Anda di halaman 1dari 34

Accepted Manuscript

Development of a wake model for a Darrieus-type straight-bladed vertical axis


wind turbine and its application to micro-siting problems

H.F. Lam, H.Y. Peng

PII: S0960-1481(17)30721-8

DOI: 10.1016/j.renene.2017.07.094

Reference: RENE 9069

To appear in: Renewable Energy

Received Date: 26 June 2016

Revised Date: 22 July 2017

Accepted Date: 23 July 2017

Please cite this article as: H.F. Lam, H.Y. Peng, Development of a wake model for a Darrieus-type
straight-bladed vertical axis wind turbine and its application to micro-siting problems, Renewable
Energy (2017), doi: 10.1016/j.renene.2017.07.094

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT

1 Development of a wake model for a Darrieus-type straight-bladed vertical axis wind

2 turbine and its application to micro-siting problems

3 Lam H.F.1 and Peng H.Y.*1,2


4 1 Department of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong SAR

5 2 School of Civil and Environmental Engineering, Harbin Institute of Technology, Shenzhen, China

6 Abstract

7 Recent wake studies suggest that vertical axis wind turbines (VAWTs) exhibit a potential

8 increase in packing density in wind farm situations. In this study, an analytical wake model fit

9 for straight-bladed VAWTs was developed and applied to the micro-siting problem of VAWTs.

10 The wake asymmetry of VAWTs in the horizontal direction distinguished itself from the

11 axisymmetric wake of horizontal axis wind turbines (HAWTs). In the proposed wake model, it

12 was assumed that the wake region at a given distance was encompassed by two semi-ellipses.

13 The velocity in the wake was derived based on continuity conservation. Wake measurements

14 were conducted in a wind tunnel, and the model parameters were derived by fitting predicted

15 results to parameters measured in experiments. The wake model was next used to evaluate the

16 park effect. Due to the asymmetry, a method to determine the position of a turbine relative to

17 its upstream counterparts was developed. To faithfully reflect the wind conditions on sites,

18 distributions of both wind direction and speed were considered. The layout design of VAWTs

19 was first performed on regular wind farm sites. Moreover, the cluster configuration was

20 generalized to irregular wind farms. The winding number method was adapted specifically for

21 the irregular constraints.

22 Keywords: wake model, wake asymmetry, VAWT, micro-siting, wind tunnel test.

* Corresponding author.
E-mail address: pollenhavy@hotmail.com (Peng H.Y.)

1
ACCEPTED MANUSCRIPT

23 1. Introduction

24 Wind energy is clean, and wind is one of the most promising renewable energy sources (Tai et

25 al., 2013; Lee and Lim, 2015; Lam and Peng, 2017). To produce wind-generated electrical

26 power on a commercial scale, a wind farm consisting of dozens of wind turbines must be built.

27 Due to the power extraction and blockage effect, the wake of a wind turbine suffers a velocity

28 deficit and turbulence hike. These harsh wind conditions certainly affect the power production

29 performance of downstream wind turbines. To minimize such interference and thereby

30 maximize total power production, the placement of wind turbines must be optimized (Chen and

31 MacDonald, 2014).

32 Many efforts have been made over the past several decades to address the wake modeling and

33 micro-siting problem of horizontal axis wind turbines (HAWTs) (Jensen, 1983; Katic et al.,

34 1986; Mosetti et al., 1994; Grady et al., 2005; Saavedra-Moreno et al., 2011; Wan et al., 2012;

35 Turner et al., 2014). Jensen (1983) proposed an analytical model to estimate the average

36 velocity and radius of the HAWT wake that is now widely used in micro-siting problems. In

37 that study, a circular cone spreading out linearly was used to incorporate the wake region, in

38 which constant velocity at a given downstream distance was assumed. Marmidis et al. (2008)

39 conducted the optimal placement of wind turbines using Jensens wake model and Monte Carlo

40 simulation. In that study, a wind farm site was split into a great number of cells, which were the

41 possible turbine positions. The cluster configurations can also be solved without dividing a

42 wind farm site into grids. Kusiak and Song (2010) formulated the expected power production

43 under statistical distributions of both wind direction and speed. In that study, Jensens wake

44 model was applied to estimate wake losses, and the evolution strategy (ES) was used for

45 optimization. Turner et al. (2014) developed mixed integer linear and quadratic optimization

46 formulations by using Jensens wake model. The new mathematical formulations produced

47 more symmetrical layouts of turbines with slightly more power compared to previous methods.

48 However, research efforts focused on the wake characteristics of vertical axis wind turbines

49 (VAWTs) have been rather limited. To date, no wake model has been available to estimate the

2
ACCEPTED MANUSCRIPT

50 wake spreading and velocity of VAWTs. Academic communities have focused more on the

51 aerodynamic loading and power production of stand-alone VAWTs (Islam et al., 2008; Howell

52 et al., 2010; McLaren 2011; Bianchini et al., 2015). In fact, recent studies have suggested that

53 VAWTs are more suitable for applications in built environments, e.g., rooftops of buildings and

54 offshore areas (Balduzzi et al., 2012; Saeidi et al., 2013; McNaughton et al., 2014; Elkhoury et

55 al., 2015). Tjiu et al. (2015) reported that several multi-megawatt offshore Darrieus-type

56 VAWTs were currently in progress and that HAWTs were facing major challenges in the multi-

57 megawatt range, especially offshore. The authors concluded that Darrieus-type VAWTs had a

58 lower cost of energy (COE), better operational performance, and higher scalability than

59 HAWTs in offshore applications. Furthermore, some studies have claimed that VAWTs have

60 fast wake recovery, resulting in a high packing density in wind farm scenarios (Simo Ferreira,

61 2009; Kinzel et al., 2012; Hezaveh and Bou-zeid, 2013; Tescione et al., 2014). Different from

62 the axisymmetric wake of HAWTs, the wake of straight-bladed VAWTs exhibits substantial

63 asymmetry in the horizontal direction. Therefore, a circular cone linearly spreading

64 downstream, which is adopted in an HAWT, cannot represent the wake of straight-bladed

65 VAWTs. A new wake model especially suitable for straight-bladed VAWTs must be

66 developed. Doing so would properly resolve the micro-siting problem of VAWTs.

67 In this study, a wake model of straight-bladed VAWTs was developed and applied to optimize

68 the cluster configuration of VAWTs. The wake was assumed to spread out linearly, and the

69 wake asymmetry in the horizontal direction was taken into account. The velocity in the wake

70 was derived based on continuity conservation. Wind tunnel tests were conducted to update the

71 wake model. The model-predicted results agreed well with the experimental data. Next, the

72 wake model was applied to the layout design of VAWTs. A method was proposed to resolve

73 the relative position of a turbine to its upstream counterpart. Statistical distributions of wind

74 direction and speed were considered in the micro-siting problem. The covariance matrix

75 adaption based evolutionary strategy (CMA-ES) was implemented to maximize the power

76 production. Finally, the layout design was further extended to wind farms with irregular

77 boundaries.

3
ACCEPTED MANUSCRIPT

78 2. Modeling Methodology

79 2.1 Problem formulation

80 Figure 1 is a schematic plot of the proposed wake model. A Cartesian coordinate system (right-

81 hand rule) was introduced with the x axis pointing downstream, the y axis pointing windward,

82 and the z axis aligned with the tower. The origin of the coordinate system was set at the turbine

83 center at the blade mid-span level. In Figure 1, Part I is a view cut horizontally at the level of

84 the blade mid-span plane, and Part II shows the wake edges cut vertically at a downstream

85 distance of x. rW and rL are the spreading distances at the windward and leeward, respectively.

86 The straight-bladed VAWT rotated counterclockwise, as viewed from the top. The wake

87 modeling of the VAWT involved a number of important assumptions based on wind tunnel

88 tests of VAWTs, large eddy simulations of VAWTs, and what is currently known about the

89 wake of HAWTs. The assumptions are listed as follows.

90 Assumption 1: The wake region of the VAWT at a given downstream distance is represented

91 by two semi-ellipses sharing the same minor axes (see Part II in Figure 1). This assumption was

92 based on the measured velocity contour on a vertical plane in the wake of an H-rotor VAWT

93 (Peng et al., 2016).

94 Assumption 2: The two semi-cones spread out linearly downstream. Four spreading rates were

95 adopted to reflect the wake asymmetry, i.e., kW, kL, kU, and kD at the windward, leeward, upward,

96 and downward, respectively. This assumption was based on the engineering wake model (Peng

97 et al., 2016). It differed from the wake modeling of HAWTs, in which only one circular cone

98 was used to represent the spreading of the wake edge (Jensen, 1983; Katic et al., 1986).

99 Assumption 3: The wake was asymmetrical in the horizontal direction and symmetrical against

100 the blade mid-span plane (Tescione et al., 2014, Peng et al., 2016; Lam and Peng, 2016). This

101 indicated that kL did not equal kW, but kD equaled kU.

102 Assumption 4: At a given distance downstream, the velocity in the wake region was constant.

103 This assumption made use of the simplification applied in the wake modeling of HAWTs

104 (Jensen, 1983; Katic et al., 1986). Though this assumption does not reproduce the exact velocity
4
ACCEPTED MANUSCRIPT

105 profiles downstream, it is of practical interest in engineering applications in the micro-siting

106 problems to be addressed in this study.

107 Assumption 5: For the upward and downward spreading, the wake edges started from z = H/2

108 at x = 0. In contrast, due to stronger expansion in the horizontal direction, the wake edges started

109 from bD at the windward and leeward. Moreover, the spreading constant, b, was generally

110 larger than 1/2 due to the stronger influence of the blade rotation at the blade mid-span plane,

111 as reflected in the engineering wake model (Peng et al., 2016).

112

113 Figure 1. Schematic of the wake model of the VAWT

114 According to the continuity conservation at turbine position (x = 0) and at a downstream

115 distance x, the following equation was constructed:

H 1 H 1
116 bv0 rW rU rL rD b U rW rU rL rD v (1)
2D 2 D 2

117 where is the air density; v0 denotes the wake velocity at x = 0; U is the free-stream velocity; v

118 is the wake velocity; rW, rL, rU, and rD represent the semi-axes of ellipses as functions of x at

119 the windward, leeward, upward, and downward, respectively; and || || is the operator of a norm

120 on the one-dimensional vector space.

121 The wake edges spreading in the horizontal direction at the symmetry plane (z = 0) were

5
ACCEPTED MANUSCRIPT

122 expressed as follows:

x
rW kW b
D
123 (2)
x
rL k L b
D

124 The wake edges spreading in the vertical direction at y = 0 were calculated as follows:

x 1
rU kU
D 2
125 (3)
x 1
rD k D
D 2

126 where = H/D is the aspect ratio of the turbine, and kU and kD are equal in magnitude. The

127 wake velocity at a particular downstream distance was derived as follows:

H
r r rL rD U U v0
W U
D
b
128 v (4)
rW rU rL rD

129 Substituting Eqs. (2) and (3) into Eq. (4) and normalizing the velocity by the free-stream

130 velocity, U, the normalized wake velocity, v, was attained as follows:

2b vV
131 v' 1 (5)
x x
kW k L 2b 2kU
D D

132 where v equals (Uv0)/U, i.e., the velocity deficit at the turbine center (x = 0). v was used to

133 predict the wake losses of individual turbines in wind parks and was essential to the layout

134 design.

135 2.2 Wind tunnel tests

136 To identify the values of the model parameters, i.e., kW, kL, b, v, kU, in Eqs. (2), (3), and (5),

137 wake measurements were performed in a wind tunnel. The wind tunnel at City University of

138 Hong Kong is 2.0 m high and 2.5 m wide. The VAWT with five straight blades was placed in

139 the wind tunnel, as shown in Figure 2. The diameter, D, of the turbine was 0.3 m, and the blade

140 depth, H, was 0.3 m. As a result, the blockage ratio of the turbine was 1.8% < 10%, and hence

6
ACCEPTED MANUSCRIPT

141 no correction for the blockage effect was required (Ross and Altman, 2011; Chen and Liou,

142 2011). The height of the turbine from the ground to the level of the blade mid-span, h, was 1.0

143 m. As such, the blade mid-span was placed at the mid height of the wind tunnel. The turbine

144 was placed 6 m downstream of the wind entrance. A 4-m distance, within which the wind tunnel

145 had good flow uniformity, was reserved for the wake measurements.

146

147 Figure 2. Straight-bladed VAWT in the wind tunnel

148 The VAWT with cambered blades performed outstandingly well at self-starting. No motor was

149 required to drive the turbine to rotate, and the turbine operated naturally in the wind. This made

150 the wake measurements much closer to the actual behavior in the field. A four-hole cobra probe

151 was used to measure the wake velocity. It was mounted onto a three-dimensional traversing

152 system (TDTS) to automatically cover the target-measured points. The sampling rate was 3,000

153 Hz, and the measurement duration was 30 s. The measurement duration is approximately 100

154 times the rotational speed of the turbine, and hence is enough to capture the stationary wake

155 characteristics. The wake was measured along horizontal lines at the level of the blade mid-

156 span plane, against which the wake flow pattern was approximately symmetrical. Moreover,

157 the wake pattern at the blade mid-span plane, i.e., the symmetry plane, was crucial to the

158 placement of multiple VAWTs.

7
ACCEPTED MANUSCRIPT

159 The measurement campaign was implemented at two different blade speed ratios (BSRs): 1 =

160 0.24 and 2 = 0.42. Four downstream distances of the turbine, i.e., x = 2D, x = 4D, x = 6D, and

161 x = 8D (D for the turbine diameter), were measured. For further details of the instruments and

162 experimental setup, please refer to the study by Peng et al. (2016).

163 Figure 3 presents the normalized stream-wise velocities, Ux/U (Prfl.), and their associated

164 mean values (Const.) along the y direction at z = 0 at two different BSRs at four downstream

165 distances. In this study, the wake edge is defined in a way that Ux/U equals unity on the edge,

166 suggesting a full wake recovery. There are two edges separating the wake from the free-stream

167 flow, with one at the windward and the other at the leeward. Figure 3a clearly reflects that the

168 near wake suffers severe velocity deficits, and the velocity profile deviates greatly from the

169 average value. As it approaches the far wake, e.g., x = 4D, the velocity gradually recovers due

170 to the flow mixing, and the average value becomes more and more representative of the wake

171 velocity. This kind of wake development further supports the assumption of the constant

172 velocity in addressing micro-siting problem. Moreover, great wake asymmetry is observed from

173 the Ux/U profiles, differing significantly from the axisymmetric wake of HAWTs (Vermeer et

174 al., 2003; Gonzlez et al., 2010). At x = 2D downstream, the maximum velocity deficit (MVD)

175 position shifts 0.5D from y = 0 toward windward. The wake flow motion toward windward

176 continues as the wake spreads out. At x = 8D downstream, the position of the MVD migrates

177 approximately 0.75D from x = 0 toward windward.

8
ACCEPTED MANUSCRIPT

1.2 1.2

1 1
Velocity (Ux/U)

Velocity (Ux/U)
0.8 0.8

0.6 0.6

(a) (b)
0.4 0.4
-2 -1 0 1 2 -2 -1 0 1 2
Lateral distance (y/D) Lateral distance (y/D)
=0.24, Prfl. =0.24, Const. =0.42, Prfl. =0.42, Const.
1.2 1.2

1 1
Velocity (Ux/U)

Velocity (Ux/U)

0.8 0.8

0.6 0.6

(c) (d)
0.4 0.4
-2 -1 0 1 2 -2 -1 0 1 2
178 Lateral distance (y/D) Lateral distance (y/D)

179 Figure 3. Stream-wise velocities along the y direction at z = 0 (velocities from 2 = 0.42 were

180 extracted from (Peng and Lam, 2016)): (a) x = 2D, (b) x = 4D, (c) x = 6D, and (d) x = 8D

181 Based on the results from the wind tunnel tests, the model parameters of the proposed wake

182 model of the straight-bladed VAWT were determined. First, the least squares method was used

183 to fit Eq. (2) with the measured edges at the windward and leeward, and kW, kL, and b were then

184 obtained. kW, kL, and b were then substituted into Eq. (4). Regression analysis of Eq. (4) with

185 the measured velocities in the wake gave v and kU. Finally, Table 1 shows the parameters of

186 the wake model at two BSRs. The asymmetrical wake spreading is quantitatively reflected in

187 the spreading rates: kW and kL. At the windward, the spreading rate, kW, was about 0.10

188 compared with the spreading rate, kL, of around 0.065 at the leeward. In addition, the spreading

189 rate, kU, in the vertical direction was around 0.05, smaller than that in the horizontal direction.

190 The spreading distance at the symmetry plane was approximately 0.8D, larger than the turbine

191 radius as expected. Moreover, the average velocity deficit, v, at the turbine center (x = 0) was

9
ACCEPTED MANUSCRIPT

192 approximately 0.40.

193 Table 1. Model parameters of the wake model at different BSRs

kW kL b v kU

1 = 0.24 0.101 0.072 0.79 0.36 0.047

2 = 0.42 0.104 0.063 0.83 0.39 0.044

194 Figure 4 presents the matching between the predictions from the proposed wake model and the

195 wind tunnel test results at 1 = 0.24. In Figure 4(a), good agreement between the spreading

196 edges of the wake model and that of the experiments is achieved. In Figure 4(b), the predicted

197 average velocities agree well with the measured data. The wake of the VAWT recovered

198 incredibly fast, and v reached more than 85% at x = 6D and approximately 90% at x = 8D

199 downstream. As a rule of thumb, VAWTs in wind parks can be spaced somewhere between 6D

200 apart in the prevailing direction. The fast wake recovery indicated a high packing density of

201 VAWTs in a given area, giving a high output of electricity power per unit area. The good

202 agreement between the measured and model-predicted results suggests that the proposed wake

203 model was capable of well predicting the wake characteristics of the straight-bladed VAWT.

2
(a)
Lateral distance (y/D)

1
Measurement @Windward
Wake model
0
Measurement @Leeward
Wake model
-1

-2
2 3 4 5 6 7 8
Wake distance (x/D)

0.9
(b)
Average velocity (v')

0.85

0.8
Measurement
Wake model
0.75
2 3 4 5 6 7 8
204 Wake distance (x/D)

10
ACCEPTED MANUSCRIPT

205 Figure 4. Comparisons between the predictions of the wake model and the measurement

206 results: (a) wake edges and (b) wake velocity

207 2.3 Wake loss of multiple wakes

208 The position of a given VAWT can be represented by (xi, yi), i = 1, 2, , N, where N is the

209 number of turbines. In a wind farm scenario, there are dozens to even hundreds of wind turbines.

210 At a given wind direction and wind speed, the mutual interference of turbines can be assessed

211 based on the proposed wake model. Without loss of generality, Figure 5 shows the schematic

212 for the evaluation of the effect of turbine j (at B) on turbine i (at A). The two turbines rotated

213 counterclockwise as indicated by the arrows on the circles. The vector BC represents the wind

214 direction, i.e., (cos (), sin ()) in relation to the global coordinate system. Note that equals

215 0 when the wind blows in the x direction and 90 when the wind direction is along the y axis.

216 The vector BA is (xi xj, yi yj). Therefore, the angle, , between BC and BA is:

xi x j cos yi y j sin
217 cos 1 , ji (6)
l

218 where l = ((xj xi)2 + (yj yi)2)1/2 is the distance between the two turbines. As a result, the

219 projected distance of BA in the wind direction is la = ||(xj xi) cos () + (yj yi) sin ()||. BC is

220 represented as (la cos (), la sin ()). The magnitude of the vector CA is lb = l sin ().

11
ACCEPTED MANUSCRIPT

221

222 Figure 5. Schematic for the calculation of the wake interference

223 Unlike the situation for HAWTs, one additional step was needed to check whether turbine i was

224 affected by turbine j for VAWTs. It was necessary to check whether turbine i stood at the

225 windward or leeward of turbine j before evaluating the wake effect. The reason is that the wake

226 edges, which are asymmetrical in the horizontal direction, expand differently toward windward

227 and leeward. In this study, the checking process was converted into a mathematical problem by

228 evaluating the cross product of BC and CA. With xc = xj + la cos () and yc = yj + la sin (), CA

229 was expressed as (xi xj la cos (), yi yj la sin ()). Consequently, the cross product, , of

230 BC and CA was written as the following formal determinant:

i j k
231 la cos la sin 0 (7)
xi x j la cos yi y j la sin 0

232 Through mathematical manipulation, Eq. (7) was further simplified as follows:

233 la y y cos x x sin k


i j i j (8)

234 As a result, there were three possible outcomes (I-III) for the relationship between turbines i

235 and j.

12
ACCEPTED MANUSCRIPT

236 I: If = la ((yi yj) cos () (xi xj) sin ()) > 0 and cos () > 0, turbine i located at the windward

237 of turbine j. Whether turbine i was located in the wake of turbine j could be determined by

238 comparing lb with rW.

239 II: If < 0 and cos () > 0, turbine i located at the leeward of turbine j. Whether turbine i

240 suffered the wake loss from turbine j could be determined by comparing lb with ||rL||.

241 III: If = 0 and cos () > 0, turbine i was in the wake of turbine j.

242 If turbine i was in the wake of turbine j, the velocity loss, v, could be calculated according to

243 Eq. (5):

2b vV
244 vi' 1 vij' (9)
x x
kW k L 2b 2kU
D D

245 where vij' is the normalized velocity at the position of turbine i, which experienced the wake of

246 turbine j. Under wakes of multiple turbines, the total velocity deficit at the position of turbine i
247 was calculated based on energy conservation (Manwell et al., 2009; Kusiak and Song, 2010):
1
N
2 2
248 vi' 1 vij' (10)
j 1, j i

249 3. Wind Farm Layout Design

250 3.1 Problem formulation

251 3.1.1 Power curve of the VAWT

252 The power curve, P = f (v), describes the power production of a VAWT at different wind speeds,

253 v. The power curve is, in fact, a piecewise function. When the wind speed is smaller than the

254 cut-in speed, vcut-in, no electricity power is produced due to insufficient torque. If the wind speed

255 is larger than vcut-in and smaller than the rated output speed, vrated, the power curve increases

256 rapidly. Note that the turbine power output reaches Prated when the wind speed is at vrated.

257 Between vrated and the cut-out speed, vcut-out, the braking system starts to work and keeps the

258 power output fixed at Prated. Beyond vcut-out, the braking system stops the turbine from working

13
ACCEPTED MANUSCRIPT

259 to prevent disastrous damage, and hence no power is generated. In this study, instead of using

260 a linear function, a modified sigmoid function was used to describe the growing power curve,

261 as suggested by Kusiak and Song (2010). As a result, the power curve was proposed to have

262 the following form:

0, v vcut in
P0
, vcut in v vrated
263 f v 1 ec0 v v1/2 (11)
P , v
rated v vcut out
rated
0, v vcut out

264 where P0 is a power constant, c0 is a constant, and v1/2 is the wind speed, at which half of P0 is

265 obtained. These model parameters could be updated with the power curve provided by the

266 manufacturer. The modified sigmoid function was capable of capturing the essential feature of

267 the power curve in the growing phase.

268 3.1.2 Wind direction and wind speed

269 On a wind farm site, wind direction and speed continually change both temporally and spatially.

270 A common way to describe this information is to use annual statistical distributions of wind

271 direction, p(), and wind speed, pv(v,). In this study, it was assumed that wind speeds shared

272 the same distribution at a given wind direction. Furthermore, the wind distribution was assumed

273 to be uniform in space so that they could be applied to the whole wind farm. As a result, the

274 expected power output of an individual turbine could be calculated as follows (Kusiak and

275 Song, 2010):

360
276 E Pi p f v p v, dvd
v (12)
0 0

277 where E(Pi) is the expectation of the power production of turbine i. Integrating Eq. (12)

278 analytically was an impractical task. Through discretization of the wind direction and speed,

279 the expected power production could be approximated by numerical integration.

280 In reality, wind originates from various directions ranging from 0 to 360. In this study, the

14
ACCEPTED MANUSCRIPT

281 wind directions of 0 and 90 were aligned with the x and y axes, respectively (as defined in

282 Figure 5). To implement numerical integration, the wind direction was discretized into N bins

283 with equal intervals. The center points of the bins were 1, 2, , N, and the associated

284 probability of each bin was 1, 2, ..., N.

285 According to the literature, wind speed is commonly assumed to follow a Weibull distribution

286 (Kusiak and Song, 2010; Eroglu and Sekiner, 2012; Wagner et al., 2013). In fact, this

287 assumption holds for many wind farm sites in a long-term situation (Manwell et al., 2009). In

288 this study, the Weibull distribution of the wind speed was expressed as follows:

b'
b' 1 v

b'
v a'
e , v0
289 pv v, a' , b' a' a' (13)


0, v 0

290 where a is the scale parameter, and b is the shape parameter, both of which are functions of

291 the wind direction and could be updated according to the measured wind data. By analogy, the

292 wind speed was discretized into Nv bins with equal widths. The center points of the bins were

293 v1, v2, , vNv, and the corresponding cumulative probability of each bin was 1, 2, ..., Nv. They

294 were evaluated by the cumulative distribution function as follows:

1 e v a' b' , v 0
295 F v, a' , b' (14)
0, v 0

296 3.1.3 Constraints and objective function

297 Considering a rectangular wind farm site, the coordinates of the turbine positions must satisfy

298 the following conditions:

L 2 xi L 2
299 , i 1, 2 L , N (15)
W 2 yi W 2

300 where L is the length of the area, and W is the width of the area. To ensure sufficient turbine

301 spacing for maintenance and the prevention of disastrous interactions, the distance, l, between

302 any two turbines must be larger than or equal to the minimum distance, lmin.
15
ACCEPTED MANUSCRIPT

x x y y
1
2 2 2
303 l i j i j lmin , ji (16)

304 In this study, with the discretization of wind distribution, Eq. (12) was transformed from an

305 integration formula into a summation,

N Nv
306 E Pi st f vis ,t (17)
s 1 t 1

307 where s represents the bin number of wind direction, t denotes the bin number of wind speed,

308 and vis ,t is the wake velocity at turbine i with a wind direction of s and a wind speed of vt.
309 Substituting Eq. (10) into Eq. (17), the expected power production of an individual turbine was

1

N Nv
s ,t N 2

E Pi st f U 1 1 vij
' s ,t 2
310 (18)

s 1 t 1 j 1, j i

311 where Us,t is the free-stream velocity with the direction of s and magnitude of vt, and v 'ijs ,t is the

312 velocity at turbine i in the wake of turbine j with a wind direction of s and a wind speed of vt.

313 Finally, the total power production, E(P), of all of the turbines on the wind farm was formulated

314 as follows:

1

s ,t ' s ,t 2
N N Nv N 2
315 E P st f U 1 1 vij (19)

i 1 s 1 t 1 j 1, j i

316 Maximizing the total power production is equivalent to minimizing its reciprocal, i.e., 1/E(P).

317 Hence, the objective function was defined as follows:

1
318 F z (20)
E P

319 where z comprises the turbine coordinates, equal to (xi, yi, , xNT , y NT ). The wind farm layout

320 design was converted to a mathematical constrained optimization problem, in which F(z) was
321 minimized in the presence of inequality constraints, i.e., Eqs. (15) and (16).
16
ACCEPTED MANUSCRIPT

322 3.2 Optimization algorithm

323 In the literature, there are various algorithms for solving the micro-siting problem of multiple

324 turbines, such as Monte Carlo simulation (Marmidis et al., 2008), genetic algorithms (Chen et

325 al., 2015), and the ES (Kusiak and Song, 2010). In this study, the CMA-ES (Hansen, 2006), a

326 more powerful variant of the ES that considers the correlation of turbine positions, was used to

327 solve the optimization problem. The CMA-ES has been applied to the HAWT micro-siting

328 problem and performed exceptionally well (Wagner et al., 2011). The algorithm is briefly

329 covered here. For detailed formulations, please refer to a study by Hansen (2006).

330 The general form of a solution, z, at population k and generation g of the CMA-ES is denoted

331 z kg = ( x1, gk , y1, gk , x2, gk , y2, gk ,, xN g,k , yN g,k )T with z kg A 2 N . A population at generation g+1,
332 z kg 1 , can be drawn from a multivariate normal distribution with a mean m(g) A 2 N and a

333 covariance matrix ((g))2C(g) A 2 N 2 N :

334 z kg 1 : N C ,
m g , g
2
g
k 1, 2, L , n (21)

335 where (g) is the step size at generation g, and n is the total population considered. In this study,

336 Eq. (20) was subsequently evaluated using samples given by Eq. (21). By sorting the objective

337 functions, i.e., Eq. (20), in ascending order, the top e populations of z(g+1) were selected as the

338 elites for recombination to obtain the mean. The mean of the search distribution was attained

339 by taking a weighted average of the elites:

e
340 m g 1 i z i:gn1 (22)
i 1

341 where e equals n/2 rounded toward negative infinity, i is the weighting factor with the sum of

all weighting factors equal to unity (Hansen, 2006), z i:n


g 1
342 is the ith elite selected from the n

343 populations. The covariance matrix of the search distribution at generation g+1 was

344 correspondingly adapted:

17
ACCEPTED MANUSCRIPT

ccov

T
C g 1 1 ccov C g Pc g 1 Pc g 1
cov
345 z g 1 m g z g 1 m g T (23)
1 e
ccov 1 i
i:e i:e
g g
cov
i 1

346 where ccov is the learning rate for the covariance matrix adaption, cov is the weighting

347 parameter, and Pc A 2 N is the evolution path. To improve the overall efficiency of the CMA-
348 ES, the step size of C(g) at generation g+1 was updated in the course of optimization:

P g 1
c
349 g 1
g
exp 1 (24)
d


E N 0, I


350 where c is the learning rate for step size control, d is the damping parameter for step size

351 control, P A 2 N is the conjugate evolution path, and || || is the Euclidean norm operator.

352 To avoid wasting computational effort, a new stopping criterion monitoring the convergence of

353 the algorithm was defined. If the following criterion was fulfilled, the algorithm was terminated:

g
354 (25)
1

355 where is the minimum step size ratio. Eq. (25) was also an indicator of convergence.

356 The optimization process of the placement of VAWTs is summarized as follows.

357 1. Initialization of a list of algorithmic parameters. These are parameters related to the

358 following.

359 Wind turbine: number, N; diameter, D; coordinates, (xi, yi); and power curve, f (v).

360 Wake model: spreading rates, kW, kL, and kU; spreading constant, b; and velocity deficit, v.

361 Wind farm: L W (length width) and minimum turbine spacing, lmin.

362 Wind distributions: wind direction, p(), and wind speed, pv(v).

363 CMA-ES (Hansen, 2006): maximum generation number, G; population size, n; elite size,
364 e; mean, m(1); covariance matrix, C(1); step size, (1); minimum step size ratio, .

18
ACCEPTED MANUSCRIPT

365 2. Generate n populations by Eq. (21), in which mutations by covariance matrix are made. If

366 either Eq. (15) or Eq. (16) is violated for samples at a given population number, a new set

367 of samples is produced until all of the constraints are fulfilled.

368 3. Evaluate Eq. (20) for each population, sort the function values in ascending order, and select

369 the top e populations as elites.

370 4. Recombine the elites using Eq. (22) and update the mean, m(g+1), for the next generation.

371 5. Adapt the covariance matrix, C(g+1), and the step size, (g+1), by Eqs. (23) and (24),

372 respectively.

373 6. Repeat procedures 2-5 until either the stopping criterion, i.e., Eq. (25), is satisfied or the

374 maximum generation number, G, is reached. The mean of elites in the final generation is

375 used as the solution.

376 4. Application to Wind Farms

377 In this section, the optimal placement of turbines is demonstrated in three wind farm scenarios.

378 The characteristic parameters of the wind turbine in all of the scenarios are kept same and shown

379 in Table 2.

380 Table 2. Characteristic parameters of the VAWT

D Prated vrated vcut-in vcut-out P0 v1/2 c0

3.0 m 3.0 kW 12 m/s 3 m/s 25 m/s 3,390 w 8.7 m/s 0.62

381 Based on Eq. (11), the power curve of this wind turbine is plotted in Figure 6.

19
ACCEPTED MANUSCRIPT

3.5

3
Power production (kW)

2.5

1.5

0.5

0
3 5 7 9 11 13 15 17 19 21 23 25 26
382 Wind speed (m/s)

383 Figure 6. Power curve of the VAWT

384 Table 3 presents the parameters of the wake model for the layout design. The wind direction is

385 split into 24 equal-width bins, each of which represents 15. The wind speed from 0 to 28 m/s

386 is similarly discretized into 14 equal-width bins. For the CMA-ES, the population size is set as

387 n = 4+ln(32N), and the elite number equals e = n/2. Note that both n and e are rounded toward

388 negative infinity. The maximum generation number is G = 1,000, which is large enough to

389 ensure the convergence of the optimization in this study. This set of CMA-ES parameters is

390 applied in all of the scenarios.

391 Table 3. Model parameters of the wake model

kW kL b v kU

0.10 0.065 0.80 0.40 0.045

392 4.1 Wind farm scenario 1: Verification

393 To test the capability of the layout design algorithm, simple wind distributions with a dominant

394 wind direction ranging between 30 and 60 were adopted (see Figure 7). Figure 7(a) presents

395 the wind direction distribution from 0 to 360, and the values along the radial direction show

396 the probability of occurrence. Figure 7(b) presents the wind speed distribution sampled from a

20
ACCEPTED MANUSCRIPT

397 Weibull distribution with a = 12 and b = 2.

90 0.50 0.15
(a) (b)
120 60

0.12
150 0.25 30

Probability density
0.09
180 0
0.06

210 330
0.03

240 300
270 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27
398 Wind speed (m/s)

399 Figure 7. Distributions of wind direction and speed: (a) wind direction and (b) wind speed

400 For the purposes of illustration, four wind turbines were selected to optimize the cluster

401 configuration. The wind farm occupied a 24 m 24 m area. Figure 8 shows the initial and

402 optimized layouts. To form the initial layout, turbines were placed at equal distances from each

403 other, and a distance of 1D from the wind farm boundaries was reserved to prevent frequent

404 violations of constraints in the course of computation. In the optimized configuration, turbines

405 1 and 4 were shifted toward the upper left and lower right corners, respectively. Turbines 2 and

406 3 were placed such that the four turbines were roughly perpendicular to the prevailing wind

407 direction. This layout minimized the mutual interference of wind turbines to the greatest extent

408 possible and meanwhile fully used the available land area of the site. Therefore, no wake losses

409 were triggered in this configuration, and the expected power production reached the maximum.

410 The results clearly prove the efficacy of the optimization algorithm. However, for the placement

411 of dozens of turbines with a relatively flat wind direction distribution, intuitive judgment was

412 not adequate, and layout design was desperately needed.

21
ACCEPTED MANUSCRIPT

Initial layout Optimal layout


4
1

3 1 3

2
Width of the wind farm (y/D)

3
1

-1

-2
2

-3 2 4

-4
-4 -3 -2 -1 0 1 2 3 4
413 Length of the wind farm (x/D)

414 Figure 8. Configurations of turbines on the wind farm

415 4.2 Wind farm scenario 2: Complex wind direction

416 To simulate a more complex scenario, a layout design of 12 turbines involving a relatively flat

417 wind direction distribution was adopted. The wind farm size was set as 36 m 48 m. Figure 9

418 shows the wind direction and speed distributions. The wind blew from three major directions

419 with an interval of approximately 120. A Weibull distribution with a = 12 and b = 2 was

420 adopted as the wind speed distribution.

22
ACCEPTED MANUSCRIPT

90 0.3
0.15
(a) (b)
120 60
0.2
0.12
150 30

Probability density
0.1
0.09
180 0
0.06

210 330
0.03

240 300
270 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27
421 Wind speed (m/s)

422 Figure 9. Distributions of wind direction and speed: (a) wind direction and (b) wind speed

423 Figure 10 presents the initial and optimized layouts. Similar to scenario 1, in the optimized

424 configuration, the algorithm tended to place turbines at the boundaries to decrease mutual

425 wakes. According to the results of these two scenarios, turbines are best placed near the

426 boundaries of wind farms, with as large as possible interval distances. If a prevailing wind

427 direction exists, the turbines should be aligned to be perpendicular to that direction. To
i
428 quantitatively assess the wake loss of an individual turbine, the wake loss coefficient, CWL , was

429 defined as follows:

PI E Pi
430 i
CWL (26)
PI

431 where PI is the power production of a turbine under ideal conditions (no wake loss). The power

432 production and wake loss coefficient of each turbine were calculated, and they are listed in

433 Figure 11. It is clear that the power outputs of all of the turbines increased by a noticeable

434 amount (see Figure 11(a)) after optimization. To clearly set out the reduction of mutual

435 interference, Figure 11(b) presents the wake loss coefficients in the initial and optimal layouts.

436 The first four wind turbines suffered relatively lower wake losses in both the initial and

437 optimized layouts than the remaining turbines. The four turbines were located upstream of the
23
ACCEPTED MANUSCRIPT

438 other turbines when the wind blew in the direction with the largest probability. Therefore, they

439 had the largest expected power output and hence the lowest wake losses.

Initial layout Optimal layout


6
1 4 7 10
1
4 7 10
4
Width of the wind farm (y/D)

2
8

2
0 5 8 11

2 11

5
-2

-4 9
9 12
3 6
6 12
-6 3
-8 -6 -4 -2 0 2 4 6 8
440 Length of the wind farm (x/D)

441 Figure 10. Layouts of turbines on the wind farm

442 To evaluate the overall wake loss of the wind farm, Eq. (26) was rewritten as follows:

NPI E P
443 CWL (27)
NPI

444 where CWL is the overall wake loss coefficient of the wind farm. In the initial layout, an

445 evaluation of Eq. (27) gave CWL = 13.5%. In comparison, CWL decreased considerably to 4.3%

446 in the optimized layout. From the power production perspective, after the configuration of the

447 turbines was optimized, the total power output increased by 10.6%. Therefore, the effectiveness

448 of the optimization algorithm and the significance of the wind farm layout design were clearly

449 demonstrated.

24
ACCEPTED MANUSCRIPT

(a) Initial Optimized Ideal


1.9
Power production (kW)

1.8

1.7

1.6

1.5

1.4
1 2 3 4 5 6 7 8 9 10 11 12
Turbine No.
(b) Initial Optimized
20
Wake loss coefficient (%)

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12
450 Turbine No.

451 Figure 11. Power production and wake loss coefficients of individual turbines on the wind

452 farm: (a) power production and (b) wake loss coefficient

453 4.3 Wind farm scenario 3: Irregular site

454 In reality, a wind farm site does not necessarily have a regular shape. In this study, the layout
455 design was applied to a wind farm site with an irregular shape. In general, the boundary of an
456 irregular wind farm site can be approximated by a polygon with finite vertices. A polygon, C,

457 connected by an array of n0 vertices, i.e., V0, V1, , Vn0 = V0, can be seen as a piecewise linear

458 curve. Therefore, whether a turbine falls within the irregular wind farm can be converted into a

459 point-in-polygon problem. The winding number approach was introduced to solve the micro-

460 siting problem on irregular wind farm sites.

461 The winding number, (T, C), of a point, T, in relation to a polygon, C, was expressed as follows

462 (Hormann and Agathos, 2001):

1 n0 1
1 n0 1
TVi , TVi 1
463 T , C
2
i 2
arccos sign i (28)
i 0 i 0 TVi TVi 1

25
ACCEPTED MANUSCRIPT

464 where i is the winding angle between the vectors of TVi and TVi+1, and < > is the operator of

465 an inner product. In Eq. (28), sign(i), which was used to determine the winding orientation
466 from vertex Vi to vertex Vi+1 around the point T, was written as follows:

TVi x TVi x1
467 sign i sign (29)
TV y TVi y1
i

468 where TVi x and TVi y represent the x and y components of the coordinates of TVi , respectively,

469 and TVix1 and TViy1 represent the x and y components of the coordinates of TVi1 , respectively.
470 As a result, the winding number of a polygon (wind farm boundary) around a point (wind

471 turbine) was evaluated. If the winding number, (T, C), was not zero, the wind turbine was

472 considered to fall within the wind farm boundary. Otherwise, the turbine was outside the wind

473 farm site. Consequently, the constraint, (T, C) 0, for irregular wind farm sites replaced Eq.

474 (15) and further integrated with the layout design algorithm. Unlike the first two scenarios, in

475 this case, the initial locations of the turbines were generated randomly to accommodate the

476 irregular wind farm boundary constraints.

4
V3 V2

2
Width of the wind farm (y/D)

1
V1
0 V6 = V0

-1

-2

-3
V4 V5
-4
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
477 Length of the wind farm (x/D)

478 Figure 12. Wind farm site with an irregular boundary


26
ACCEPTED MANUSCRIPT

479 Figure 12 presents an irregular wind farm site with a concave L shape. Some building

480 rooftops have this kind of shape. In Figure 12, the wind farm site is represented by a polygon
481 with six vertices. Six wind turbines were nominated to be placed on this irregular wind farm.

482 Wind distributions collected from historical wind data (Kusiak and Song, 2010) were used in

483 this scenario. The wind speed distribution was approximated by a Weibull distribution with a

484 = 13 and b = 2. Figure 13 shows the distributions of the wind direction and speed. It can be

485 seen that the prevailing wind direction was approximately within the range of 75-105.

90 0.14
(a) (b)
120 60
0.12

150 0.5 30 0.1


Probability density

0.08
180 0
0.06

210 330 0.04

0.02
240 300
270 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27
Wind speed (m/s)
486

487 Figure 13. Distribution of wind direction and speed

488 After implementing the layout design algorithm with the new boundary constraints, the optimal

489 layout could be determined. Figure 14 presents the initial and optimized layouts of the VAWTs

490 on the L-shaped wind farm. In the optimized layout, the VAWTs are positioned on the

491 boundaries of the wind farm. As was proved quantitatively in the previous wind farm scenarios,

492 this kind of placement minimized the mutual interference of VAWTs and hence maximizes the

493 overall power production. Furthermore, as the prevailing wind blew in the y direction, the

494 downstream turbines 3 and 4 shifted away from the wakes of turbines 5 and 6. In addition,

495 turbine 1 moved to a position to get rid of the wake of turbine 2 when wind blew in the

496 prevailing direction. The efficacy of the layout design for the irregular wind farm site was

27
ACCEPTED MANUSCRIPT

497 clearly demonstrated.

4
4 1
3
1
3
4
3
2
Width of the wind farm (y/D)

1 2
2
5
0

-1

-2

6
-3
5 6 Initial layout
-4 Optimal layout
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
498 Length of the wind farm (x/D)

499 Figure 14. Layout of VAWTs on the irregular wind farm

500 5. Conclusion

501 In this study, a wake model was developed for straight-bladed VAWTs. To address the wake

502 asymmetry in the horizontal direction, the wake region was represented by two semi-elliptical

503 cones. In contrast, the wake was assumed to spread symmetrically in the vertical direction. The

504 velocity in the wake was derived based on continuity conservation. To derive the model

505 parameters via regression analyses, wind tunnel tests were performed at different BSRs. The

506 model-predicted results agreed well with the experimental data. The asymmetry was

507 quantitatively reflected in the spreading rates at the windward and leeward. The wake velocity

508 recovered very fast, and VAWTs in wind parks were suggested to be spaced somewhere of 6D

509 apart in the prevailing direction.

510 In the micro-siting problem, both wind direction, p(), and wind speed, pv(v), distributions

511 were considered to faithfully represent wind properties. A modified sigmoid function was used

28
ACCEPTED MANUSCRIPT

512 to capture the essential feature of a power curve in the growing phase. The cross-production

513 factor, , was introduced to determine whether a turbine fell to the windward or leeward of its

514 upstream counterparts. The expected power production, E(P), of a given number of turbines

515 was eventually formulated. The CMA-ES showed good performance in solving the micro-siting

516 problem. Several scenarios were considered to verify and illustrate the proposed turbine

517 placement method. In the simple scenario 1, the efficacy of the layout design was manifestly

518 proved. In the complicated scenario 2, the power output of the wind farm was raised by 10.6%

519 after optimization. Moreover, the overall wake loss coefficient, CWL, was decreased to 4.3%

520 from 13.5% in the initial layout. Hence, the significance of the layout design was clearly

521 demonstrated. In scenario 3, the cluster configuration was generalized to irregular wind farm

522 sites. The winding number method performed effectively to handle the irregular constraints and

523 could be easily integrated with an optimization algorithm.

524 In the future, the BSR and turbulence intensity are suggested being considered to develop a

525 more advanced wake model as they might have certain impacts on the wake characteristics.

526 Acknowledgments

527 The work described in this study was supported by a grant from the Research Grants Council

528 of the Hong Kong Special Administrative Region, China (Project No. 9041889 [CityU

529 115413]). Gratitude is also given to Mr. CHAN Kwok Keung, Mr. TAI Wing Hing, Prof.

530 CHEUNG Chun Kuen, and Dr. HE Yun Cheng at the wind tunnel facility for their expertise

531 and assistance.

532 References

533 Tai F.Z., Kang K.W., Jang M.H., Woo Y.J., Lee J.H., 2013, Study on the analysis method for the vertical-axis
534 wind turbines having Darrieus blades, Renewable Energy, 54, 26-31.

535 Lee Y.T. and Lim H.C., 2015, Numerical study of the aerodynamic performance of a 500 W Darrieus-type vertical-
536 axis wind turbine, Renewable Energy, 83, 407-415.

537 Lam H.F., Peng H.Y., 2017, Measurements of the wake characteristics of co- and counter-rotating twin H-rotor
538 vertical axis wind turbines, Energy, 131, 13-26.

539 Chen L., MacDonald E., 2014, A system-level cost-of-energy wind farm layout optimization with landowner

29
ACCEPTED MANUSCRIPT

540 modeling, Energy Conversion and Management, 77, 484494.

541 Jensen N.O., 1983, A note on wind generator interaction, RISO National Laboratory, Roskilde, Denmark,
542 Technical report Riso-M-2411.

543 Katic I., Hjstrup J., Jensen N., 1986, A simple model for cluster efficiency, EWEC, 86.

544 Mosetti G., Poloni C., Diviacco B., 1994, Optimization of wind turbine positioning in large wind farms by means
545 of a genetic algorithm, Journal of Wind Engineering and Industrial Aerodynamics, 51, 105-116.

546 Grady S.A., Hussaini M., Abdullah M.M., 2005, Placement of wind turbines using genetic algorithms, Renewable
547 Energy, 30, 259-270.

548 Saavedra-Moreno B., Salcedo-Sanz S., Paniagua-Tineo A., Prieto L., Portilla-Figueras A., 2011, Seeding
549 evolutionary algorithms with heuristics for optimal wind turbines positioning in wind farms, Renewable
550 Energy, 36, 2838-2844.

551 Wan C., Wang J., Yang G., Gu H., Zhang X., 2012, Wind farm micro-siting by Gaussian particle swarm
552 optimization with local search strategy, Renewable Energy, 48, 276-286.

553 Turner S.D.O., Romero D.A., Zhang P.Y., Amon C.H., Chan T.C.Y., 2014, A new mathematical programming
554 approach to optimize wind farm layouts, Renewable Energy, 63, 674-680.

555 Marmidis G., Lazarou S., Pyrgioti E., 2008, Optimal placement of wind turbines in a wind park using Monte Carlo
556 simulation, Renewable Energy, 33, 1455-1460.

557 Kusiak A. and Song Z., 2010, Design of wind farm layout for maximum wind energy capture, Renewable Energy,
558 35, 685-694.

559 Islam M., Ting D., Fartaj A., 2008, Aerodynamic models for Darrieus-type straight-bladed vertical axis wind
560 turbines, Renewable and Sustainable Energy Reviews, 12, 1087-1109.

561 Howell R., Qin N., Edwards J., Durrani N., 2010, Wind tunnel and numerical study of a small vertical axis wind
562 turbine, Renewable Energy, 35, 412-22.

563 McLaren K.W., 2011, A numerical and experimental study of unsteady loading of high-solidity vertical axis wind
564 turbines, Ph.D. Thesis, McMaster University, Canada.

565 Bianchini A., Ferrara G., Ferrari L., 2015, Design guidelines for H-Darrieus wind turbines: Optimization of the
566 annual energy yield, Energy Conversion and Management, 89, 690-707.

567 Balduzzi F., Bianchini A., Carnevale E.A., Ferrari L., Magnani S., 2012, Feasibility analysis of a Darrieus vertical-
568 axis wind turbine installation in the rooftop of a building, Applied Energy, 97, 921-929.

569 Saeidi D., Sedaghat A., Alamdari P., Alemrajabi A.A., Aerodynamic design and economical evaluation of site
570 specific small vertical axis wind turbines, Applied Energy, 101, 765-775.

571 McNaughton J., Billard F., Revell A., 2014, Turbulence modelling of low Reynolds number flow effects around a
572 vertical axis turbine at a range of tip-speed ratios, Journal of Fluids and Structures, 47, 124-138.

573 Elkhoury M., Kiwata T., Aoun E., 2015, Experimental and numerical investigation of a three-dimensional vertical-

30
ACCEPTED MANUSCRIPT

574 axis wind turbine with variable-pitch, Journal of Wind Engineering and Industrial Aerodynamics, 139, 111-
575 123.

576 Tjiu W., Marnoto T., Mat S., Ruslan M.H., Sopian K., 2015, Darrieus vertical axis wind turbine for power
577 generation II: Challenges in HAWT and the opportunity of multi-megawatt Darrieus VAWT development,
578 Renewable Energy, 75, 560-571.

579 Simo Ferreira J.S., 2009, The near wake of the VAWT: 2D and 3D views of the VAWT aerodynamics, Ph.D.
580 Thesis, Delft University of Technology, Netherlands.

581 Kinzel M., Mulligan Q., Dabiri J.O., 2012, Energy exchange in an array of vertical-axis wind turbines, Journal of
582 Turbulence, 13.

583 Hezaveh S.H. and Bou-zeid E., 2013, Large eddy simulation of vertical axis wind turbine to optimize farm design,
584 66th Annual Meeting of the APS Division of Fluid Mechanics, 58, Pennsylvania.

585 Tescione G., Ragni D., He C., Simo Ferreira C.J., van Bussel J.W., 2014, Near wake flow analysis of a vertical
586 axis wind turbine by stereoscopic particle image velocimetry, Renewable Energy, 70, 47-61.

587 Peng H.Y., Lam H.F., Lee C.F., 2016, Investigation into the wake aerodynamics of a five-straight-bladed vertical
588 axis wind turbine by wind tunnel tests, Journal of Wind Engineering and Industrial Aerodynamics, 155, 23-
589 35.

590 Lam H.F. and Peng H.Y., 2016, Study of wake characteristics of a vertical axis wind turbine by two- and three-
591 dimensional computational fluid dynamics simulations, Renewable Energy, 90, 386-398.

592 Ross I. and Altman A., 2011, Wind tunnel blockage corrections: Review and application to Savonius vertical-axis
593 wind turbines, Journal of Wind Engineering and Industrial Aerodynamics, 99, 523-538.

594 Chen T.Y. and Liou L.R., 2011, Blockage corrections in wind tunnel tests of small horizontal-axis wind turbines,
595 Experimental Thermal and Fluid Science, 35, 565-569.

596 Vermeer L.J., Sorensen J.N., Crespo A., 2003, Wind turbine wake aerodynamics, Progress in Aerospace Sciences,
597 39, 467-510.

598 Gonzlez J.S., Rodriguez A.G.G., Mora J.C., Santos J.R., Payan M.B., 2010, Optimization of wind farm turbines
599 layout using an evolutive algorithm, Renewable Energy, 35, 1671-1681.

600 Peng H.Y. and Lam H.F., 2016, Turbulence effects on the wake characteristics and aerodynamic performance of
601 a straight-bladed vertical axis wind turbine by wind tunnel tests and large eddy simulations, Energy, 109,
602 557-568.

603 Manwell J.F., McGowan J.G., Rogers A.L., 2009, Wind energy explained: Theory, design and application, 2nd ed.,
604 London: John Wiley & Sons.

605 Eroglu Y. and Sekiner S.U., 2012, Design of wind farm layout using ant colony algorithm, Renewable Energy,
606 44, 53-62.

607 Wagner M., Veeramachaneni K., Neumann F., OReilly U., 2011, Optimizing the Layout of 1000 Wind Turbines,
608 European Wind Energy Association, Annual Event.

31
ACCEPTED MANUSCRIPT

609 Chen Y., Li H., He B., Wang P., Jin K., 2015, Multi-objective genetic algorithm based innovative wind farm layout
610 optimization method, Energy Conversion and Management, 105, 1318-1327.

611 Hansen N., 2006, The CMA evolution strategy: A comparing review, StudFuzz, 192, 75-102.

612 Hormann K. and Agathos A., 2001, The point in polygon problem for arbitrary polygons, Computational
613 Geometry: Theory and Applications, 20, 131-144.

32
ACCEPTED MANUSCRIPT

Highlights

Development of an analytical wake model for H-rotor VAWTs.


Implementation of wind tunnel tests for the updating of the model parameters.
Approach to determine the position of a VAWT relative to its upstream counterparts.
Application of the proposed wake model to the micro-siting problem of VAWTs.
Extension of the cluster configuration of VAWTs to irregular wind farm sites.

Anda mungkin juga menyukai