Académique Documents
Professionnel Documents
Culture Documents
Energy
journal homepage: www.elsevier.com/locate/energy
a r t i c l e i n f o a b s t r a c t
Article history: Most wind turbine generators installed in large wind farms are variable speed types which operate at the
Received 12 December 2014 maximum power point tracking mode in order to increase the power generation. Due to this fact and
Received in revised form regarding the random nature of the wind speed, the output power of the wind farm fluctuates randomly.
27 January 2016
Fluctuating power affects network operation and needs to be smoothed. In order to mitigate the output
Accepted 28 January 2016
Available online 24 February 2016
power fluctuations of a wind farm, a 4-step coordinated control technique based on ADALINE (ADAptive
Linear NEuron) is proposed in this paper which uses a small BESS (Battery Energy Storage System) ca-
pacity. At first the on-line tracking of the WFOP (Wind Farm Output Power) is carried out by ADALINE.
Keywords:
Wind farm power fluctuation
Afterwards, two constraints for maximum permissible fluctuations are imposed on the ADALINE output.
ADALINE (ADAptive Linear NEuron) Two states of charging feedback control strategies are implemented in the third and fourth steps.
Coordinated control Reducing the battery capacity in proposed coordinated control technique is fulfilled through the accurate
BESS (battery energy storage system) tracking performed by ADALINE and also by maintaining the level of BESS saved energy within the
batteries safe performance region performed by state of charging feedback control strategies. Simulation
results run by real data verify that the performance of the proposed approach is considerably better than
the basic approach.
© 2016 Elsevier Ltd. All rights reserved.
1. Introduction tracking mode [2]. Due to this fact and regarding the random nature
of wind speed, the output power of the wind farm fluctuates. The
Global warming along with persuasion of environmental orga- effect of fluctuations on the system stability is more significant
nizations to decrease greenhouse gases on the one hand, and when the wind farm capacity or the wind power production part in
increasing demand for electrical energy and its growing final price the whole power system increases. This is a challenge which may
on the other hand have attracted the attention of governments and lead to the power system collapse in unsuitable situations [3].
different societies to renewable energy resources [1]. Wind is one of So far, various researches have been conducted in order to mitigate
the renewable energy resources possessing characteristics dis- fluctuations of WFOP (Wind Farm Output Power).
tinguishing it from others. However, emergence of the wind energy A large majority of studies demonstrated that wind power fluc-
in electrical networks and increasing the power production in wind tuations can be mitigated by blade pitch control [4]. However, me-
farms resulted in some new challenges too. Because of their high chanical stress may damage blades if the pitch angel control operates
final cost compared with other parts of the power generation fast enough to mitigate short-term fluctuations. Utilizing ESSs (En-
network, wind turbines need to work at the maximum power point ergy Storage Systems) by different technologies has been recognized
as another acceptable solution to smooth wind power fluctuations
[5e7]. In Refs. [8] and [9] some approaches based on pumped storage
and compressed air energy saving have been studied which have a
* Corresponding author.
E-mail addresses: mohsen.jannati@shahreza.ac.ir (M. Jannati), hosseinian@aut. low dynamic adjustment speed and it is not suitable for large wind
ac.ir (S.H. Hosseinian), vahidi@aut.ac.ir (B. Vahidi), liguojie@sjtu.edu.cn (G.-j. Li). farms. Implementing SMES (Superconducting Magnetic Energy
http://dx.doi.org/10.1016/j.energy.2016.01.100
0360-5442/© 2016 Elsevier Ltd. All rights reserved.
2 M. Jannati et al. / Energy 101 (2016) 1e8
two constraints define maximum variation values as 2 percent and 4.1. ADALINE architecture
7 percent of the wind farm rated power per 1-min window and per
30-min window respectively. ADALINE is a two-layer neural network with n inputs and one
The most stringent constraints belong to the dual power ramp output which the output is a linear combination of the inputs. It
constraints. In this study two power ramp constraints are taken was first proposed by Widrow and Hoff, Stanford University [16,27].
into account; as the test network is a wind farm located in China, Main features of the ADALINE are as follows:
dual constraints used in china are chosen to be used in this study:
1. It can be trained on-line based on changing inputs and the target
1) The maximum variation in the COP is set to 2 percent of the response.
wind farm rated power (Prated) per 1-min window (PFMC-1min) 2. It is a self adaptive algorithm which can be applied to the
i.e. maxfDPCo j1min g Prated x1 % where x1 ¼ 2. weights training.
2) The maximum variation in the COP is set to 7 percent of Prated per 3. It has a simple structure making it easily implemented on
30-min window (PFMC-30 min) i.e. maxfDPCo j30min g Prated hardware.
x30 % where x30 ¼ 7.
In order to track a given signal (or its harmonic components for
It should be remarked that the maximum variation in instan- various applications) on-line, an adaptive linear neuron can be
taneous COP per each 30-min window may increase up to 10 implemented. To ponder the subject, consider an arbitrary signal
percent temporarily. containing a harmonic component as follows:
In this step, the optimum SOC performance region of batteries is During the training phase, if the output ~
zðkÞ equals to the target
defined and then a COD method is determined which contributes in signal, discrete Fourier transform coefficients of z(k) refer to the
fixing batteries in their defined optimum performance region while weight vector q(k). In the training phase, the weights should be
reducing the system necessity for battery capacity. Therefore, when chosen so that the error between ADALINE output and target signal
the battery SOC approaches to the optimum performance region, is minimized. Widrow-Hoff training rule based on the Least Square
the feedback control strategy prolongs its entrance to this region Error approach is used to train ADALINE [28]. The weights are
(more details have been addressed in Section 4.3). corrected as follows:
zðkÞ ¼ ~
zðkÞ ¼ qTo XðkÞ (6)
After the error converges to zero, the weight vector qo in (6)
equals to:
recognized as an on-line approach [29]. Since in this study Dt ¼ 5 sec, m and n values are assigned 12 and
The considered signal in this research is the waveform of WFOP. 360 respectively.
Ideally it is expected to be similar to a DC waveform; however it has Considering PFMC-1min and PFMC-30min, the COP value of the
unpredictable fluctuations practically. Assuming that u ¼ 0, next sample should be within a specific range; otherwise the
z(t) ¼ PADA(t), ~zðtÞ ¼ PWF ðtÞ and e(t) ¼ DP(t) ¼ PADA(t)PWF(t), the mentioned constraints are not satisfied. To clarify the issue,
following equation is achieved: waveforms of PCo has been shown in Fig. 4 for 30 min hypotheti-
cally. Specifically for (k1)th sample while having windows WP1Co
DPðkÞ ¼ eðkÞ ¼ PADA ðkÞ PWF ðkÞ (8)
and WP30Co
, the maximum power fluctuations during one and 30-min
time periods (ε1 and ε30) are obtained as follows:
where PADA(k), PWF(k) and DP(k) indicate the output power of
ADALINE, the WFOP and their difference in kth sample respectively. n o n o
The kth sample is the sample at time t, i.e. t ¼ k Dt where Dt max WP1Co min WP1co
represents the sampling period. ε1 ¼ 100 (12)
Prated
Moreover, the needed battery capacity at time t or in the kth
sample is calculated by (9): n o n o
max WP30
co
min WP30
co
X
k X
k ε30 ¼ 100 (13)
CðkÞ ¼ DPðiÞ Dt ¼ eðiÞ Dt (9) Prated
i¼1 i¼1 Regarding the description in Section 3, it's obvious that ε1 x1
Equation (9) indicates that the needed battery capacity until the and ε30 x30.
kth sample is achieved via multiplying the sampling period by the It should be noted that although the kth sample have not taken
sum of the network learning error function. k range can be assigned by both the PFMC-1min and PFMC-
yet, PCo
The waveform of the wind farm output alluded in Section 5 k is outside a specified range, conditions
30min. It means that if PCo
along with the waveforms of ADALINE output have been shown
in Fig. 3 that are related to a 16 h time period.
As it can be inferred from Fig. 3, the FLF method has not been
very successful in the tracking of the wind farm output power.
Although the FLF graph is smoother, it needs more BESS capacity for
the smoothing process. However, the ADALINE method has a better
tracking result. It has accomplished the smoothing purpose with
less BESS capacity. Therefore, the ADALINE method is more cost-
effective than the FLF method. Moreover, it is observed that delay
of the ADALINE output is less than that of the FLF output.
The PCo data is saved into one and 30-min windows. Upon
receiving new data, the oldest data existing in the window is
removed and the new data is added. Assume that COP data win-
dow (up to kth sample) for one and 30 min is formulated as in (10)
and (11): Fig. 4. Waveforms of the COP for 30 min hypothetically.
M. Jannati et al. / Energy 101 (2016) 1e8 5
)
ε1 x1 and ε30 x30 are not satisfied for the kth sample. Therefore, m
bk 1 X b kþi P
b kþi
k 2½P k k k P ¼ P (21)
Co1 ; P Co1
considering PFMC-1min, we have: PCo where P Co1 E;REF ADA WF
m i¼1
and P k Co1 are the upper and lower bounds respectively. Likewise,
k k
k where P b b
k 2½P k WF and P ADA are the predicted WFOP and ADALINE
considering PFMC-30min, we have: PCo Co30 ; P Co30 where bk
k respectively. P E;REF represents the state of COD reference for the kth
P Co30 and P k Co30 are the upper and lower bounds respectively. bk bk
sample. P E;REF < 0 and P E;REF > 0 are indicative of COD for the kth
Upper and lower bounds shown in Fig. 4 are calculated by following sample respectively.
equations: b k (the predicted SOC at kth sample) is calculated:
Now S OC
n o
k b k < 30 and P
1. If SOCmin < S OC bk
P Co1 ¼ min WP1Co þ Prated x1 % (14) E;REF > 0, discharging is allowed but
k
PCo declines.
n o b k < SOCmax , no new constraint is determined for
2. If SOCmin < S OC
P kCo1 ¼ max WP1Co Prated x1 % (15) COD at kth sample.
b k < SOCmax and P
3. If 70 < S OC bk < 0, charging is allowed but P k
E;REF Co
n o increases.
k
P Co30 ¼ min WP30
Co
þ Prated x30 % (16)
These constraints prolong the entrance time of batteries to their
n o optimum performance region. In other words, the total needed
P kCo30 ¼ max WP30
Co
Prated x30 % (17) capacity for smoothing purposes is reduced. Fig. 5 shows the
strategy of this step schematically.
Now considering (18), (19) is achieved: k in this strategy is under the following constraints:
Concisely, PCo
8
> k k
> P Co PCo S OC
> k
b k þ Pk b k 2ðSOC ; 30Þ& P
b
>
> if S OC min E;REF > 0
>
> 30 Co
>
>
>
< Pk b k 2ðSOC ; SOCmax Þ
k Co if S OC min
PCo ¼ (22)
> k
> k
>
> P PCo
>
> Co b k 70 þ P k
S OC if S OC bk
b k 2ð70; SOCmax Þ& P <0
>
> 30 Co E;REF
>
>
:
Go SCFC 2 O:W
h k
i h k
i h k
i 4.4. SCFC-2 (state of charging feedback control-2)
P kCo ; P Co ¼ P kCo1 ; P Co1 ∩ P kCo30 ; P Co30 (18)
By implementing SCFC-1, the battery SOC is set to be within the
h k
i optimum region. As explained previously, adjusting the SOC in the
k
PCo 2 P kCo ; P Co (19) ranges of [10,20]% or [80, 90]% doesn't hurt the battery specifically
for short time periods. In this step, COD is allowed but PFMC-30min
where PCok represents the COP in kth sample. Concisely, considering becomes milder. As shown in Fig. 6, moderating this constraint is
the data up to (k1)th sample and PFMC-1min and PFMC-30min, achieved by selecting x30. Indeed, by appointing a variable value for
PCo in kth sample is under the following constraint: x30 in a given short time, the needed battery capacity is reduced
8 again. In this strategy: xnew
> Pk if k
PCo P k Co 30 2½x30 ; x30 where x30 ¼ 7 and
>
< Co x30 ¼ 10 . Also, the algorithm of this strategy is as follows:
k
k
PCo ¼ PCok
if PCo 2 P k Co ; P Co
k
(20)
>
>
: k k k
P Co if PCo P Co
5) Now (16) and (17) are rewritten as follows: Being located in Shanxi Province, China, the under-study 99 MW
wind farm contains 66 DFIGs (Doubly-Fed Induction Generators)
n o
k 30 each of which has 1.5 MW capacity. Each DFIG unit utilizes both
P Coxnew ¼ min W P þ Prated xnew
30 % (25)
30 Co
boost programs and has 0.69 kV export voltage which then boosts
to 35 kV. All the DFIG units are connected together and to the power
n o
grid via a collection line. Then the voltage boosts to 110 kV. Fig. 7
P kCoxnew ¼ max WP30 Prated xnew
30 % (26)
30 Co shows the under-study wind farm. The analysis of the proposed
k technique is performed based on the real data of the given WFOP
where P Coxnew
30
and P k Coxnew are new upper and lower bounds with the sampling period of 5 s.
30
respectively. It is worth mentioning that 1-min constraint is usually
satisfied by a proper tracking.
Thus, the following term applies to the COP at kth sample: 5.2. Wind farm PFS through the proposed technique
h k
i In order to analyze the performance of the proposed coordi-
k
PCo 2 P kCoxnew ; P Coxnew
30
(27) nated control method, the real data of a whole day related to WFOP
30
Table 1
The required BESS capacity allocation through the proposed approach and FLF
approach.
ADALINE FLF
Moreover, ADALINE is used for tracking in the proposed method [6] Uehara A, Pratap A, Goya T, Senjyu T, Yona A, Urasaki N, et al. A coordinated
control method to smooth wind power fluctuations of a PMSG-based WECS.
which is trained in less than 20 msec. on the other hand, the
IEEE Trans Energy Convers 2011;26(2):550e8.
required time for running both methods is very small so that their [7] Abbey C, Joos G. Supercapacitor energy storage for wind energy applications.
operation can be realized online. Eventually, the total time is much IEEE Trans Ind Appl 2007;43(3):769e76.
smaller than the sampling period of the wind power (the sampling [8] Barton JP, Infield DG. Energy storage and its use with intermittent renewable
energy. IEEE Trans Energy Convers 2004;19(2):441e8.
period is 5 s according to the Section 5.1). Therefore, the calculation [9] Liyan Q, Wei Q. Constant power control of DFIG wind turbines with super-
speed is sufficiently fast to satisfy online operation requirements in capacitor energy storage. IEEE Trans Ind Appl 2011;47(1):359e67.
both methods. Since the FLF-based method has a simple structure [10] Pan WX, Fan YW, Ju L, Gao AL. Optimal selection of pumped storage system
capacity in wind farms. Electr Technol 2008;3:120e4.
as well, the proposed ADALINE-based method has no advantage [11] Tan J, Li GJ, Tang ZW. Power control and benefit analysis based on compressed
over it in computational cost point of view. air energy storage in wind farms. Autom Electr Power Syst 2011;8:33e7.
Another point is that constraints imposed by the utilities for [12] Tomkoi A, Takahashi R, Murata T. Smoothing control of wind power generator
output by superconducting magnetic energy storage system. In: Proceedings
renewable PFS have direct effect on the required BESS capacity. of International Conference on Electrical Machines and Systems, Seoul, Korea;
Dual constraints which impose the most stringent conditions in- 2007. p. 302e7.
crease the system need for BESS capacity, but provide smoother [13] Jia HX, Zhang Y, Wang YF. Application of energy storage technology in wind
power systems. Renew Energy 2009;27:10e5.
active power for injecting to the grid. Thus, selecting different [14] Ribeiro PF, Johnson BK, Crow ML, Arsoy A, Liu YL. Energy storage systems for
constraints leads to different results in various manuscripts. advanced power application. IEEE Trans Power Syst 2001;89(12):1744e56.
In the other words, in a specific country or utility if the PFMC- [15] Boutoubat M, Mokrani L, Machmoum M. Control of a wind energy conversion
system equipped by a DFIG for active power generation and power quality
1min constraint is utilized rather than the mentioned dual con-
improvement. Renew Energy 2013;50:378e86. Elsevier.
straints (PFMC-1min and PFMC-30min), the required BESS capacity [16] Tanabe T, Sato T, Tanikawa R, Aoki I, Funabashi T, Yokoyama R. Generation
decreases considerably. The constraints are defined by the utilities scheduling for wind power generation by storage battery system and mete-
based on the power grid structure and the capacity of available orological forecast. In: Proc. 21st Century IEEE Power Energy Soc. Gen.
Meeting e Convers. Del. Electr. Energy, Jul.; 2008. p. 1e7.
batteries. [17] Abbey C, Strunz K, Joos G. A knowledge-based approach for control of two-
level energy storage for wind energy systems. IEEE Trans Energy Convers
6. Conclusion 2009;24(2):539e47.
[18] Jiang Q, Gong Y, Wang H. Battery energy storage system dual-layer control
strategy for mitigating wind farm fluctuations. IEEE Trans Power Syst
In order to mitigate the output power fluctuations of a wind 2013;99:1e11.
farm, a coordinated control based on ADALINE was proposed in this [19] Member Q, Wang H. Two-time-scale coordination control for a battery energy
storage system to mitigate wind power fluctuations. IEEE Trans Energy
paper by use of a low BESS capacity. The proposed coordinated Convers 2013;28(1):52e61.
technique contains 4-step including: tracking by ADALINE, [20] Jiang Q, Hong H. Wavelet-based capacity configuration and coordinated
applying PFMC-1min and PFMC-30min, and finally performing the control of hybrid energy storage system for smoothing out wind power
fluctuations. IEEE Trans Power Syst 2013;28(2):1363e72.
strategies SCFC-1 and SCFC-2. Compared with the basic FLF based [21] Xiaojuan H, Fang Ch, Xiwang C, Yong L, Xiangjun L. A power smoothing
approach, the accurate and fast tracking of the WFOP performed by control strategy and optimized allocation of battery capacity based on hybrid
ADALINE contributes to decrease the required capacity for energy storage energy technology. Energies 2012;5:1593e612.
[22] Xiangjun L, Dong H, Li W, Xiaokang L. Control strategy of battery state of
saving in the system. In the next step two constraints defined by
charge for wind/battery hybrid power system. In: Proc. 2010 IEEE Int. Symp.
utilities are applied which are very significant factors in allocating Ind. Electron., Bari, Italy; 2010. p. 2723e6.
the required BESS for the wind farm. Two SCFC strategies make the [23] Yoshimoto K, Nanahara T, Koshimizu G. New control method for regulating
best possible use of the battery safe performance region by con- state-of-charge of a battery in hybrid wind power/battery energy storage
system. In: Proceedings of the 2006 IEEE PES Power Systems Conference and
trolling the level of remained energy in the BESS. For performance Exposition (PSCE '06), Atlanta, GA, USA; 2006. p. 1244e51.
analysis, real output power data of a typical wind farm was tested [24] Sercan T, Mesut EB, Subhashish B, Alex H. Validation of battery energy storage
by the proposed technique and the basic FLF based approach. It is control for wind farm dispatching. In: Proceedings of the IEEE Power and
Energy Society General Meeting, Minneapolis, MN, USA; 2010. p. 1e7.
concluded through the simulation results that the required BESS [25] Esmaili A, Nasiri A. Power smoothing and power ramp control for wind en-
capacity in the ADALINE based approach (with or without use of ergy using energy storage. In: Energy Conversion Congress and Exposition
SCFC strategies) is less than the FLF based approach. (ECCE); 2011. p. 922e7.
[26] Fox B, Flynn D, Bryans L, Jenkins N, Milborrow D, Malley MO, et al. Wind
power integration: connection and system operation aspects. IET Power En-
References ergy Ser 2007;50.
[27] Widrow B, Hoff ME. Adaptive switching circuits. In: WESTCON Convention,
[1] Mohibullah Imdadullah, Ashraf I. Estimation of CO2 mitigation potential Record Part IV; 1960. p. 96e104.
through renewable energy generation. In: Power and Energy Conference, [28] Widrow B, Lehr MA. 30 years of adaptive neural networks: perceptrons,
PECon '06. IEEE; 2006. p. 24e9. adaline and backpropagation. IEEE Proc 1990;78(9):1415e42.
[2] Abdeddaim S, Betka A. Optimal tracking and robust power control of the DFIG [29] Alcantara FJ, Salameron P. A new technique for unbalance current and voltage
wind turbine. Int J Electr Power Energy Syst 2013;49:234e42. Elseveir. estimation with neural networks. IEEE Trans Power Del 2005;20(2):852e8.
[3] Jannati M, Hosseinian SH, Vahidi B, Li Guo-Jie. A survey on energy storage [30] Vahidi B, Jannati M, Hosseinian SH. A novel approach to adaptive single phase
resources configurations in order to propose an optimum configuration for autoreclosure scheme for EHV power transmission lines based on learning
smoothing fluctuations of future large wind power plants. Renew Sustain error function of ADALINE. Soc Model Simul Int 2008;84(12):601e10.
Energy Rev 2014;29:158e72. Elsevier. SIMULATION.
[4] Devaraj D, Jeevajyothi R. Impact of fixed and variable speed wind turbine [31] Wang ZQ, Manry MT, Schiano JL. LMS learning algorithms: misconceptions
systems on power system voltage stability enhancement. In: IET Conference and new results on convergence. In: IEEE Trans. Neural Networks; 2000.
on Renewable Power Generation; 2011. p. 1e9. p. 47e56.
[5] Senjyu T, Sakamoto R, Urasaki N, Funabashi T, Fujita H, Sekine H. Output [32] Babazadeh H, Wenzhong G, Xiaoyu W. Controller design for a hybrid energy
power leveling of wind turbine generator for all operating regions by pitch storage system enabling longer battery life in wind turbine generators. In:
angle control. IEEE Trans Energy Convers 2006;21(2):467e75. North American Power Symposium (NAPS). IEEE; 2011. p. 1e7.