Applied Energy 222 (2018) 485–496

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Equivalent modeling of wind energy conversion considering overall effect of T

pitch angle controllers in wind farm
⁎
Yuqing Jina, Ping Jua, , Christian Rehtanzb, Feng Wua, Xueping Pana
a
College of Energy and Electrical Engineering, Hohai University, Nanjing 210098, PR China
b
Institute of Energy Systems, Energy Efficiency, and Energy Economics, TU Dortmund University, Dortmund 44227, Germany

H I GH L IG H T S G R A P H I C A L A B S T R A C T

• Awindtransfer function represents the

energy conversion process with
spatial effects.
• Athecompensation wind speed represents
overall effect of all pitch angle
controllers.
• Simplest way to build equivalent
model of complete wind energy con-
version process.
• Proposed model is identified and va-
lidated by measurements from an ac-
tual wind farm.
• Example of using proposed model to
convert forecasted wind speed into
output power.

A R T I C LE I N FO A B S T R A C T

Keywords: The correct electric power fluctuation process of a wind farm is essential for the operation and studying of the
Wind farm power grid integrated with wind power. A complete wind energy conversion model, that should take both the
Wind energy conversion spatial effects and the overall effect of all the pitch angle controllers into account, is required to convert the wind
Spatial effect fluctuation into the electric power fluctuation. Although some equivalent models have been proposed in pre-
Pitch angle controller
vious studies, there is no simple solution for equivalently modeling the non-synchronous actions of the pitch
Equivalent modeling
Wind power forecast
angle controllers in individual wind turbine generators. This study found the relationship between the overall
effect of all the pitch angle controllers and the spatial effects of the wind farm and presented a theoretical
derivation of the frequency-domain equivalent modeling method. The proposed modeling method is the simplest
way to obtain the equivalent model of the complete wind energy conversion process of a wind farm with
consideration of all the spatial effects and the overall effects of all the pitch angle controllers. The only input
signal of the proposed equivalent model is the speed of the wind before entering the wind farm, which is called
the “original incoming wind speed” and the only output signal is the total power of the wind farm. In this
proposed equivalent modeling method, a discrete transfer function for representing both the wind energy
conversion process, and all the spatial effects of a wind farm, can be obtained, first through a wind process below
the rated wind speed. Second, a compensation factor for calculating a “compensation wind speed” of the original
incoming wind can be obtained through a wind process that partly exceeds the rated wind speed. Using this
compensation wind speed, a compensation power with negative values can be obtained to represent the total
reduced power caused by the pitch angle controllers in individual wind turbine generators. A frequency-domain
equivalent model has been identified and validated by field measurements of an actual wind farm. The nor-
malized root-mean-squared error of the model is less than 8% over the entire wind process, and the maximum

⁎
Corresponding author.
E-mail addresses: jyq16@hhu.edu.cn (Y. Jin), pju@hhu.edu.cn (P. Ju), christian.rehtanz@tu-dortmund.de (C. Rehtanz), wufeng@hhu.edu.cn (F. Wu), xueping_pan@163.com (X. Pan).

https://doi.org/10.1016/j.apenergy.2018.03.180
Received 6 September 2017; Received in revised form 9 March 2018; Accepted 30 March 2018
0306-2619/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).
Y. Jin et al. Applied Energy 222 (2018) 485–496

error of the power ramp rate in 10 min is less than 5 MW. Finally, an example is provided to demonstrate the
online use of the proposed model to convert the forecasted wind speed into output power of a wind farm in ultra-
short-term wind power forecast.

Nomenclature PN rated power of WTG

R radius of wind turbine
Abbreviations vN rated wind speed of WTG
ρ air density
CFEM complete frequency-domain equivalent model, includes
wind energy conversion process, all spatial effects, and Variables
overall effects of all pitch angle controllers in a wind farm
CWS compensation wind speed, represents effect of a pitch D compensation factor to calculate CWS of v0(t)
angle controller Dn wind speed reduction of vn(t) compared to v0(t)
DDPMSG direct-drive permanent magnet synchronous generator H(ω) frequency response function of wind energy conversion
DFIG doubly-fed induction generator process and spatial effects in a wind farm
EPV error of peak value of output power H(z) discrete-time transfer function of H(ω)
EPRR-1 maximum error of power ramp rate in 1 min Pe0+(t) output power of single WTG obtained from v0(t) without
EPRR-10 maximum error of power ramp rate in 10 min considering PAC
FEM frequency-domain equivalent model, includes wind en- Pen+(t) output power obtained from vn(t) without considering PAC
ergy conversion process and all spatial effects of wind Pen−(t) compensation power obtained from vn−(t), represents ef-
farm fect of PAC
MPPT maximum power point tracking PeΣ(t) total output of wind farm
NMAE normalized mean absolute error PeΣ+(t) sum of output power of all WTGs, without considering
NRMSE normalized root-mean-squared error PACs
OIWS original incoming wind speed, represents speed of wind PeΣ−(t) compensation power of entire wind farm, represents
before entering a wind farm overall effect of all PACs
PAC(s) pitch angle controller(s) PMi(t) output power of wind farm starting from Mi, produced by
WTG(s) wind turbine generator(s) wind speed from M0 to Mi
PMi+m(t) output power of wind farm from Mi to m hours in future,
Indices calculated from forecasted wind speed
td time lag
n index of WTG Tn total time lag of vn(t) relative to v0(t)
v0(t) OIWS
Parameters v′0−(t) CWS of OIWS
v1(t) wind speed of first WTG along wind direction, can be used
CPMax maximum value of power coefficient as v0(t)
m time range of wind speed forecast vn(t) wind speed of n-th WTG
M0 wind starting moment vn−(t) CWS of vn(t)
Mi wind speed forecast moments
N total number of WTGs

1. Introduction
Getting the correct fluctuation characteristic of the wind power is
essential for the dispatching, control, and electric market of the power
grids with high wind power penetration [1–4]. Power fluctuations at
wind farms are caused by the incoming wind. There are usually two
methods to get the wind speed fluctuation in different time scales. One
method is using the wind speed models, such as the four-component
composite model [5], mean value and turbulence composite model [6],
stochastic differential equations based continuous wind speed model
[7], and the Weibull distribution based model [8,9]; the other method
is through the wind speed forecast [10–14]. However, a large wind
farm may comprise hundreds of wind turbine generators (WTGs) and
cover a great area. The spatial effects of this will lead to significant
differences between the fluctuation characteristic of the incoming wind
and that of the output power, especially in a small time scale [15].
Therefore, an appropriate wind energy conversion model is important
to obtain the correct electric power fluctuation characteristics of a wind
farm.
To obtain a complete wind energy conversion model, the spatial
effects on a wind farm should be taken into account. Among the spatial
effects, the wake effect [16,17] represents the decrease in the wind
speed in the wind direction, the wind shear [18–20] represents the
wind speed change caused by undulating terrain, the tower shadow
[19,20] represents the airflow disturbance caused by the WTG tower,
and the time-lag effect [21,22] represents the time postponement of the
wind moving over a certain distance. The spatial effects could also lead
to non-synchronous actions of the pitch angle controllers (PACs) in
individual WTGs when the wind fluctuates over the rated wind speed of
the WTG. However, to construct a detailed time-domain model to re-
present both the spatial distribution of the WTGs, and the ground en-
vironment of the wind farm, is a highly complex task because of the
great number of parameters required; parameters are difficult to obtain
accurately and comprehensively [23]. Even if all the required para-
meters are obtained, the detailed model will be too complex. Therefore,
building an equivalent model of the wind energy conversion process is
more feasible and practical.
Among the existing equivalent models, the equivalent power curve
of the wind farm is a widely used time-domain equivalent model for
converting the forecasted wind speed into electric power [24–28]. The
international standard IEC 61400-12-1 gives a procedure for obtaining
the power curve from field measurements. The 10-min average wind
speed data are typically used, and a long term data accumulation is
required [28–30]. When the equivalent power curve is used for a wind

486
Y. Jin et al.
farm, it cannot be used in time intervals of seconds. In small time scales,
the equivalent modeling in the time domain will be significantly more
complex. The smaller the time scale is, the greater will be the impact of
non-synchronized variation in the output powers of individual WTGs.
Ref. [31] presented a transfer function in the frequency domain to re-
present the spatial smoothing of the output power caused by the time-
lag effect. Ref. [32] applied a Gaussian filter to the power curve to take
the smoothing effect into account. Ref. [33] provided a simulation
model for power fluctuations from large wind farms (hereinafter re-
ferred to as the detailed model). The input to this model are the power
spectral densities of the wind speeds at the hub height of each WTG,
and then a correlated power spectral densities matrix and inverse
Fourier transform are adopted to obtain the wind speeds of each WTG
in the time domain. Finally, the total output power of the wind farm is
obtained by aggregating the outputs of the individual WTGs. A fast
simulation method for this detailed model is introduced in [34]. Ad-
ditionally, based on the detailed model, [35] proposed an aggregated
model, that uses the average wind speed obtained in the frequency
domain to take the place of individual wind speeds, and uses the ag-
gregated wind turbine model to replace the individual wind turbine
models. Ref. [36] applied the aggregated model to evaluate the reduced
power under extreme wind conditions. Ref. [37] combined the
equivalent power curve and the aggregated model to assess the power
variation by considering failure outages of the WTGs. The correlated
power spectral densities matrix, whose number of elements is the
square of the number of WTGs in a wind farm, is the core of both the
detailed model and the aggregated model. In a previous study by the
author [38], a frequency-domain equivalent model (FEM) was proposed
to represent both the wind energy conversion process and all spatial
effects of a wind farm. However, [38] did not find the theoretical basis
and the solution for how to equivalently model the overall effect of all
PACs in the wind farm, and also did not validate the FEM by field
measurements.
In general, there is no simple solution for the equivalent modeling of
the non-synchronous actions of the PACs of all the WTGs, as it is an
uncertain process that varies with different wind fluctuation processes.
This issue could lead to the complication of the equivalent model, or to
an unsatisfactory simulation precision. Following on from a previous
study [38], this study found the relationship between the overall effect
of all the PACs and the spatial effects of the wind farm and gave a
theoretical derivation of the equivalent modeling method. The deriva-
tion results indicated that the overall effect of all the PACs can be re-
presented by a compensation wind speed (CWS) of the original in-
coming wind speed (OIWS). A simplified frequency-domain equivalent
model of the complete wind energy conversion process of a wind farm
was then obtained. The OIWS, that is the speed of the wind before
entering the wind farm, is the only input signal of the proposed model.
Compared to the equivalent power curve method, the proposed model
can be adopted for use in time intervals of seconds, and the long term
data accumulation is not required. The effectiveness of the proposed
frequency-domain equivalent model is validated by field measurements
of an actual wind farm. Finally, an example demonstrating the use of
the proposed model in ultra-short-term wind power forecast is
Fig. 1. Relationships between OIWS, time-lag, and speed reduction.
487
Applied Energy 222 (2018) 485–496
presented.
2. Simplified wind speed model for each WTG
When the wind moves through the wind farm, the spatial effects will
lead to the non-synchronous variation in the wind speeds in each WTG.
The spatial effects models, including the wake effect model, wind shear
model, tower shadow model, and the time-lag model, were summarized
in [38], and will, therefore, not be repeated here. It should be noted
that the Jensen wake model, whose effectiveness has been validated in
[16], is adopted to derive a linear wind speed model for individual
WTGs. The linear relationship between the wind speeds of WTGs will be
beneficial to both the derivation of the equivalent modeling method of
the overall effect of all the PACs, and the wind energy conversion
process in a wind farm.
Fig. 1 shows the relationships between the OIWS, the time-lag, and
the speed reduction caused by the spatial effects. The v0(t) obtained
from the wind measurement mast, as shown in Fig. 1, can be regarded
as the OIWS. If the v0(t) cannot be obtained, the wind speed v1(t)
measured in the first WTG along the wind direction can be used as the
OIWS, as v1(t) also indicates the speed of the wind before entering the
wind farm. When neglecting the impact of the land surface, there is no
speed reduction between v1(t) and v0(t), but a time delay.
When taking the spatial effects into account, the wind speed vn(t) of
each WTG can be approximately represented as a linear model:
⎧ vn (t ) = Dn v0 (t −Tn )
⎪ n n
, n = 1…N
⎨ Dn = ∏ di, Tn = ∑ tdi
⎪ (1)
⎩ i=1 i=1
where N is the number of WTGs, n is the index of a WTG, Dn is the wind
speed reduction compared to v0(t), and Tn is the total time lag relative
to v0(t). The time-lag td is calculated as [21,22]:
td = x / v (2)
where x is the distance the wind moves through, and v is the mean
value of wind speed.
According to the equations of the spatial effects [38] and the time
lag, the wind speed reduction Dn and the time lag Tn, are both related to
the relative position of individual WTGs, and are affected by the wind
direction. An example is shown in Fig. 2. In this wind direction, the
wind speeds v1(t), v2(t), and v3(t) are not reduced compared to v0(t), but
have different time lags. From the wake effect and tower shadow, v4(t)
has both a speed reduction and time lag compared to v0(t). When the
wind direction changes, the wind speed of each WTG can also be re-
presented by Eq. (1). In addition, according to Eq. (2), the time lag is
related to not only the relative position of the WTGs, but also the mean
value of the wind speed. Therefore, the value of Dn can be equal to, or
less than, one, and the value of Tn can be equal to, or greater than, zero.
Although the parameters Dn and Tn in Eq. (1) will vary with changes
in the wind direction or the mean value of the OIWS, their value can be
taken as constant over a certain variation range of wind directions and
mean values of the OIWS. This is a basic assumption of the following
Y. Jin et al.
Fig. 2. Impact of changes in wind direction.
derivation and has been validated by field measurements.
3. FEM of wind energy conversion
For the doubly-fed induction generator (DFIG)-based or the direct-
drive permanent magnet synchronous generator (DDPMSG)-based
WTGs, the power coefficient of the wind turbine is regarded as being
maintained at the maximum value, CPMax, because of the maximum
power point tracking (MPPT) control when the incoming wind speed is
less than the rated wind speed vN of a WTG [39,40]. In this situation,
the PAC will not act. From Eq. (1), when the OIWS v0(t) is below vN
across the entire wind process (from when the wind starts blowing,
until it stops), all the WTGs are regarded as working in MPPT mode. For
wind farms with significantly undulating terrain, if the v0(t) is measured
from a higher position than the hub height of any other WTGs, the wind
shear effect will also not affect this conclusion. The FEM of the wind
energy conversion of a wind farm is derived under this situation. The
detailed derivation process can be found in [38], and a brief in-
troduction is given below.
Ignoring the power loss in the collector network and WTGs, the sum
of the output power of all WTGs, PeΣ+(t), is:
N the energy conversion process [33,35], they will be combined au-
1 3
PeΣ + (t )= ∑ ⎡
⎣2
ρπR2vn (t ) CPMax⎤
⎦
n=1
N
1 3 3
= ∑ ⎡
⎣2
ρπR2Dn v0 (t −Tn ) CPMax⎤
⎦ speed into electric power, or the measured output power of the first
n=1 (3)
where ρ is the air density, and R is the radius of the wind turbine. We
assume that the electric power obtained from a single WTG from the
OIWS v0(t) can be represented by Pe0+(t) as follows:
1 an input, or by using electric power as an input, are not the same.
Pe0 + (t ) = ρπR2v03 (t ) CPMax
2 (4)
It should be noted that when v0(t) is less than the cut-in wind speed,
the value of v0(t) should be set to zero to obtain the correct output
power. When ignoring all the spatial effects, the ideal output of a wind
farm is N·Pe0+(t). A frequency response function H(ω) can then be ob-
tained to represent the wind energy conversion process and all the
spatial effects on a wind farm. The H(ω) can be calculated as:
488
Applied Energy 222 (2018) 485–496
N
H (ω) = F [PeΣ + (t )] F ⎡ ρπR2v03 (t ) CPMax⎤
⎣2 ⎦ (5)
where F[·] is the Fourier Transform. According to the introduction
in Section 2, an obvious variation in the wind direction or the mean
value of the OIWS will lead to significant changes in all the spatial
effects. Across a certain variation range of the wind direction, and for
those wind processes with a similar mean value of the OIWS, the same
H(ω) can be used to calculate the electric power fluctuation within an
acceptable error. The structure of the FEM based on the H(ω) is shown
in Fig. 3.
In this FEM, A·v03(t) represents converting the wind speed into the
electric power. However, because of the high inertia of the wind tur-
bine, small and high-frequency wind fluctuations will be smoothed. In
[33,35], a rotor wind model, in the frequency domain, is used to re-
present the energy conversion process and the smoothing effect of a
single WTG. The smoothing effect of the wind turbine is discussed here:
(a) If the cube of the OIWS v0(t) was taken as the input when calcu-
lating the H(ω), the smoothing effect of the wind turbine would be
included in the H(ω). Because the frequency response function of
the smoothing effect and that of the spatial effects are in series in
tomatically when calculating H(ω) according to the theory of linear
systems.
(b) If there was a suitable model of a single WTG to convert the wind
windward WTG could be obtained, N·Pe0+(t) could be used directly
as the input, and replace A·v03(t), as shown in Fig. 3. In this case, the
smoothing effect of the wind turbine would not need to be con-
sidered.
(c) According to (a) and (b), the H(ω) obtained by using wind speed as
Therefore, if the H(ω) was obtained from the wind speed signal, the
input signal should always choose wind speed when applying the
FEM. Similarly for the H(ω) obtained from the electric power signal.
It should be noted that, when calculating the H(ω), only the OIWS
v0(t) and the output power PeΣ+(t) of the wind farm need to be mea-
sured. The wind speed reduction Dn and the time lag Tn of each WTG
Fig. 3. Structure of wind farm FEM.
Y. Jin et al. Applied Energy 222 (2018) 485–496

need not be measured. By using this FEM, the troublesome detailed The PeΣ−1(t) in Eq. (10) can be further rewritten as:
modeling of the spatial effects and the wind energy conversion inside N N
the wind farm in the time domain are avoided. ⎛ 1 ⎞⎛ ⎞
PeΣ − 1 (t )= ⎜ ∑ ⎡ 2 ρπR2Dn3 vN3 CPMax⎤ ⎟·⎜N ∑ Dn3⎟
However, in the FEM, the effect of the non-synchronous PACs ac- ⎣ ⎦
⎝n=1 ⎠⎝ n=1 ⎠
tions of the WTGs have not been included. Therefore, the FEM is not N N
⎡1 2 3 3 ⎤
complete and can only be used when the OIWS is less than the vN. To = ∑ ⎢ 2 ρπR Dn (NvN ∑ Dn3) CPMax⎥
obtain a complete FEM (CFEM), the equivalent modeling method of the n=1 ⎣ n=1 ⎦ (11)
overall effect of the PACs will be derived in the next section. A compensation factor, D, is then defined as:
N
4. Equivalent model of the overall effect of the PACs
D= 3 ∑ Dn3 / N
n=1 (12)
Because of the spatial effects, the variations in the wind speeds of
each WTG are non-synchronous; that is neither the start time nor the The D is the mean cube root of Dn. Just as the values of Dn are
duration of the actions of the PACs in different WTGs are the same. The regarded as constant across a certain range of wind direction, as men-
derivation in this section is not focused on the action sequence of the tioned above, so is D. However, as the values of Dn are difficult to obtain
PACs, but the function of the PACs that reduces the surplus power from accurately by theoretical derivation, identifying the value of D from the
the incoming wind. The aim of the derivation is to obtain a proper field measurements is a viable method.
equivalent model that is compatible with the existing FEM, as shown in Substituting Eq. (12) into Eq. (11), we have:
Fig. 3, so that the FEM can be completed without requiring significant N
1
modifications. PeΣ − 1 (t ) = ∑ ⎡ ρπR2Dn3 (vN / D)3CPMax⎤
n=1 ⎣2 ⎦ (13)
4.1. Equivalent model of single PAC When substituting Eq. (13) back to Eq. (10), the total reduced power
PeΣ−(t) can be obtained as:
When the incoming wind speed is higher than vN, the WTG will exit
N
the MPPT mode, and its PAC will reduce the power coefficients by in- 1
PeΣ − (t ) = ∑ ⎡ ρπR2Dn3 [(vN / D)3−v03 (t −Tn )] CPMax⎤
creasing the pitch angle to keep the output of the WTG as its rated value n=1 ⎣2 ⎦ (14)
PN. According to the principle of the WTG, the PN can be calculated as:
According to Eq. (3) and the FEM, as shown in Fig. 3, the output of
1 the FEM is PeΣ−(t) when taking [(vN / D)3−v03 (t )] as the input. Therefore,
PN = ρπR2vN3 CPMax
2 (6) we define a new CWS v′0-(t) of the OIWS v0(t) as:
3
We can add and subtract the vn (t) in the Eq. (6) as:
⎧ 0, v0 (t ) ⩽ vN
v0′− (t ) =
1 3 3 3
⎧ PN = 2 ρπR2 [vn (t ) + vN −vn (t )] CPMax = Pen + + Pen − ⎨ 3 (vN / D)3−v03 (t ) , v0 (t ) > vN (15)
⎩
⎪ 1
Pen + = 2 ρπR2vn3 (t ) CPMax From Eqs. (14) and (15), we have:
⎨
⎪ Pen − = 1 ρπR2 [vN3 −vn3 (t )] CPMax N
⎩ 2 (7) 1
PeΣ − (t ) = ∑ ⎡ ρπR2Dn3 v0′−3 (t −Tn ) CPMax⎤
where Pen+(t) has a positive value and represents the electric power n=1 ⎣2 ⎦ (16)
absorbed from vn(t) without taking the effect of the PAC into account, We can see that Eq. (16) is similar to Eq. (3) when substituting v03 (t )
and Pen−(t) has a negative value and reduces the surplus power in with v0′−3 (t ) . As the FEM, as shown in Fig. 3, realizes the function of Eq.
Pen+(t) to represent the effect of the PAC. We can define vn−(t) as the (3), the PeΣ−(t) can be obtained directly when inputting v0′−3 (t ) to this
CWS that produces the negative power Pen−(t). The vn−(t) only appears FEM. This means that the non-synchronous actions of all the PACs can
when vn(t) is greater than vN, that corresponds exactly to the action time be equivalently modeled as the reduced power PeΣ−(t) produced by the
of the PAC, therefore vn−(t) can be calculated as: CWS v0′− (t ) calculated in Eq. (15).

⎧ 0, vn (t ) ⩽ vN
vn − (t ) = 4.3. Complete model of wind energy conversion process
⎨ 3 vN3 −vn3 (t ) , vn (t ) > vN (8)
⎩
Therefore, the effect of a single PAC in the n-th WTG is equivalently Combined with Eqs. (3) and (16), the total output of the wind farm
modeled as the negative power Pen−(t) produced by its CWS vn−(t). PeΣ(t) can be obtained as:

⎧ PeΣ (t ) = PeΣ + (t ) + PeΣ − (t )

4.2. Equivalent model of all PACs ⎪ N
⎪ PeΣ + (t ) = ∑ ⎡ 1 ρπR2Dn3 v03 (t −Tn ) CPMax ⎤
⎣2 ⎦
According to Eqs. (1) and (8), when the PAC takes effect, we have: n=1
⎨ N
⎪ 1
vn3− (t ) = vN3 −Dn3 v03 (t −Tn ) (9) ⎪ PeΣ − (t ) = ∑ ⎡ ρπR2Dn3 v0′−3 (t −Tn ) CPMax⎤
⎣2 ⎦
⎩ n = 1 (17)
Therefore, the total reduced power PeΣ−(t) from all the PACs can be
calculated in the time domain as: To complete the wind energy conversion model, the core of the FEM
need not be modified, but the input of the FEM should be expanded.
N
⎧ 1 According to Fig. 3, the structure of the CFEM is shown in Fig. 4.
⎪ PeΣ − (t ) = ∑ ⎡ ρπR2vn3− (t ) CPMax⎤ = PeΣ − 1−PeΣ − 2
⎣2 ⎦
⎪ n=1
⎪ N 4.4. Scheme for identification of CFEM
1
PeΣ − 1 (t ) = ∑ ⎡ 2 ρπR2vN3 CPMax ⎤
⎨ ⎣ ⎦
⎪ n=1 The frequency response function H(ω), and the compensation factor
N
⎪ 1 D, are the core of this CFEM, and should be obtained from the field
⎪ PeΣ − 2 (t ) = ∑ ⎡ ρπR2Dn3 v03 (t −Tn ) CPMax⎤
⎣2 ⎦ measurements. For convenience, the H(ω) can be represented by a
⎩ n = 1 (10)
transfer function. The order of the transfer function, and the

489
Y. Jin et al. Applied Energy 222 (2018) 485–496

compensation factor D, can be optimized by following the steps shown

in Fig. 5. In this scheme, the System Identification Toolbox of Matlab is
used to identify the parameters of the transfer function. As both the
input signal and the output signal are discrete-time data, the System
Identification Toolbox will output the discrete-time transfer function H
(z). To optimize the compensation factor D, any optimization method or
exhaustive attack method can be used.
To evaluate the accuracy of the output of the CFEM, the normalized
mean absolute error (NMAE) and the normalized root-mean-squared
error (NRMSE) are used. They are calculated as:

L
⎧ Pe Σ (t ) − Pcalc (t ) 2
⎪ NRMES=
⎪
1
L
∑
t=1
( Pinst ) × 100%

⎨ L
|Pe Σ (t ) − Pcalc (t )|
⎪ NMAE= 1 ∑ × 100%
⎪ L Pinst
⎩ t=1 (18)

where Pinst is the installed capacity of the wind farm, L is the number of
data points, PeΣ(t) is field measured data, and Pcalc(t) is the calculated
output power.
In the identification scheme shown in Fig. 5, two wind fluctuation
processes, with similar wind directions, will be used. In the first, the
wind fluctuation process should not exceed the rated wind speed vN.
This wind fluctuation process is used to calculate the transfer function
H(z) with the specified model order. The maximum wind speed in the
second wind fluctuation process should be greater than vN at times.
Using the second wind fluctuation process and the H(z) obtained
through the first wind fluctuation process, the compensation factor D Fig. 5. Scheme for identification of CFEM.
can be determined. With different model orders H(z) and D, differ, as
does the accuracy of the CFEM. After testing several model orders,
approximately 275°–360°, as shown in Fig. 7. The wind speed that is
choose the H(z) and D as the result of the identification scheme that has
regarded as the OIWS, v0(t), is shown in Fig. 8, and its mean value is
the best accuracy in the second wind fluctuation processes.
4.732 m/s. Fig. 9 shows the corresponding output of the wind farm
under Wind-1.
5. Validation with field measurements A further wind fluctuation process, known as Wind-2 is used to
optimize the compensation factor D defined in Eq. (12). The wind di-
5.1. Introduction to studied wind farm rection and wind speed are shown in Figs. 10 and 11, respectively. This
wind fluctuation process has a similar wind direction variation range as
The field measurements were collected from an onshore wind farm Wind-1, despite the difference in the detailed change process. However,
named Ling-Yang, located on the coast of the Yellow Sea in China with the mean value of Wind-2 reaches 7.544 m/s, and is greater than that of
48 1.5 MW DFIGs. Its structure is shown in Fig. 6. The colored lines in Wind-1. The negligible effect of the mean value difference of the OIWS
the figure indicate the feeders in the wind farm. The wind direction will be discussed in Section 5.4. Fig. 12 shows the corresponding output
arrow shown in Fig. 6 is approximately the wind direction of the field of the wind farm under Wind-2.
measurements to be introduced in this study. The detailed wind di- According to the CFEM identification scheme introduced in Section
rection changing curves will be presented below. Under this direction, 4.4, we test the model order of H(z) from 1 to 10 and set the search
the wind speed at DFIG #41 is chosen as the OIWS v0(t). The time in- range of the compensation factor D from 0.5 to 1. The result of the
terval of the measured data is 5 s. identification is:

0.05657z
5.2. Identification of CFEM ⎧ H (z ) = z − 0.9253
⎨ D = 0.9384 (19)
⎩
A wind fluctuation process, known as Wind-1, is used to calculate
the discrete transfer function H(z). Its maximum wind speed is lower These results indicate that the best order of the transfer function H
than the rated wind speed vN of 12 m/s. According to the nacelle po- (z) is one in this study. Its frequency spectrum is shown in Fig. 13.
sition data of DFIG #41, the wind direction variation range is With the identified CFEM, the NRMSE and NMAE of Wind-2 attain

Fig. 4. Complete model of wind energy conversion process.

490
Y. Jin et al. Applied Energy 222 (2018) 485–496

Fig. 6. Geographic wiring diagram of Ling-Yang wind farm.

Fig. 7. Wind direction variation range of Wind-1.

Fig. 9. Output of wind farm under Wind-1.

Fig. 8. Wind speed of Wind-1. Fig. 10. Wind direction variation range of Wind-2.

5.3. Validation of CFEM

the minimum values of 4.048% and 2.808%, respectively. Fig. 14 shows The following two wind fluctuation processes are used to validate
the calculated curves of PeΣ+(t) and PeΣ−(t), and the comparison of the the CFEM obtained above. In the first wind fluctuation process, known
actual and calculated total output power curves is shown in Fig. 15. as Wind-3, the maximum wind speed is less than vN, therefore, only the

491
Y. Jin et al.
Fig. 11. Wind speed of Wind-2.
Fig. 12. Output of wind farm under Wind-2.
Fig. 13. Frequency spectrum of H(z).
spatial effects will influence the output power of the wind farm.
However, in the second wind fluctuation process, known as Wind-4, the
maximum wind speed exceeds vN, therefore, both the spatial effects and
the PACs will play a role in the wind energy conversion process.
Except for the error indicators NRMSE and NMAE, the error of the
peak value of the output power (EPV) and the maximum error of the
power ramp rate are also compared. In the Chinese national standard
GB/T 19963-2011 on the Technical Rules for Connecting Wind Farm to
Power System, the “active power change” in 1 min and in 10 min are
considered. Therefore, the maximum errors of the power ramp rate in
1 min (EPRR-1) and in 10 min (EPRR-10) are compared. These three
492
Applied Energy 222 (2018) 485–496
Fig. 14. Calculated curves of PeΣ+(t) and PeΣ−(t).
Fig. 15. Comparison of calculated and measured output power.
error indicators are calculated as:
max(P
calc (t )) − max(P (t ))
eΣ
⎧ EPV=| | × 100%
⎪ max(Pe Σ (t ))
⎨ EPRR− 1 = |max(| P e Σ (t )−Pe Σ (t + 1min)|)−max(|Pcalc (t )−Pcalc (t + 1min)|)|
⎪ EPRR−10 = |max(|PeΣ (t )−PeΣ (t + 10min)|)−max(|Pcalc (t )−Pcalc (t + 10min)|)|
⎩
(20)
where PeΣ(t) and Pcalc (t) have the same meaning as in Eq. (18).
5.3.1. Wind speed under rated value
Figs. 16 and 17 show the features of the Wind-3 with a mean value
of only 3.660 m/s. Three output power curves are compared in Fig. 18.
The blue curve is the actual output, while the red one is calculated by
the CFEM. If the spatial effects are neglected, the output power can be
obtained as 48 (total number of the WTGs in this wind farm) times the
power of DFIG #41, shown as the dark gray curve in Fig. 18.
It can be seen that the spatial effects visibly decrease the output
power of the wind farm. The power calculated by the CFEM has similar
change tendencies as the actual power curve, that means that the CFEM
represents the spatial effects well. Table 1 presents the error indicators.
We also found that the effect of the time-lag is not significant. The
possible reason is the space scale of the wind farm is not sufficiently
great in the wind direction shown in Fig. 6.
5.3.2. Wind speed beyond rated value
Wind-4 has a mean value of 8.235 m/s, and has a similar direction
as Wind-1, as shown in Fig. 19. There is a longer period when the wind
speed is greater than vN, as shown in Fig. 20. The comparison of the
measured and calculated output power is shown in Fig. 21, and the
error indicators are presented in Table 2.
We can see that both the calculated power curve and the error in-
dicators prove the effectiveness of the proposed CFEM shown in Fig. 4.
5.4. Discussion of variation in wind direction and mean value of OIWS
The four wind fluctuation processes 1–4 discussed above have si-
milar variation ranges of the wind direction, so that the CFEM could be
identified and validated. In the above examples, the variation range of
Y. Jin et al. Applied Energy 222 (2018) 485–496

small, the effect of the variation in the mean value of the OIWS on the
application of the CFEM should be further studied, and validated, on
some larger wind farms.
In addition, attention should be paid to the accuracy of the mea-
sured wind speed, as the wind speed is the only input of the CFEM. In a
wind farm, there are typically two ways to obtain the wind speed data.
One is from a wind measurement mast, and the other is from the WTG
itself. In terms of the wind speed measuring device, the ultrasonic-based
or LiDAR-based devices have better accuracy than cup anemometer
[41,42]. Additionally, the effective wind speed estimation technology,
also known as soft sensor technology, is another way to improve the
accuracy of the wind speed measured in the WTG [43,44].
Fig. 16. Wind direction variation range of Wind-3.

5.5. Example of application of CFEM

The following is an example of online usage where the CFEM is used

for ultra-short-term wind power forecast. We assume that the wind
starts at moment M0, and the forecast takes place at moment Mi (i = 1,
2, 3, …). The forecasted wind speed data cover from Mi to m hours in
the future. The general steps to transform the forecasted wind speed to
the output of the wind farm are illustrated in the flow chart in Fig. 22.
The ultra-short-term wind speed forecast in the State Grid of China
refers to the wind forecast for 0–4 h in future, with data point intervals
less than 15 min. In this example, the Wind-4 used in the Section 5.3.2
is used here. The aim of this example is to show how to online use the
CFEM, therefore, no wind speed forecast method or its accuracy will be
Fig. 17. Wind speed of Wind-3. discussed. The forecasted wind speed will be taken directly from the
data of Wind-4, and the time interval of the data points of Wind-4 will
change to that of the 15-min average wind speed.
We assume that the forecast moment is at the end of the 16th hour,
and the duration of the measured wind speed data is 16 h, as shown by
the blue curve in Fig. 23. In this curve, the data after the 16th hour are
set to zero. The red curve in Fig. 23 is the forecasted wind speed for the
following 4 h (17th–20th hours). In this forecasted wind speed curve,
the data of the first 16 h, and the data after the 20th hour are set to zero.
We input the measured wind speed data of the 1st–16th hours to the
CFEM, and obtain the output power of the wind farm, as shown by the
blue curve in Fig. 24, including both the solid and dotted portions. The
solid blue curve is the output power in the first 16 h. The dotted blue
curve is the output power PMi(t) of the wind farm starting after the 16th
hour, that results from the wind speed in the first 16 h.
Fig. 18. Comparison of output power under Wind-3.
When the forecasted wind speed data of the 17th–20th hours to the
CFEM are input, we can get the output power of the wind farm, as
Table 1 shown by the red curve in Fig. 24, including the solid and dotted por-
Error indicators under Wind-3.
tions. The solid red curve is the forecasted output power PMi+m(t) of the
Error indicator Output power wind farm in 17th–20th hours. The dotted red curve is the output power
starting after the 20th hour which results from the wind speed in the
Using CFEM Without using CFEM
17th–20th hours.
NRMSE 0.92% 3.08% According to the flow chart shown in Fig. 22, the forecasted output
NMAE 0.67% 2.07% power of the wind farm can be obtained by adding PMi(t) and PMi+m(t)
EPV 2.51% 54.75% in the 17th–20th hours. This time range is shown in Fig. 25 as the green
EPRR-1 0.49 MW 4.00 MW
rectangle, and the sum of PMi(t) and PMi+m(t) as the green curve. The
EPRR-10 0.17 MW 3.41 MW

the wind direction is approximately 85°. This is a significant variation

range, but the determined errors, as presented in Tables 1 and 2, are
acceptable. In general, there are particular prevailing wind directions at
the wind farm, therefore, we do not need to calculate the CFEM for each
direction, that renders the application of this model convenient.
As for the obvious change in the mean value of the OIWS, the
measured field data show that it does not negatively impact the ap-
plication of the proposed CFEM. In fact, the time-lag continuously
changes in the wind fluctuation process, and the time-lag model in Eq.
(2) may be too simple to represent its exact effect on the wind energy
conversion process. As the scale of the studied wind farm is relatively
Fig. 19. Wind direction variation range of Wind-4.

493
Y. Jin et al. Applied Energy 222 (2018) 485–496

Fig. 20. Wind speed of Wind-4.

Fig. 21. Comparison of output power under Wind-4.

Table 2
Fig. 22. Flow chart of applying CFEM in wind power forecast.
Error indicators under Wind-4.
Error indicator Output Power

Using CFEM Without using CFEM

NRMSE 7.75% 18.01%

NMAE 5.30% 13.40%
EPV 5.85% 3.59%
EPRR-1 0.81 MW 25.83 MW
EPRR-10 4.60 MW 31.97 MW

green curve is the forecasted output power of the wind farm. By con-
tinuing the loop in the flow chart shown in Fig. 22, the CFEM can be
used at the next forecast moment.
The above example shows the online use of the CFEM. In addition, Fig. 23. Measured and forecasted wind speed data.
for the offline application of CFEM, the wind speed data of the entire
wind process used as OIWS should first be calculated from an appro-
priate wind speed model. The output of the wind farm can then be
obtained by inputting the wind speed data to the CFEM shown in Fig. 4.

6. Conclusion

A novel method for equivalent modeling of the wind energy con-

version process of a wind farm, taking into consideration all the spatial
effects and the overall effect of all the pitch angle controllers, was
proposed. Based on the simplified linear wind speed model for each
wind turbine generator in a wind farm, the wind energy conversion
process, and all spatial effects, can be represented by a frequency-do-
main model. The function of a single pitch angle controller can be
equivalently modeled as a compensation wind speed of the incoming Fig. 24. Output power of measured and forecasted wind speed.
wind of the wind turbine generator, that produces an electric power
with negative values to decrease the surplus power in the incoming wind speed, a simple frequency-domain equivalent model of the wind
wind. Furthermore, the overall effect of all the pitch angle controllers farm was obtained. The model included the wind energy conversion
can be equivalently modeled as a compensation wind speed of the process, all the spatial effects, and the overall effect of all the pitch
original incoming wind. With this equivalent method of compensation angle controllers in a wind farm. The only input signal of this model

494
Y. Jin et al.
Fig. 25. Forecasted output power of wind farm.
was the original incoming wind speed of the wind farm, and the only
output signal was the total power of the wind farm. The model had two
key components, one the discrete transfer function representing the
wind energy conversion process and all the spatial effects, and the other
a compensation factor to calculate the compensation wind speed of the
original incoming wind speed. Both these components could be iden-
tified from field measurements. The effectiveness of this frequency-
domain equivalent model was proved, and the given example showed
the potential of its application.
There are also certain aspects that require further study. First, how
to represent the coordinate rotation of the space caused by wind di-
rection change with analytical expressions should be studied, thus we
do not need to calculate many models for different wind directions.
Second, the effect of time-lag on the accuracy of the equivalent model
in larger wind farms should be further evaluated as the time-lag did not
appear to have a visible effect on the wind farm in this study. Third,
how to adapt to significant wind direction changes when online ap-
plying the equivalent model should be analyzed. Fourth, how to further
modify the equivalent model to adapt the outages of WTGs caused by
either extreme wind conditions or failures should be studied.
Acknowledgements
This work was supported by the National Natural Science
Foundation of China [51407059, 51422701, 51190102, 51137002],
and the “111” Project of Renewable Energy and Smart Grid [B14022].
References
[1] Ren G, Liu J, Wan J, Guo Y, Yu D. Overview of wind power intermittency: Impacts,
measurements, and mitigation solutions. Appl Energy 2017;204:47–65. http://dx.
doi.org/10.1016/J.APENERGY.2017.06.098.
[2] Liu F, Bie Z, Liu S, Ding T. Day-ahead optimal dispatch for wind integrated power
system considering zonal reserve requirements. Appl Energy 2017;188:399–408.
http://dx.doi.org/10.1016/J.APENERGY.2016.11.102.
[3] Wang Q, Wu H, Florita AR, Brancucci Martinez-Anido C, Hodge B-M. The value of
improved wind power forecasting: grid flexibility quantification, ramp capability
analysis, and impacts of electricity market operation timescales. Appl Energy
2016;184:696–713. http://dx.doi.org/10.1016/j.apenergy.2016.11.016.
[4] Exizidis L, Kazempour SJ, Pinson P, de Greve Z, Vallée F. Sharing wind power
forecasts in electricity markets: a numerical analysis. Appl Energy 2016;176:65–73.
http://dx.doi.org/10.1016/j.apenergy.2016.05.052.
[5] Anderson PM, Bose A. Stability simulation of wind turbine systems. PAS-
102:3791–5 IEEE Trans Power Appar Syst1983. http://dx.doi.org/10.1109/TPAS.
1983.317873.
[6] Nichita C, Luca D, Dakyo B, Ceanga E. Large band simulation of the wind speed for
real time wind turbine simulators. IEEE Trans Energy Convers 2002;17:523–9.
http://dx.doi.org/10.1109/TEC.2002.805216.
[7] Zárate-Miñano R, Anghel M, Milano F. Continuous wind speed models based on
stochastic differential equations. Appl Energy 2013;104:42–9. http://dx.doi.org/10.
1016/J.APENERGY.2012.10.064.
[8] Chaurasiya PK, Ahmed S, Warudkar V. Comparative analysis of Weibull parameters
for wind data measured from met-mast and remote sensing techniques. Renew
Energy 2018;115:1153–65. http://dx.doi.org/10.1016/J.RENENE.2017.08.014.
[9] Akdağ SA, Bagiorgas HS, Mihalakakou G. Use of two-component Weibull mixtures
495
Applied Energy 222 (2018) 485–496
in the analysis of wind speed in the Eastern Mediterranean. Appl Energy
2010;87:2566–73. http://dx.doi.org/10.1016/J.APENERGY.2010.02.033.
[10] Wang J, Du P, Niu T, Yang W. A novel hybrid system based on a new proposed
algorithm—multi-objective whale optimization algorithm for wind speed fore-
casting. Appl Energy 2017;208:344–60. http://dx.doi.org/10.1016/J.APENERGY.
2017.10.031.
[11] Song J, Wang J, Lu H. A novel combined model based on advanced optimization
algorithm for short-term wind speed forecasting. Appl Energy 2018;215:643–58.
http://dx.doi.org/10.1016/J.APENERGY.2018.02.070.
[12] Sun S, Qiao H, Wei Y, Wang S. A new dynamic integrated approach for wind speed
forecasting. Appl Energy 2017;197:151–62. http://dx.doi.org/10.1016/J.
APENERGY.2017.04.008.
[13] Li C, Xiao Z, Xia X, Zou W, Zhang C. A hybrid model based on synchronous opti-
misation for multi-step short-term wind speed forecasting. Appl Energy
2018;215:131–44. http://dx.doi.org/10.1016/J.APENERGY.2018.01.094.
[14] Zhao J, Guo Z-H, Su Z-Y, Zhao Z-Y, Xiao X, Liu F. An improved multi-step fore-
casting model based on WRF ensembles and creative fuzzy systems for wind speed.
Appl Energy 2016;162:808–26. http://dx.doi.org/10.1016/J.APENERGY.2015.10.
145.
[15] Yi D, Chongru L, Tian X, Yongxiu H. The power output characteristics of Jiuquan
wind power base and its reactive power compensation. In: 2013 IEEE PES Asia-
Pacific Power Energy Eng Conf; 2013. p. 1–5. 10.1109/APPEEC.2013.6837189.
[16] Göçmen T, van der Laan P, Réthoré P-E, Diaz AP, Larsen GC, Ott S. Wind turbine
wake models developed at the technical university of Denmark: a review. Renew
Sustain Energy Rev 2016;60:752–69. http://dx.doi.org/10.1016/J.RSER.2016.01.
113.
[17] Gao X, Yang H, Lu L. Optimization of wind turbine layout position in a wind farm
using a newly-developed two-dimensional wake model. Appl Energy
2016;174:192–200. http://dx.doi.org/10.1016/J.APENERGY.2016.04.098.
[18] Li J, Wang X, Yu X. (Bill). Use of spatio-temporal calibrated wind shear model to
improve accuracy of wind resource assessment. Appl Energy 2018;213:469–85.
http://dx.doi.org/10.1016/J.APENERGY.2018.01.063.
[19] Peng G. Influence analysis of wind shear and tower shadow on load and power
based on blade element theory. In: 2011 Chinese Control Decis Conf; 2011. p.
2809–12. 10.1109/CCDC.2011.5968689.
[20] Dolan DSL, Lehn PW. Simulation model of wind turbine 3p torque oscillations due
to wind shear and tower shadow. IEEE Trans Energy Convers 2006;21:717–24.
http://dx.doi.org/10.1109/TEC.2006.874211.
[21] Suvire GO, Mercado PE. Wind farm: dynamic model and impact on a weak power
system. In: 2008 IEEE/PES Transm Distrib Conf Expo Lat Am; 2008. p. 1–8. 10.
1109/TDC-LA.2008.4641719.
[22] Attya AB, Hartkopf T. Generation of high resolution wind speeds and wind speed
arrays inside a wind farm based on real site data. In: 11th Int Conf Electr Power
Qual Util; 2011. p. 1–6. 10.1109/EPQU.2011.6128927.
[23] Monforti F, Gonzalez-Aparicio I. Comparing the impact of uncertainties on tech-
nical and meteorological parameters in wind power time series modelling in the
European Union. Appl Energy 2017;206:439–50. http://dx.doi.org/10.1016/J.
APENERGY.2017.08.217.
[24] González-Aparicio I, Monforti F, Volker P, Zucker A, Careri F, Huld T, et al.
Simulating European wind power generation applying statistical downscaling to
reanalysis data. Appl Energy 2017;199:155–68. http://dx.doi.org/10.1016/J.
APENERGY.2017.04.066.
[25] Marvuglia A, Messineo A. Monitoring of wind farms’ power curves using machine
learning techniques. Appl Energy 2012;98:574–83. http://dx.doi.org/10.1016/J.
APENERGY.2012.04.037.
[26] Xue F, Song X-F, Chang K, Xu T-C, Wu F, Jin Y-Q. Equivalent modeling of DFIG
based wind farm using equivalent maximum power curve. In: IEEE Power Energy
Soc Gen Meet.; 2013 10.1109/PESMG.2013.6672848
[27] Taslimi-Renani E, Modiri-Delshad M, Elias MFM, Rahim NA. Development of an
enhanced parametric model for wind turbine power curve. Appl Energy
2016;177:544–52. http://dx.doi.org/10.1016/J.APENERGY.2016.05.124.
[28] Hayes BP, Ilie IS, Porpodas A, Djokic SZ, Chicco G. Equivalent power curve model of
a wind farm based on field measurement data. IEEE Trondheim PowerTech
2011;2011:1–7. http://dx.doi.org/10.1109/PTC.2011.6019318.
[29] Bardal LM, Sætran LR. Influence of turbulence intensity on wind turbine power
curves. Energy Procedia 2017;137:553–8. http://dx.doi.org/10.1016/J.EGYPRO.
2017.10.384.
[30] Lydia M, Kumar SS, Selvakumar AI, Prem Kumar GE. A comprehensive review on
wind turbine power curve modeling techniques. Renew Sustain Energy Rev
2014;30:452–60. http://dx.doi.org/10.1016/J.RSER.2013.10.030.
[31] Li P, Banakar H, Keung PK, Far HG, Ooi BT. Macromodel of spatial smoothing in
wind farms. IEEE Trans Energy Convers 2007;22:119–28. http://dx.doi.org/10.
1109/TEC.2006.889605.
[32] De Tommasi L, Gibescu M, Brand AJ. A dynamic wind farm aggregate model for the
simulation of power fluctuations due to wind turbulence. J Comput Sci
2010;1:75–81. http://dx.doi.org/10.1016/j.jocs.2010.03.011.
[33] Sorensen P, Cutululis NA, Vigueras-Rodriguez A, Jensen LE, Hjerrild J, Donovan
MH, et al. Power fluctuations from large wind farms. IEEE Trans Power Syst
2007;22:958–65. http://dx.doi.org/10.1109/TPWRS.2007.901615.
[34] Lin Jz, Gao Wz, Sun Y, Cheng L, Sorensen P. Fast simulation approaches for power
fluctuation model of wind farm based on frequency domain. In: 2012 IEEE Power
Electron Mach Wind Appl; 2012. p. 1–7. 10.1109/PEMWA.2012.6316365.
[35] Sorensen P, Cutululis NA, Vigueras-Rodriguez A, Madsen H, Pinson P, Jensen LE,
Y. Jin et al.
et al. Modelling of power fluctuations from large offshore wind farms. Wind Energy
2008;11:29–43. http://dx.doi.org/10.1002/we.246.
[36] Lin J, Sun Y, Cheng L, Gao W. Assessment of the power reduction of wind farms
under extreme wind condition by a high resolution simulation model. Appl Energy
2012;96:21–32. http://dx.doi.org/10.1016/J.APENERGY.2011.10.028.
[37] Cheng L, Lin J, Sun YZ, Singh C, Gao WZ, Qin XM. A model for assessing the power
variation of a wind farm considering the outages of wind turbines. IEEE Trans
Sustain Energy 2012;3:432–44. http://dx.doi.org/10.1109/TSTE.2012.2189251.
[38] Jin Y, Ju P, Wu F, Pan X, Chen Q, Sun L, et al. Equivalent modeling of wind farm in
frequency domain. In: IEEE Power Energy Soc Gen Meet, vol. 2014–Octob; 2014.
10.1109/PESGM.2014.6939241.
[39] Song D, Yang J, Cai Z, Dong M, Su M, Wang Y. Wind estimation with a non-standard
extended Kalman filter and its application on maximum power extraction for
variable speed wind turbines. Appl Energy 2017;190:670–85. http://dx.doi.org/10.
1016/J.APENERGY.2016.12.132.
496
Applied Energy 222 (2018) 485–496
[40] Kumar D, Chatterjee K. A review of conventional and advanced MPPT algorithms
for wind energy systems. Renew Sustain Energy Rev 2016;55:957–70. http://dx.
doi.org/10.1016/J.RSER.2015.11.013.
[41] Ghaemi-Nasab M, Franchini S, Davari AR, Sorribes-Palmer F. A procedure for ca-
librating the spinning ultrasonic wind sensors. Measurement 2018;114:365–71.
http://dx.doi.org/10.1016/j.measurement.2017.09.050.
[42] Shu ZR, Li QS, He YC, Chan PW. Observations of offshore wind characteristics by
Doppler-LiDAR for wind energy applications. Appl Energy 2016;169:150–63.
http://dx.doi.org/10.1016/J.APENERGY.2016.01.135.
[43] Jena D, Rajendran S. A review of estimation of effective wind speed based control of
wind turbines. Renew Sustain Energy Rev 2015;43:1046–62. http://dx.doi.org/10.
1016/J.RSER.2014.11.088.
[44] Yang X, Han X, Xu L, Liu Y. Soft sensor based on support vector machine for ef-
fective wind speed in large variable wind. In: 2006 9th Int Conf Control Autom
Robot Vis; 2006. p. 1–4. 10.1109/ICARCV.2006.345278.