s
200 400 600

Power P/kW
Power P/kW

Power P/kW
150 300
400
100 200
100 200
50
0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(a) Yaw coefficient (b) Yaw coefficient (c) Yaw coefficient

1200 1500 2500

v = 6m/s v = 7m/s v = 8m/s
1000 2000

Power P/kW
Power P/kW

Power P/kW
800 1000
1500
600
1000
400 500
200 500
0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(d) Yaw coefficient (e) Yaw coefficient (f) Yaw coefficient

2500 2500 2500

v = 9m/s v = 10m/s v = 11m/s
2000 2000 2000
Power P/kW

Power P/kW
Power P/kW

1500 1500 1500

1000 1000 1000
500 500 500
0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(g) Yaw coefficient (h) Yaw coefficient (i) Yaw coefficient

2500 2500 2500

2000 2000 2000
Power P/kW
Power P/kW
Power P/kW

v = 12m/s v = 13m/s v = 14m/s

1500 1500 1500

1000 1000 1000

500 500 500

0 0.2 0.4 0.6 0.8 1 1.2 0 0

0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
(j) Yaw coefficient (k) Yaw coefficient (l) Yaw coefficient

Fig. 9. KDE analysis of yaw coefficient and output power.

0.50
v=4m/s
v=5m/s 0.5
Power coefficient

0.45
Power coefficient

v=6m/s 0.4
0.40 v=7m/s 0.3
0.2
0.35 0.1
0
0.30 1.2 1
0.8 6.5 7
0.25 Yaw
coeff
0.6
0.4 5.5 6 )
0.2 0 4 4.5 5 ed v m/s
/(
0 0.2 0.4 0.6 0.8 1 1.2 icien
t Wind spe
(a) Yaw coefficient (b)

Fig. 10. Power coefficient under yaw.

Table 4
Yaw effect index under different wind speed.

v 3 m/s 4 m/s 5 m/s 6 m/s 7 m/s 8 m/s 9 m/s 10 m/s 11 m/s 12 m/s 13 m/s 14 m/s

Gv¼vi 22.4% 17.3% 30.4% 32.5% 31.3% 34.2% 19.6% 14.3% 8.7% 0.06% 1.5% 0.3%
 J. Dai et al. / Energy 149 (2018) 684e696
for Fig. 8.
Based on Eq. (13) and Eq. (14), the effect of yaw under different
wind speeds is shown in Table 4. The effect of yaw on output power
below the rated wind speed is very significant. The output power
drop is 14.3% when yaw coefficient increases from 0 to 1.2 under
10 m/s wind speed, 34.2% under 8 m/s wind speed which is
maximum drop value in all wind speed conditions. In other words,
when wind speed is 8 m/s, the effect of yaw on output power is
maximum, and the yaw effect is different under different wind
speed. Generally speaking, with the increase of wind speed from
3 m/s to 8 m/s, the yaw effect index G has an upward trend. With
the increase of wind speed from 8 m/s to 12 m/s, the yaw effect
index Gshows a downward trend.
Comparing to data fitting method, the effect trend of yaw based
on KDE method is consistent with that based on LSF method in
general. But there are some differences in local relation, for
example, the trend of KDE curve in Fig. 9(g) is much different with
the data fitting curve in the corresponding Fig. 8(g). This is because
KDE gets each point using local data, but LSF gets a function
expression using global data in each Figure. In addition, it can also
be seen that data amount under higher wind speed is less than that
under lower wind speed because this mountain wind farm is
dominated by low wind speed.
6. Torque characteristic investigation under yaw
6.1. Torque characteristic investigation
Under the action of the external air flow, an axial force and a
tangential force will be generated on the blade. Bending moment
caused by the axial force causes blade and tower bending; me-
chanical torque caused by the tangential forces enables rotor
rotation. In SCADA data for the turbine used in this paper, the
parameter corresponding to the aerodynamic force is the generator
electromagnetic torque, because the mechanical torque of the wind
rotor is equal to the generator electromagnetic for direct drive wind
turbines under steady state. When investigating the effect of yaw
on rotor torque, the constraint conditions include not only the same
wind speed but also the same rotor speed. As illustrated in Fig. 11,
both rotor speed and rotor torque are intermediate (internal) var-
iables, the change of rotor speed has a direct effect on rotor torque.
Thus, rotor torque will be affected by wind speed, yaw angle and
rotor speed, that means
T ¼ f ðv; g; uÞ
Since function in Eq. (21) is rather sophisticated, currently, there
are still empirical and numerical methods to describe it. If the effect
of yaw is not considered, based on blade element momentum
(BEM) theory an approximate expression of rotor torque is [48,49]
Intermediate variables
Yaw angle Rotor speed
Output power
P T
Wind speed Rotor torque T
T f v, ,
Fig. 11. Input, output and intermediate variables.
693
Z    
1 urð1 þ a0 Þ vð1  aÞ
T¼ rc C sin arctan
2 cos2 f l
urð1 þ a0 Þ
  
vð1  aÞ
 Cd cos arctan dr dr (22)
urð1 þ a0 Þ
where, c is local chord length of blade element;r is local radius;f is
local flow angle;Cl and Cd are lift coefficient and drag coefficient,
respectively; a and a0 are axial induction factor and tangential in-
duction factor which require an iterative solution and need to be
empirically modified according to yaw angle g.
From Eq. (22), an important conclusion can be derived that is
both wind speed change and rotor speed change will lead to the
change of rotor torque. Further, if the effect of yaw is taken into
account, the effective wind speed acting on the wind turbine blade
will change. So, theoretically, the change of yaw coefficient has an
effect on rotor torque.
Using the extracted SCADA data, scattered points of rotor torque
vs rotor speed and yaw coefficient are shown in Fig. 12 under
different wind speed. In order to better observe the scattered point
distribution, the corresponding fitting surfaces are also given in the
figure.
Apparently, rotor torque increases with the increase of rotor
speed in Fig. 12(a), (b) and (c), where wind speed is below the rated
wind speed. In Fig. 12(d), rotor torque decreases with the increase
of rotor speed, where wind speed is above the rated wind speed.
The reason for this phenomenon is that MPPT control is employed
when wind speed is below its rated value and constant power
control is employed when wind speed is above its rated value.
If both wind speed and rotor speed are unchanged, the change of
yaw coefficient does not lead to large rotor torque fluctuation in
Fig. 12. To further observe this phenomenon, some 2D pattern
scattered points of yaw coefficient and rotor torque are illustrated
in Fig. 13.
6.2. Discussion of torque characteristic
Using the function T ¼ fT ðxÞjðv¼vk ;u¼ul Þ , the relationship of yaw
coefficient and rotor torque can be expressed as
fT ðxÞjðv¼vk ;u¼ul Þ ¼ cv x þ dv (23)
where, cv anddv are undetermined coefficient.
Based on the least square algorithm, the corresponding
parameter values ofcv anddv are shown in Table 5.
Obviously, under the same wind speed and rotor speed, the
change of the yaw coefficient has little effect on the rotor torque in
(21) Fig. 13, which seems to be in conflict with the theoretical analysis
by Eq. (22). To explain this phenomenon, it should be pointed out
that in the actual control strategy of the turbine, the judgment basis
of power tracking control is not based on the wind speed change,
but the “speed-power” curve. That is to say, the rotor speed and
output power are measured and used to judge whether meeting the
set “speed-power” curve, which is the basis for real-time adjust-
ment of rotor speed. In this way, corresponding to a rotor speed, the
corresponding output power is determined, then the correspond-
ing rotor torque is determined ðT ¼ P=uÞ.
7. Conclusions
Yaw will change the effective wind speed acting on the wind
rotor, which will have an effect on the wind turbine performance. In
this paper, the power coefficient has been expanded as a function of
four factors (tip speed ratio, pitch angle, yaw angle, the change rate
694 J. Dai et al. / Energy 149 (2018) 684e696

Fig. 12. Scattered points of rotor torque vs rotor speed and yaw coefficient.

200 250 250

200 200 T =1.3 +220

150 v = 4m/s v = 4.5m/s
Torque T/kN
Torque T/kN

T =0.41 +200
Torque T/kN

T =0.13 +180 v = 4m/s

150 =0.84rad/s 150 =0.84rad/s
=0.80rad/s
100
100 100
50
50 50
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) Yaw coefficient (b) Yaw coefficient (c) Yaw coefficient

350 400 400

300
250 T =0.99 +280 v = 5m/s 300 T =3.1 +340 300
Torque T/kN
Torque T/kN

v = 5.5m/s
Torque T/kN

v = 5.5m/s T =1.9 +370

200 =0.99rad/s =1.10rad/s =1.15rad/s
200 200
150
100 100 100
50
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(d) Yaw coefficient (e) Yaw coefficient (f) Yaw coefficient

Fig. 13. Relationship between yaw coefficient and rotor torque.

Table 5
Parameter values ofcv and.dv

ðv; uÞ (4 m/s, 0.8 rad/s) (4 m/s, 0.84 rad/s) (4.5 m/s, 0.84 rad/s) (5 m/s, 0.99 rad/s) (5.5 m/s, 1.1 rad/s) (5.5 m/s, 1.15 rad/s)

cv 0.13 0.41 1.3 0.99 3.1 1.9

dv 180 200 220 280 340 370

of input wind energy). The loss factor of energy capture (power coefficient.
coefficient) is subdivided into aerodynamic loss factor and inertia When investigating the effect of yaw on output power, external
loss factor, which has extended the understanding of power wind is regarded as the system input; output power of the
 J. Dai et al. / Energy 149 (2018) 684e696
generator is regarded as the system output. Below the rated wind
speed, if the wind speed keeps constant but yaw coefficient in-
creases, the corresponding output power decreases. For the
sampled data in this paper, with the increase of wind speed from
3 m/s to 8 m/s, the yaw effect index approximatively has an upward
trend. With the increase of wind speed from 8 m/s to 12 m/s, the
yaw effect index shows a downward trend. When inflow wind
speed is close to the rated wind speed or above it, the effect of yaw
on output power will become weaker because pitch control has a
stronger effect on output power than yaw in this region.
Since the “speed-power” curve is employed to realize the wind
energy tracking control in the wind turbine, the rotor speed, and
output power are measured and used to judge whether meeting the
set “speed-power” curve. So, corresponding to a rotor speed, the
corresponding output power is determined, the corresponding
torque rotor is also determined. In this way, this kind of strong
control mechanism of wind turbine weakens the effect of yaw to
rotor torque. As a result, under the same wind speed and rotor
speed, the change of the yaw coefficient has little effect on rotor
torque.
Following this research, the effect investigation of yaw on wind
turbine performance will be further improved through using more
wind farm SCADA data, and moreover, the investigation will be
further explored by combining SCADA data and CFD analysis. In
addition, the investigation of different concepts of wind turbines
has not been considered in the current research presented in this
paper. In the future, different concepts of wind turbines will be
distinguished when investigating the effect of yaw on wind turbine
performance. Also, more field influence factors will be considered
as well. All these new research achievements would be reported in
subsequent papers.
Acknowledgements
This work is supported by National Key Research and Develop-
ment Program of People's Republic of China (grant number
2016YFF0203400), National Natural Science Foundation of People's
Republic of China (grant number 51675175, 51475160), Hunan
Provincial Natural Science Foundation(grant number 2016JJ5024)
and Scientific Research Fund of Hunan Provincial Education
Department (grant number 15B084).
Nomenclature
g yaw angle
c wake skew angle
fv wind direction angle
fs rotor axis angle
f local flow angle
x yaw coefficient
gðKTÞ current yaw angle
T sample period
gb yaw angle base
Cp power coefficient
l tip speed ratio
b blade pitch angle
u rotor speed
uc cut-off frequency
uv rate of wind energy
R radius of wind rotor
r local radius
v wind speed
v1 wind speed in front of wind rotor
vd wind speed flowing through wind rotor
695
vk selected wind speed value
P power
P power set
r air density
t inertia time constant of wind rotor
Ca aerodynamic loss factor
CJ inertia loss factor
fh ðxÞ kernel density function
K($) kernel function
h bandwidth
Xi sample set
x,y,u variation
A, B, C, D matrix coefficients
a0 ; …; an undetermined coefficient
av ; bv ; cv ; dv undetermined coefficient
k0 ; k1 undetermined coefficient
m,n, l numerical variable
F symbol for the least squares method
Gv¼vk yaw effect index
S sweep area of the wind rotor
c local chord length
Cl lift coefficient
Cd drag coefficient
a axial induction factor
a0 tangential induction factor
References
[1] Baghaee HR, Mirsalim M, Gharehpetian GB, Talebi HA. Reliability/cost-based
multi-objective Pareto optimal design of stand-alone wind/PV/FC generation
microgrid system. Energy 2016;115:1022e41.
[2] Jowder FA. Wind power analysis and site matching of wind turbine generators
in Kingdom of Bahrain. Appl Energy 2009;86(4):538e45.
[3] Ucar A, Balo F. Evaluation of wind energy potential and electricity generation
at six locations in Turkey. Appl Energy 2009;86(10):1864e72.
[4] Esteban M, Leary D. Current developments and future prospects of offshore
wind and ocean energy. Appl Energy 2012;90(1):128e36.
[5] Gharali K, Johnson DA. Numerical modeling of an S809 airfoil under dynamic
stall, erosion and high reduced frequencies. Appl Energy 2012;93:45e52.
[6] Castellani F, Vignaroli A. An application of the actuator disc model for wind
turbine wakes calculations. Appl Energy 2013;101:432e40.
[7] Seim F, Gravdahl AR, Adaramola MS. Validation of kinematic wind turbine
wake models in complex terrain using actual windfarm production data.
Energy 2017;123:742e53.
[8] Abo-Khalil AG, Lee D-C. MPPT control of wind generation systems based on
estimated wind speed using SVR. IEEE Transns Ind Electron 2008;55(3):
1489e90.
[9] Zeng Q, Chang L, Shao R. Fuzzy-logic-based maximum power point tracking
strategy for Pmsg variable-speed wind turbine generation systems. Confer-
ence Fuzzy-logic-based maximum power point tracking strategy for Pmsg
variable-speed wind turbine generation systems. IEEE, p. 000405e000410.
[10] Pan T, Ji Z, Jiang Z. Maximum power point tracking of wind energy conversion
systems based on sliding mode extremum seeking control. Conference
Maximum power point tracking of wind energy conversion systems based on
sliding mode extremum seeking control. IEEE, p. 1e5.
[11] Gonza
lez L, Figueres E, Garcera
G, Carranza O. Maximum-power-point
tracking with reduced mechanical stress applied to wind-energy-conversion-
systems. Appl Energy 2010;87(7):2304e12.
[12] Ouyang T, Kusiak A, He Y. Predictive model of yaw error in a wind turbine.
Energy 2017;123:119e30.

A, Sriramula S, Krishnan N, Nichols J. Ultimate loads and response
[13] Morato
analysis of a monopile supported offshore wind turbine using fully coupled
simulation. Renew Energy 2017;101:126e43.
[14] Castellani F, Astolfi D, Sdringola P, Proietti S, Terzi L. Analyzing wind turbine
directional behavior: SCADA data mining techniques for efficiency and power
assessment. Appl Energy 2017;185:1076e86.
[15] Cortina G, Sharma V, Calaf M. Investigation of the incoming wind vector for
improved wind turbine yaw-adjustment under different atmospheric and
wind farm conditions. Renew Energy 2017;101:376e86.
[16] Micallef D, Sant T. A review of wind turbine yaw aerodynamics. In: Wind
turbines-design, control and Applications. InTech; 2016.
[17] del Campo V, Ragni D, Micallef D, Diez FJ, Ferreira CJS. Estimation of loads on a
horizontal axis wind turbine operating in yawed flow conditions. Wind En-
ergy 2015;18(11):1875e91.
[18] Medici D, Alfredsson P. Measurements on a wind turbine wake: 3D effects and
bluff body vortex shedding. Wind Energy 2006;9(3):219e36.
696 J. Dai et al. / Energy 149 (2018) 684e696
[19] Trujillo JJ, Seifert JK, Würth I, Schlipf D, Kühn M. Full field assessment of wind
turbine near wake deviation in relation to yaw misalignment. Wind Energy
Sci 2016;(1):41e53.
[20] Marathe N, Swift A, Hirth B, Walker R, Schroeder J. Characterizing power
performance and wake of a wind turbine under yaw and blade pitch. Wind
Energy 2016;19(5):963e78.
[21] Jime
Crespo A, Migoya E. Application of a LES technique to characterize

nez A,
the wake deflection of a wind turbine in yaw. Wind energy 2010;13(6):
559e72.
[22] Grant I, Parkin P. A DPIV study of the trailing vortex elements from the blades
of a horizontal axis wind turbine in yaw. Exp Fluids 2000;28(4):368e76.
[23] Pesmajoglou SD, Graham J. Prediction of aerodynamic forces on horizontal
axis wind turbines in free yaw and turbulence. J Wind Eng Ind Aerodyn
2000;86(1):1e14.
[24] Elgammi M, Sant T. A new stall delay algorithm for predicting the aero-
dynamics loads on wind turbine blades for axial and yawed conditions. Wind
Energy 2017;20:1645e63.
[25] Micallef D, Ferreira CS, Sant T, Gv Bussel. Experimental and numerical
investigation of tip vortex generation and evolution on horizontal axis wind
turbines. Wind Energy 2016;19(8):1485e501.
[26] Adaramola M, Krogstad P-Å. Experimental investigation of wake effects on
wind turbine performance. Renew Energy 2011;36(8):2078e86.
[27] Kress C, Chokani N, Abhari R. Downwind wind turbine yaw stability and
performance. Renew Energy 2015;83:1157e65.
[28] Hansen A, Butterfield C, Cui X. Yaw loads and motions of a horizontal axis
wind turbine. J Sol energy Eng 1990;112(4):310e4.
[29] Bassett K, Carriveau R, Ting DSK. Vibration response of a 2.3 MW wind turbine
to yaw motion and shut down events. Wind Energy 2011;14(8):939e52.
[30] Jeong M-S, Kim S-W, Lee I, Yoo S-J, Park K. The impact of yaw error on
aeroelastic characteristics of a horizontal axis wind turbine blade. Renew
energy 2013;60:256e68.
[31] Gebraad P, Teeuwisse F, Wingerden J, Fleming PA, Ruben S, Marden J, et al.
Wind plant power optimization through yaw control using a parametric
model for wake effectsda CFD simulation study. Wind Energy 2016;19(1):
95e114.
[32] Gebraad P, Thomas JJ, Ning A, Fleming P, Dykes K. Maximization of the annual
energy production of wind power plants by optimization of layout and yaw-
based wake control. Wind Energy 2017;20(1):97e107.
[33] Farret FA, Pfischer L, Bernardon D. Active yaw control with sensorless wind
speed and direction measurements for horizontal axis wind turbines. Con-
ference Active yaw control with sensorless wind speed and direction mea-
surements for horizontal axis wind turbines. IEEE, p. I25/1-I/6.
[34] Kragh KA, Fleming P. Rotor speed dependent yaw control of wind turbines
based on empirical data. Conference Rotor speed dependent yaw control of
wind turbines based on empirical data. p. 9e12.
[35] Mikkelsen T, Angelou N, Hansen K, Sjo € holm M, Harris M, Slinger C, et al.
A spinner-integrated wind lidar for enhanced wind turbine control. Wind
Energy 2013;16(4):625e43.
[36] Fleming P, Scholbrock A, Jehu A, Davoust S, Osler E, Wright A, et al. Field-test
results using a nacelle-mounted lidar for improving wind turbine power
capture by reducing yaw misalignment. Conference Field-test results using a
nacelle-mounted lidar for improving wind turbine power capture by reducing
yaw misalignment, vol. 524. IOP Publishing, p. 012002.
[37] Theodoropoulos S, Kandris D, Samarakou M, Koulouras G. Fuzzy regulator
design for wind turbine yaw control. Sci World J 2014;(2014). Article ID
516394.
[38] Yesilbudak M, Sagiroglu S, Colak I. A novel intelligent approach for yaw po-
sition forecasting in wind energy systems. Int J Electr Power & Energy Syst
2015;69:406e13.
[39] McKay P, Carriveau R, Ting DSK. Wake impacts on downstream wind turbine
performance and yaw alignment. Wind Energy 2013;16(2):221e34.
[40] Burton T, Sharpe D, Jenkins N, Bossanyi E. Wind energy handbook. New York:
John Wiley & Sons Ltd; 2005.
[41] Slootweg J, Polinder H, Kling W. Dynamic modelling of a wind turbine with
doubly fed induction generator. Conference dynamic modelling of a wind
turbine with doubly fed induction generator, Vancouver, BC, Canada. p.
644e649.
[42] Dai J-C, Hu Y-P, Liu D-S, Wei J. Modelling and analysis of direct-driven per-
manent magnet synchronous generator wind turbine based on wind-rotor
neural network model proceedings of the institution of mechanical engi-
neers. Part A J Power Energy 2012;226(1):62e72.
[43] Sultan A-Y, Charabi Y, Gastli A, Al-Alawi S. Assessment of wind energy po-
tential locations in Oman using data from existing weather stations. Renew
Sustain Energy Rev 2010;14(5):1428e36.
[44] Lakshmanan K, Ortner R, Ryabko D. Improved regret bounds for undiscounted
continuous reinforcement learning. Conference improved regret bounds for
undiscounted continuous reinforcement learning, Lille. p. 524e532.
[45] Liu Z, Shi R, Shen L, Xue Y, Ngan KN, Zhang Z. Unsupervised salient object
segmentation based on kernel density estimation and two-phase graph cut.
Multimedia, IEEE Trans 2012;14(4):1275e89.
[46] Bessa RJ, Miranda V, Botterud A, Wang J, Constantinescu M. Time adaptive
conditional kernel density estimation for wind power forecasting. Sustain
Energy, IEEE Trans 2012;3(4):660e9.
[47] Dai J, Liu D, Wen L, Long X. Research on power coefficient of wind turbines
based on SCADA data. Renew Energy 2016;86:206e15.
[48] Lanzafame R, Messina M. Fluid dynamics wind turbine design: critical anal-
ysis, optimization and application of BEM theory. Renew Energy 2007;32(14):
2291e305.
[49] Dai JC, Hu YP, Liu DS, Long X. Aerodynamic loads calculation and analysis for
large scale wind turbine based on combining BEM modified theory with dy-
namic stall model. Renew Energy 2011;36(3):1095e104.