Energy and Power Engineering, 2013, 5, 387-392
doi:10.4236/epe.2013.54B075 Published Online July 2013 (http://www.scirp.org/journal/epe)
A Control Strategy for Smoothing Active Power
Fluctuation of Wind Farm with Flywheel Energy Storage
System Based on Improved Wind Power Prediction
Algorithm*
J. C. Wang, X. R. Wang#
School of Electrical Engineering, Southwest Jiaotong University, Chengdu, China
Received March, 2013
ABSTRACT
The fluctuation of active power output of wind farm has many negative impacts on large-scale wind power integration
into power grid. In this paper, flywheel energy storage system (FESS) was connected to AC side of the doubly-fed in-
duction generator (DFIG) wind farm to realize smooth control of wind power output. Based on improved wind power
prediction algorithm and wind speed-power curve modeling, a new smooth control strategy with the FESS was pro-
posed. The requirement of power system dispatch for wind power prediction and flywheel rotor speed limit were taken
into consideration during the process. While smoothing the wind power fluctuation, FESS can track short-term planned
output of wind farm. It was demonstrated by quantitative analysis of simulation results that the proposed control strat-
egy can smooth the active power fluctuation of wind farm effectively and thereby improve power quality of the power
grid.
Keywords: Wind Power Generation; FESS; Wind Power Prediction; Improved Time-series Algorithm; Active Power
Smooth Control
1. Introduction
With large-scale wind power integration into grid, the
fluctuation of active power output of wind farm has
turned the operation and control of power system much
more difficult [1]. It’s shown by researches that the wind
power fluctuation of various time scales will have dif-
ferent impacts on different aspects of power system, such
as power quality, system reserve capacity, energy dis-
patching, etc. [2, 3]. Therefore, the study on smooth con-
trol of active power of wind farm contributes to improve
the power supply reliability of wind power integration
and the level of accepting large-scale wind power by
system [4].
As an effective measure to smooth the fluctuation of
wind power, energy storage technology has drawn more
and more attention [5]. Making the output active power
of wind turbine to track given smooth power curve by
controlling generator speed and pitch would reduce wind
energy utilization efficiency [6, 7]. Instead, energy stor-
age device doesn’t have such a weakness. In references
[8, 9], the Battery Energy Storage System (BESS) has
smoothed the wind power fluctuation. However, the
BESS has some shortcomings, such as high cost, heavy
environmental pollution and high period maintenance
expense etc. [3, 10]. By contrast, flywheel energy storage
system (FESS) is features with rapid charging and dis-
charging, long periodic service life and environmental
pollution free, which is applicable to smooth short-term
wind power fluctuation [11]. In reference [12, 13], FESS
is used to realize smooth control over active power out-
put of wind farm. A FESS is designed and a reference
power calculation method based on filter is studied to
determine the operation state of the FESS in reference
[13]. However, in these control methods, the role of
FESS to track planned output of wind power has not
been taken into consideration and corresponding control
strategies are seldom discussed in research.
This paper proposed a new FESS control strategy ap-
plicable to smooth short-term wind power fluctuation on
the platform of doubly-fed induction generator (DFIG)
and FESS power generation system. Reference power of
FESS is given based on improved ultrashort-term wind
power prediction algorithm and power curve modeling.
*Project support refers to acknowledgements.
#Corresponding author.

J. C. WANG, X. R. WANG
388
The energy storage system can track short-term planned
output power of wind farm and smooth the wind power
fluctuation simultaneously. A simulation system referred
to reference [13] was established on the Matlab/Simulink
for analysis, which proved the correctness and effective-
ness of the proposed smooth control strategy.
2. Power Generation System with FESS
Refer to Figure 1 for the structure of power generation
system with FESS [14]. The system is composed of DFIG
wind farm, synchronous generator (power grid) as well
as load and FESS. FESS is connected to AC side of the
wind farm converter. The FESS contains drive motor,
flywheel rotor, bearing support system, converters, etc.
High-speed permanent magnet synchronous motor
(PMSM) is adopted as drive motor for FESS to ensure
wide speed range and rapid dynamic response. Flywheel
rotor acceleration to charge and its deceleration to dis-
charge can be realized by controlling the PMSM. Thus,
mutual transfer between electric energy and mechanical
energy can be achieved. Converters of both the FESS and
the DFIG can be controlled to implement maximum
power point tracking (MPPT) of the wind turbine and
smooth the output power of wind farm to some extent.
3. Active Power Smooth Control
3.1. Wind Power Prediction Algorithm
The smooth control strategy is based on ultrashort-term
wind power prediction algorithm. FESS is used to track
the predicted wind power [15] and to smooth the high
frequent and irregular output power fluctuation. The ul-
trashort-term wind power prediction algorithm mainly
includes persistence approach [16], time-series algorithm
[17], wind speed–power curve modeling [18], etc. Time-
series algorithm was used in this paper to predict wind
speed, owning to the persistence of wind speed and other
atmospheric conditions within a short period [16] and
fewer input parameters necessary for time-series algo-
rithm. Then, in accordance with the relationship between
actual wind power and corresponding sequential wind
speed, a model was built to convert predicted wind speed
into predicted wind power.
Refer to the following for the process of wind speed
prediction algorithm.
Step1. Wind speed from supervisory control and
data acquisition (SCADA) of wind farm is used as input.
Step-length sequential wind speed before current time t is
read as input and stored in the data structure of Map.
()vt
Step2. Initialize weight parameters and learning rate
with least square method; normalize data in Map.
Step3. Predict wind speed at time t+1 with time-series
algorithm and reverse normalize results.
Step4. Update weights; t = t+1.
If multistep prediction is processed, the predicted wind
speed at time t+i shall be used as actual wind speed at
time t+i for input to obtain the predicted wind speed of
i-length after time t.
From the perspective of overall property, the power
characteristic piecewise function for wind turbine in ref-
erence [19] can be simplified as Equation (1):
(),0
() 0,
cut off
cut off
fvv v
Pv vv








(1)
Wind Power characteristic curve is when actual
wind speed is between 0 and cut-out wind speed.
can be fitted through scatter diagram of actual wind
power and corresponding sequential wind speed. Trust
region method based on nonlinear minimization is used
for curve fitting. Figure 2 ( section) is shown the
scatter points of actual wind speed and power as well as
the fitting curve of a typical DFIG (Rated capacity
1.5MW and rated wind speed 15m/s).
()fv
()fv
()fv
The fitted function is Equation (2):
32
()-0.00077230.0391- 0.42981.197fvvv v  (2)
Residual sum of squares (RSS), R-square, adjusted
R-square, and root-mean-square error (RMSE) are intro-
duced as indexes for evaluating curve fitting effect [18].
Figure 1. Power generation system with FESS.
11.5 12 12.513 13.514 14.5 15
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Win ds peed ( m / s)
Power (p u)
Scatt e r of Win d spe ed an d Power
Fitting Curve
Figure 2. Fitting curve of a typical DFIG.

J. C. WANG, X. R. WANG 389
RSS = 6.418, R-square = 0.9774, Adjusted R-square =
0.9771, RMSE = 0.2119. The closer RSS and RMSE to 0
and R-square and adjusted R-square to 1, the better the
curve fitting effect will be. Therefore, it can be seen that
the simplified model can fit wind speed–power scatter
favorably, converting predicted wind speed to predicted
power of wind turbine.
Consider the complicated real-time dispatching of
wind farm, the following improvements have been made
based on the aforementioned ultrashort-term prediction
algorithm.
1) Simulate a stop command when the wind turbine is
under overhaul condition. If this command works, no
ultrashort-term prediction will be carried out.
2) Suppose that the real-time power of wind turbine at
current time t is 0 and the planned stop time is t2 (t2>t),
if (t2-t) is smaller than the selected length of prediction
step, persistence approach shall be used to predict within
time range (t2-t), i.e. the predicted power equals to
real-time power of previous moment.
3) The phenomenon that wind speed is smaller than
cut-in speed leads to the stop of wind turbine. Probability
model shall be used for determination in this situation.
Analyze the points that wind turbine power was 0 and
corresponding wind speed was less than cut-in wind
speed (3m/s) from Jan. 2012 to Sept. 2012 of one wind
farm (hereinafter referred to as stop point). Find out 3
consecutive wind speed points (named as the point 1, the
point 2 and the point 3, the point 3 is the point closest to
the stop point on time axis) before stop points. Each 3
points makes a group. Figure 3 has demonstrated the
frequency number distribution drawing of all 3 consecu-
tive points in front of the stop points of one wind turbine
in wind farm from Jan. 2012 to Sept. 2012.
It can be seen from above figure that the wind speed of
almost all points is less than 4m/s. Calculate the descent
rate between 3 points in each group, i.e. (Point 1-Point
05 10 15
0
500
1000
1500
2000
2500
3000
3500
Frequency N um be r
Poin t 1
Winds peed ( m /s )05 10 15
0
500
1000
1500
2000
2500
3000
3500
Win ds peed ( m/ s)
Frequency N um be r
Point 2
05 10 1
5
0
500
1000
1500
2000
2500
3000
3500
Wind speed (m /s)
Frequency N um be r
Poin t 3
Figure 3. 3 consecutive points in front of the stop points
wind speed freque nc y number distribution.
2)/Δt, (Point 2-Point 3)/Δt. Δt is sample interval. It’s
shown that almost all rate of descent is larger than +1.1.
Other wind turbines of the wind farm have the similar
distribution characteristics. Therefore, suppose that the
real-time wind speed of single wind turbine at current
time t is wind speed [t], if two consecutive wind speed
sampling points before t is wind speed [t-Δt] and wind-
speed [t-2Δt] satisfy Equation (3), the predicted power at
time t+1 is 0.
[]4 /
[]4/
[-2 ]4 /
([]- [])/1.1
([2]-[])/
windspeed tms
windspeed ttms
windspeed ttms
windspeed ttwindspeed tt
windspeedttwindspeed ttt


 



 


1.1
(3)
3.2. Smooth Control Strategy
The principle for controlling the active power of wind
farm with FESS is as follows: when wind energy is af-
fluent, surplus energy will be converted into kinetic en-
ergy and be stored in the flywheel through charging
process with increasing rotate speed. When wind power
decreases or active power of the system is deficient,
lowing rotate speed to release stored energy to stabilize
the output power of wind farm around target value. Sup-
pose that actual output power of wind farm is and
the target power of wind farm with FESS is, the
reference power for FESS is Equation (4):
wf
P
exp ect
P
exprefect wf
PP P

 (4)
when , flywheel is operating as electromotor;
when
0
ref
P
0
ref
P

, flywheel is operating as generator. In
order to make the wind farm output as target power
exp ect , i.e. the key is the calculation method for exp ect
and the control of reference power of FESS. Generally,
in traditional exp ect calculation, a relatively smooth
curve can be obtained from actual wind farm output
power through low pass filter [20, 21].
P P
P
wf
A new exp ect smooth calculation method was pro-
posed in this paper based on forenamed ultrashort-term
wind power prediction algorithm. Firstly, the average
wind speed is defined as Equation (5):
PP
average
v
(())/
t
average t
vvtd

t


 (5)
wherein, is current time, t

is integration interval
and t

.
()vt is calculated by actual wind speed series through
above mentioned prediction algorithm. Input average
wind speed calculated by equation (5) into fitted power
curve to obtain exp ect , i.e. exp ,
ref of FESS shall be controlled to output compound
power as the smoothed power . Thus, the wind
()fv P()
ect average
Pfv
ect
P
exp
P

J. C. WANG, X. R. WANG
390
power fluctuation is leveled. The prediction horizon is F
and its length is not less than control horizon m. In this
paper, F=m=30s. The prediction sample interval is inte-
gral multiple of simulation step of the system. Figure 4
has displayed the actual wind speed curve of wind farm
and corresponding predicted wind speed and average
wind speed. There is an integrating process from the 0s
to around 1.2s.
The smoothness of expect can be changed through
adjusting integration interval
P

. As for wind farm with
multiple sets of wind turbines, if the distribution form of
wind turbines and wake effect are not be taken into ac-
count, approximately exp average (fv)
ect
Pk

, wherein, k
represents sets of wind turbines. The essence of given
reference power is the smoothing of predicted power.
Combined with wind power prediction technology, FESS
track the planned output power of wind farm to same
degree and smooth the irregular fluctuation of wind
power. The accuracy for exp ect to track the planned
output of wind farm is decided by the precision of wind
power prediction. The introduction of FESS reduces the
increased system reserve capacity due to the error of
wind power prediction and smoothes the wind power
output simultaneously, which contributes to the operation
management of power system dispatching department to
wind farm.
P
3.3. Control of the FESS
PMSM, used as drive motor for FESS, is adopted the
control method of maximum torque per ampere (MPTA)
[22] to make the stator current vector orthogonal with
rotor axis in reverse direction. i.e. and electro-
magnetic torque is
0
sq
i
fsq
i1.5
em
TP

. A linear relation is
formed between electromagnetic torque and stator cur-
rent on q axis component. Control
s
q to adjust the tor-
que e, thereby, the reference power ref of FESS can
be controlled. Converter at motor side is controlled by
space vector pulse width modulation (SVPWM). And
converter at power grid side is adopted the double
closed-loop control of voltage and current.
iP
T
When calculating reference power, rotor speed limit of
flywheel shall be taken into consideration. It’s set that
FESS can operate normally only when the rotor speed
is higher than minimum rotor speed and reference
power is less than or equals to 0, or the rotor speed is
lower than the maximum rotor speed and reference pow-
er is larger than 0 [23].
w
To avoid repeated accelerating of flywheel, new con-
trol rules are set as: (1) when FESS is operating normally,
reference torque of the FESS is ;
is friction coefficient and m is rated rotor speed of
the FESS; (2) When the rotor speed reaches maximum
speed and , the reference torque is
/
ref refm
TPwBw
ref m
TBw
Bw
0
ref
P

;
(3) When 0
ref
P

and rotor speed reduces to minimum
speed, the reference torque is 0 and the FESS stops gen-
erating.
4. Simulation
Established the simulation platform is shown in Figure 1
in Matlab/Simulink. Refer to simulation parameters set-
ting in reference [13]. The wind farm contains 12 DFIGs
with a rated capacity of 1.5 MW and rated wind speed 15
m/s. Power grid is composed of 1 synchronous generator
(SG). The maximum output power of FESS is 6MW and
simulation time is 30 s. Refer to Table 1 for other main
simulation parameters.
Under wind speed fluctuation shown in Figure 4, ac-
tual power output curve of wind farm without FESS and
compound power curve of wind farm with FESS using
aforementioned control strategy are shown in Figure 5.
Surge current of wind turbine will generate when starting
the simulation model. After 1.7 s, the system is stable.
19
05 10 15 20 25 3
0
11
12
13
14
15
16
17
18
Time (s)
Windspeed (m/s)
Pred ict ed W inds pee d
Actu al Win dspeed
Average Win d speed
Inte g rat ing Process
Figure 4. Curves of wind speed and average wind speed.
Table 1. Main parameters of the simulation system.
Simulation Parameters
Devices
Parameter Value
Rotational inertia 600 2 kg m

Initial speed 4000rpm
FESS
Friction coefficient 0.0011
Nmsrad


Rated voltage 575V
DFIG
Maximum output power 1.75MW
Rated voltage 13.8kV
Rated power 200MVA SG
Number of pole-pairs 32

J. C. WANG, X. R. WANG 391
And then, two indicators are introduced to determine
the smoothness of active power output.
1) Maximum fluctuation of compound output power of
wind power and energy storage shall not exceed 10% of
rated capacity of wind farm rated within any 20 min-
utes. During the simulation time after stable operation of
the system, the difference between maximum and mini-
mum compound output power is 9.3MW, far less than
rated capacity of wind farm 18MW. It can be concluded
from Figure 5 that smooth control strategy with FESS
has reduced the duration when the output power of wind
farm is higher than the rated power.
P
2) Define smooth coefficient [12] as Equation (6):
0
(|/| )/
t
wf rated
SmoothdPdtdtP (6)
The smaller the gradient of smooth coefficient curve,
the smoother active power output to power grid. Refer to
Figure 6 for smooth coefficient curve with/without
FESS.
51015 20 25 3
0
8
10
12
14
16
18
20
22
Time (s)
Power (MW)
without FESS
with FESS
Figure 5. Curve of active power output of wind farm with/
without FESS (From 1.7s).
05 10 15 20 25 3
0
0
0.5
1
1.5
2
2.5
3
Time (s)
S m ooth Coefficient
without FESS
w ith F ESS
Figure 6. Curves of smooth coefficient with/without FES
flu
ooth control strategy for active power
National Natural Science
[1] Y. Z. Sun, J. uence Research of
Power
, “Power Smoothing Control Strat-
al., “Smoothing Control
G. Infield, “Energy Storage and its
S.
It can be seen from above figure that there is larger
ctuation in active power output of wind farm without
FESS even though DFIG is operating under the control
strategy of MPPT. After using FESS, the proposed con-
trol strategy reduces wind power fluctuation significantly
without affecting wind energy utilization efficiency. The
smooth coefficient is only 25% of that without FESS.
5. Conclusions
In this paper, a sm
of wind farm with FESS was realized. Meanwhile, the
demand of power grid dispatch for wind power predic-
tion was also taken into consideration. A new calculation
method for FESS reference power was proposed based
on improved wind power prediction algorithm and power
curve modeling. From the quantitative analysis of simu-
lation results, it could be concluded that the proposed
control strategy would smooth the fluctuation of wind
power effectively and track short-term planned output
power of wind farm favorable. Therefore, the smooth
control strategy has improved schedulability of wind
farm, which contributes to the stable operation of power
grid. However, the effects that the precision of wind
power prediction and energy storage capacity act on the
control of active wind power need a further study.
6. Acknowledgements
This work is supported by
Foundation (NNSF) of China under Grant 50937002 and
by the Fundamental Research Funds for the Central Uni-
versities under Grant SWJTU09ZT10.
REFERENCES
Wu and G. J. Li, “Infl
Wind Power Generation on Power Systems,” Power Sys-
tem Technology, Vol. 31, No. 20, 2007, pp. 55-62.
[2] H. Holttinen and R. Hirvonen, “Wind Power in
Systems,” 1st Edition, John Wiley & Sons Ltd., Chiches-
ter, 2005.
[3] Q. Xiong and L. Liao
egy for Direct-driven Wind Power System with Flywheel
Energy Storage System,” East China Electric Power, Vol.
40, No. 9, 2012, pp.1533-1539.
[4] Y. C. Xue, N. L. Tai, K. Song, et
Strategy on Power Fluctuation of Power Filtering Based
Wind Farm,” East China Electric Power, Vol. 39, No. 3,
2011, pp. 454-458.
[5] J. P. Barton and D.
Use with Intermittent Renewable Energy,” IEEE Trans-
actions on Energy Conversion, Vol. 19, No. 2, 2004, pp.
441-448. doi:10.1109/TEC.2003.822305
[6] L. Changling, H. Banakar, S. Baike, et al., “Strategies to
Smooth Wind Power Fluctuations of Wind Turbine Gen-

J. C. WANG, X. R. WANG
392
erator,” IEEE Transactions on Energy Conversion, Vol.
22, No. 2, 2007, pp. 341-349.
doi:10.1109/TEC.2007.895401
“Power Smoothing Con-
Energy Stor-
T. Yan, “An Optimization
. Guo, “A Control Strategy for Battery En-
an, J. C. Zhang and J. H. Wang, “Improvement of
wer
ion Control of
ntrol of
eme Design for
[7] Y. Liao, J. B. He, J. Yao, et al.,
trol Strategy of Direct-driven Permanent Magnet Syn-
chronous Generator for Wind Turbine With Pitch Angle
Control and Torque Dynamic Control,” Proceedings of
the CSEE, Vol. 29, No. 18, 2009, pp. 71-77.
[8] D. Q. Bi, B. M. Ge and W. L. Wang, “VRB
age System Based Power Control of Grid-connected
Wind Farm,” Automation of Electric Power Systems, Vol.
34, No. 13, 2010, pp. 72-78.
[9] H. S. Hong, Q. Y. Jiang and Y.
Control Method of Battery Energy Storage System with
Wind Power Fluctuations Smoothed in Real Time,” Au-
tomation of Electric Power Systems, Vol. 37, No. 1, 2013,
pp. 103-109.
[10] B. Li and J. B
ergy Storage System to Level Wind Power Output,”
Power System Technology, Vol. 36, No. 8, 2012, pp.
38-43.
[11] J. P. Ru
Stability of Wind Farms Connected to Power Grid Using
Flywheel Energy Storage System,” Electric Power Sci-
ence and Engineering, Vol. 24, No. 3, 2008, pp. 5-8.
[12] X. S. Hu, C. X. Sun, R. Liu, et al., “An Active Po
Smoothing Strategy for Direct-driven Permanent Magnet
Synchronous Generator Based Wind Turbine Using Fly-
wheel Energy Storage,” Automation of Electric Power
Systems, Vol. 34, No. 13, 2010, pp. 79-83.
[13] R. Xiang, X. R. Wang and J. Tan, “Operat
Flywheel Energy Storage System Applying to Wind
Farm,” Journal of Systems Science and Mathematical
Sciences, Vol. 32, No. 4, 2012, pp. 438-449.
[14] R. Takahashi and J. Tamura, "Frequency Co Iso-
lated Power System with Wind Farm by Using Flywheel
Energy Storage System," Proceedings of the 18th Inter-
national Conference on Electrical Machines of the ICEM,
Vilamoura, 6-9 September 2008, pp. 1-6.
[15] P. Jiang and H. C. Xiong, “A Control Sch
Smoothing Wind Power Fluctuation with Hybrid Energy
Storage System,” Automation of Electric Power Systems,
Vol. 37, No. 1, 2013, pp. 122-127.
[16] M. Lei, L. Y. Shi, W. J. Chuan, et al., “A Review on the
Forecasting of Wind Speed and Generated Power,” Re-
newable and Sustainable Energy Reviews, Vol. 13, No. 4,
2009, pp. 915-920. doi:10.1016/j.rser.2008.02.002
[17] K. Chang, M. S. Ding, F. Xue, et al., “Ultra-short-term
Wind Power Prediction and its Application in Ear-
ly-warning System of Power Systems Security and Sta-
bility,” Power System Protection and Control, Vol. 40,
No. 12, 2012, pp. 19-24.
[18] J. F. Li, B. H. Zhang, G. L. Xie, et al., “Grey Predictor
Models for Wind Speed-Wind Power Prediction,” Power
System Protection and Control, Vol. 38, No. 19, 2010, pp.
151-159.
[19] W. Wangdee and R. Billinton, "Considering
Load-carrying Capability and Wind Speed Correlation of
WECS in Generation Adequacy Assessment,” IEEE
Transactions on Energy Conversion, vol. 21, No. 3, 2006,
pp. 734-741. doi:10.1109/TEC.2006.875475
[20] T. Tanabe, T. Sato, R. Tanikawa, I. Aoki, T. Funabashi
and R. Yokoyama, "Generation Scheduling for Wind
Power Generation by Storage Battery System and Mete-
orological Forecast,” Proceedings of the Power and En-
ergy Society General Meeting - Conversion and Delivery
of Electrical Energy in the 21st Century of the IEEE,
Pittsburgh, PA, 20-24 July 2008, pp. 1-7.
[21] J. V. Paatero and P. D. Lund, “Effect of Energy Storage
on Variations in Wind Power,” Wind Energy, Vol. 8, No.
4, 2005, pp. 421-441. doi:10.1002/we.151
[22] X. Z. Tang, “Vector Control System of Permanent
-Magnet Synchronous Motor,” Master Thesis, Zhejiang
University, Zhejiang, 2005.
[23] C. Chapelsky, J. Salmon and A. Knight, "Control of a
High-Inertia Flywheel as Part of a High Capacity Energy
Storage System," Proceedings of the Canadian Confer-
ence on Electrical and Computer Engineering of the
CCECE, Vancouver, BC, 22-26 April 2007, pp.
1437-1440.