K.

Thomsen
e-mail: [email protected]

Potentials for Site-Specific Design of MW Sized Wind Turbines

The potential for site specic design of MW sized wind turbines is quantied by comparing design loads for wind turbines installed at a range of different sites. The sites comprise on-shore normal at terrain stand-alone conditions and wind farm conditions together with offshore and mountainous complex terrain wind farms. The design loads are established for a 1 MW active stall regulated wind turbine with the aeroelastic code HAWC. The load analysis is limited to fatigue loads. We do not consider ultimate loads in this paper. The results illustrate the differences in design wind conditions for different sites and the related differences in design loads for the 1 MW wind turbine. Based on the difference in the design loads, the potentials for site specic design of the wind turbine main components are identied. The results show that the variation in aerodynamically driven loads and energy production can be more than 50% between the different sites. It is concluded that site specic design is feasible for some of the main components. In particular, site specic changes are feasible for the tower, nacelle components, and for the blades in the apwise direction. It is also evaluated whether the IEC61400-1 standard [see Ref. [4], International Electrochemical Commission (1999)] is representative for the different sites. A comparison with design loads based on the IEC61400-1 illustrates that the six different sites can be described by the standard design classes.

P. Fuglsang
Wind Energy and Atmospheric Physics Department Ris National Laboratory, DK-4000 Roskilde, Denmark

G. Schepers
Netherlands Energy Research Foundation, Westerduinweg 3, Petten NL-1755 ZG, The Netherlands

1 Introduction
Worldwide there is a great potential for on-shore and also offshore wind farms where a large number of units are installed. The large number of units in similar conditions will enable us to adapt wind turbines for specic site conditions to further reduce cost of energy and increase the competitiveness of wind turbines compared with other energy sources. The differences in the design of wind turbines for different sites depends on the differences in the wind turbine loads which are a result of the specic design wind conditions. Thus, it is only possible to modify a wind turbine design for a specic site if design wind conditions can be described in detail and the loads can be predicted with a high degree of accuracy. In this work, estimated design loads are presented for a wind turbine installed at six different sites. The sites comprise combinations of stand-alone and wind farm sites in combination with different terrain types. In total, six sites are considered, and the types of terrain range from normal at terrain to off-shore to complex terrain. Hence, the six sites cover variations in the mean wind speed, stand-alone conditions versus wind farm conditions, onshore versus offshore wind farms, and at terrain versus complex terrain. We have chosen to identify the potentials for site-specic design optimization by revealing the differences in loads and power production for the different sites. If the difference in loads for the different sites is large, then the basis for design changes exists. Thus, the main objective of the present work is to answer the question: Is the difference in fatigue design loads when we consider six different sites sufciently large to justify design adjustments? The approach followed is to establish the differences between the sites in terms of wind eld parameters and subsequently establish the design loads using aeroelastic simulations. This work is part of the European research project, Site-Specic

Design Optimization of Wind Turbines Based on Numerical Optimization. In this project, wind turbines are cost-optimized for different sites. The variations in loads for the different sites are evaluated in terms of component costs, which leads to redesign of the wind turbine components. This changes the assumptions for the load predictions, and through an iterative optimization procedure, the nal design of the wind turbines for the specic sites is found. In the present investigation, only results regarding differences in loads for the different sites are given. In the next section, the assumptions for the wind conditions at the six sites are described in detail. Following this, assumptions for the load simulations are given, and the results are described and discussed. Finally, a comparison with the standard IEC61400-1 design classes is carried out.

2 Assumptions for the Wind

As mentioned in the Introduction, the investigation is based on six different sites, which represent different types of terrain and operational conditions. The rst site is a reference site and represents a stand-alone wind turbine in a normal at terrain condition. The next site is similar in terms of overall wind condition, but now wind farm conditions are accounted for. The third and fourth sites represent low and high wind speed modications, respectively. The two last sites represent offshore and mountainous terrain conditions, respectively. In summary, the six different sites are abbreviations in brackets: 1. 2. 3. 4. 5. 6. On-shore normal terrain reference, ref Reference, but wind farm, refwf Wind farm with low mean wind speed, low Wind farm with high mean wind speed, high Off-shore wind farm, Off. Mountainous complex terrain wind farm, Com.

The different sites represent the variety of sites we have experience with. In all casesexcept wind climate 1wind farm operation is assumed. This is based on the fact that today almost all commer-

Table 1 Mean wind speed U mean and Weibull scale factor A for the six types of wind climate Wind climate 1 2 3 4 5 6 ref ref wf low high Off. Com. U mean m/s 7.1 7.1 6.2 8.9 8.9 7.1 A m/s 8 8 7 10 10 8

Table 2 Roughness length z 0 for different wind climate wc U hub m/s] 6 8 10 12 14 16 18 20 22 24 wc 1,2,3,4 m 0.025 0.025 0.025 0.025 0.025 0.025 0.025 0.025 0.025 0.025 wc 5 m 5.6 10 5 1.1 10 4 1.9 10 4 2.9 10 4 4.2 10 4 5.8 10 4 7.7 10 4 1.0 10 3 1.3 10 3 1.6 10 3 wc 6 m 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1

cial wind turbines are designed for use in wind farms. Thus, most of the sites include the changes in wind eld parameters caused by wind farm operation. The differences in wind conditions for the different wind climates are described using the following parameters: Mean wind speed U hub Roughness length z 0 Longitudinal turbulence intensity I u Transverse turbulence i / u , i ( v , w ) Turbulence length scale L i , i ( u , v , w ) Coherence decay factor A i , i ( u , v , w )

wind climate 1-5. For the complex terrain site, a wind vector slope of 10 is assumed, in order to account for the modied ow due to hills. 2.3 Vertical Wind Prole. The vertical shear of the wind speed can be described by the traditionally used logarithmic prole: U z U hub ln z / z 0 ln z hub / z 0 (2)

For each of the six sites, all of these parameters must be chosen. As can be seen from the list of wind eld parameters, a somewhat simple approach is followed for the modeling of the wind eld in a wind farm. It is well-known that several wind eld parameters change in a wind turbine wake, e.g., mean wind speed, wind shear, turbulence intensity, turbulence spectrum, turbulence length scales, and coherence are different in the wake than in the free wind ow. However, a recent study 1 has illustrated that with respect to fatigue loadsthe effect from all these parameters can be modeled by an increased turbulence intensity of the longitudinal wind component. This approach is followed for the design loads. With respect to power production, it is necessary to take the reduced wind speed in a wind turbine wake into account. In order to model this, we propose the Ris PARK model 2 for the fraction of time the turbine operates in wake. Then the total production is a combination of the production at reduced wind speed in the wake and the production at the free wind speed. The combination of these is determined from the wind farm lay-out and the wind direction distributions. The wind eld parameters are summarized later in this section. The rationale for the parameters is outlined in the following sections. The individual parameters describing the wind climates are described in this section and the actual parameters for the six different types of wind climate are derived. 2.1 Mean Wind Speeds. For the distribution of mean wind speeds, a Rayleigh description is used a Weibull distribution with a shape factor of 2.0. In this case, the relation between the annual mean wind speed U mean and the scale factor A is: U mean A /2. (1)

where U ( z ) is the wind speed at height z, U hub the wind speed at hub height z hub and z 0 the roughness length. The roughness length is assumed to be 0.025 m for wind climates 1-4. For the complex terrain site, a roughness length of 0.1 m is assumed. For the offshore site, the roughness increases with wind speed, and the Charnock relation 3 is proposed: z 0
A / g 2U 2 hub

ln2 z / z 0

(3)

where A 0.018 for shallow water, g is the gravity constant, is rma n constant ( 0.4) and U hub is the wind speed. the von Ka Using this expression the roughness length has been calculated for the offshore site, and for all sites the roughness lengths are summarized in Table 2. 2.4 Turbulence Parameters. The turbulence spectrum is described as a Kaimal spectrum and specied according to IEC61400-1 4: S i f 4 i2 L i / U hub
1 6 f L i / U hub (5/3)

(4)

In Table I, the parameters for each wind climate are given. None of the mean wind speed values chosen are extreme values, but the variation covers the intermediate range of mean wind speeds for realistic sites. Thus, low wind sites with even lower wind speeds and high wind sites with higher wind speeds can be found, but we believe that the differences in the chosen values are sufcient to identify the inuence on the loads. Thus, if sitespecic design is feasible with these parameters it also will be for sites with larger differences in wind parameters. 2.2 Mean Wind Vector Slope. The slope of the mean wind vector with reference to a horizontal plane is assumed to be 0 for

where S i ( f ) is the spectrum of the i component ( u , v or w ), i is the standard deviation of the i component, L i is the turbulence length scale for the i component, and U hub is the 10 minute mean wind speed. In IEC61400-1, the length scales for heights above 30 m are specied as: L u 164.4 m, L v 54.8 m and L w 13.4 m. These length scales are assumed for the wind climates 1,2,3,4 and 5. For wind climate 6 complex, the length scales are reduced by a factor of 2, based on measurements in complex terrain 5. The coherence is modeled as an expontential decay model IEC61400-1 with a decay factor of A u 8.8 for wind climates 1,2,3,4 and 5 for the longitudinal coherence. For wind climate 6, a decay factor of A u 17.6 is assumed 5. The ratios between the coherence decay parameters are chosen 6 as A v 0.75A u and A w 0.9A u . The turbulence intensity of the longitudinal turbulence component I u u / U hub depends on the roughness length as follows, assuming neutral conditions: I u 1 . ln z / z 0 (5)

Table 3 Design turbulence intensity I u,tot for different wind climate wc U hub m/s 6 8 10 12 14 16 18 20 22 24 wc 1 0.132 0.132 0.132 0.132 0.132 0.132 0.132 0.132 0.132 0.132 wc 2,3,4 0.183 0.178 0.172 0.166 0.160 0.154 0.150 0.148 0.146 0.145 wc 5 0.146 0.142 0.137 0.131 0.125 0.119 0.116 0.115 0.114 0.114 wc 6 0.205 0.201 0.195 0.190 0.185 0.180 0.177 0.175 0.173 0.172

Table 4 Summary of wind parameters. T. 2 refers to Table 2 and T. 3 refers to Table 3. No. U hub Slope z0 Lu Lv Lw Au Av Aw Iu v /u w u 1 ref 7.1 0 0.025 164.4 54.8 13.4 8.8 6.2 7.9 0.132 0.8 0.5 2 refwf 7.1 0 0.025 164.4 54.8 13.4 8.8 6.2 7.9 T. 3 0.8 0.5 3 low 6.2 0 0.025 164.4 54.8 13.4 8.8 6.2 7.9 T. 3 0.8 0.5 4 high 8.9 0 0.025 164.4 54.8 13.4 8.8 6.2 7.9 T. 3 0.8 0.5 5 Off. 8.9 0 T. 2 164.4 54.8 13.4 8.8 6.2 7.9 T. 3 0.8 0.5 6 Com. 7.1 10 0.1 82.2 27.4 6.7 17.6 12.3 15.8 T. 3 1.0 0.8

Thus, from the values of z 0 Table 2, the free ow turbulence intensity can be calculated. However, in most of the wind climates, the turbines operates in wind farm conditions. In order to take this into account, the Frandsen method 1 is followed. According to this method, the wake operation is accounted for by modifying the turbulence intensity:
2 I u ,tot I 2 f I w

(6)

where I f is the free ow turbulence intensity, I w is the wake turbulence intensity and I u ,tot is the total design turbulence intensity. The wake contribution to the turbulence intensity is calculated from: I k nC T 2 w 2 s (7)

equivalent turbulence intensity accounted for the combined effects of wake operation, but for the power production, the reduced wind speed in the wake must be taken into account. We use an approach described recently 7. In this method, the total power is a combination of the production in the free wind and the production in the wake of the upwind turbines. For the latter, we propose the Ris PARK model: U w U a 1 1 1 C T

D D 2 kx

(8)

where C T is the upwind turbine thrust coefcient, s is the relative spacing in rotor diameters and k n is a wind farm conguration parameter. A value of k n 0.4 is proposed corresponds to severe wind farm operation and a inter-turbine spacing of s 5 rotor diameters is used. Using the above expressions, the design turbulence intensities for the different wind climates are calculated, Table 3 and Fig. 1. For the ratios of transverse turbulence relative to the longitudinal turbulence, v 0.8 u and w 0.5 u are proposed for wind climate 1,2,3,4, and 5. For wind climate 6 complex, v 1.0 u and w 0.8 u are used 6. 2.5 Wind Characteristics for Power Production. For the calculation of the power production, we cannot use the same approach as used for the design loads. For the design loads, an

where U w is the reduced wind speed, U a is the free wind speed, C T is the turbine thrust coefcient, D is the rotor diameter and x is the distance to the upwind turbine. k is the wake decay constant which depends on the free ow turbulence intensity, k 0.5I u . For the assumed wind farm conguration, the fraction of time where the turbine operates in wake is calculated to be 37% same Ref. as above. This is based on a calculation of the expansion angle of the wake in a 5-diameter spaced full wind farm conguration. In principle, the turbulence intensity dependency of k, means that the fraction of time in wake depends on the turbulence intensity. Since the turbulence intensity changes with wind speed, a different fraction of wake operation time will be expected at each wind speed. However, the inuence of changing k on this fraction is found to be small and only the inuence of k on the wake wind speed will be taken into account. Thus, the total production is a combination of production at reduced wind speed 37% and production at free wind speed 63%. For the reduced wind speed, the probability of the wind speed is calculated from the free wind Weibull distribution, but the power is based on the reduced wind speed in the wake 8. The wind eld parameters are summarized in Table 4.

3 Aeroelastic Simulations
The design loads are established using the aeroelastic model HAWC 8, which is the Ris state-of-the-art aeroelastic code. It is a time-domain nite element model with a traditional blade element momentum aerodynamic model. The model divides the turbine into substructures and couples these together in coupling nodes. The elements are Timoshenko beam elements. The inow to the model is a combination of mean ow with shear and wind vector slope and a stochastic turbulence eld. 3.1 Description of the Wind Turbine. The turbine used in the load simulations is a BONUS turbine. It is a 3-bladed upwind active stall controlled machine. The turbine operates at constant speed and xed yaw. The nominal power is 1 MW and the blades are LM26.1 produced by LM Glasber A/S. The turbine is believed to be representative for the Danish concept type of MW wind turbines with a stiff three-bladed upwind rotor. Main dimensions are given in Table 5.

Fig. 1 Comparison of turbulence intensities for wind climates 1-6 wc

Table 5 Technical specications of the BONUS turbine Nominal power Rotor diameter Rotor speed Power regulation Blade length Airfoils Tower type Tower height 1 MW 54.2 m 22/15 rpm active stall 26.1 m FFA3 NACA 63200 tubular 50.0 m

Table 6 Annual energy production GWh for the six different sites No. Wind speed Weibull C Wake prob. Production Normalized 1 ref 7.1 2.0 0.00 2.48 1.00 2 refwf 7.1 2.0 0.37 2.22 0.89 3 low 6.2 2.0 0.37 1.61 0.65 4 high 8.9 2.0 0.37 3.40 1.37 5 Off. 8.9 2.0 0.37 3.40 1.37 6 Com. 7.1 2.0 0.37 2.22 0.89

3.2 Load Cases and Analysis. We only consider load cases where the turbine operates in normal power production condition. The wind speed range from 6-26 m/s is divided in intervals of 2 m/s and for each wind speed interval the loads are represented by two time series. One of these has an yaw error of 6and the other has an yaw error of 6. This is due to the fact that some yaw misalignment can be expected and we assume it to be 6and evenly divided between positive and negative values. The duration of the time series is 300 s. The resulting time series are Rainow counted and the combined life time fatigue loads are calculated by integration over all load cases. The loads considered are: Flapwise root bending moment M f la p Edgewise root bending moment M edge Rotor thrust force F thrust Rotor tilt bending moment M tilt Rotor yaw bending moment M yaw Shaft torque M sht Tower bending moment along-wind M tbL Tower bending moment sideways M tbT

4 Results
Due to the different mean wind speed distributions for the sites, signicant differences are seen in the annual energy production, Table 6. The wind farm operation condition for sites 2 to 6 inuences this as well, and comparing energy production for wc1 and wc2 it is seen that the wind farm efciency is approx. 90%. It is seen that at the high wind speed sites wc4 and wc5 the production is 37% higher than the production for the reference site. The difference in energy production between the high wind speed sites and the low wind speed site is approximately a factor of 2. In order to illustrate the different loads at the different sites, the 1 Hz damage equivalent load ranges are investigated, Fig. 2. The 1 Hz damage equivalent load is the constant-amplitude load signal causing the same damage as the real load signal - when repeated a number of times corresponding to 1 Hz. All of the analyzed load components have the same tendencies and only three loads are illustrated. For wind climate 2, 3, and 4, the loads are similar, since the wind inow is similar. The only difference between these sites is the Weibull scale parameter, which does not inuence the load at each wind speed. A large difference between the loads for the different sites is seen. The differences in the turbulence intensities versus wind speed, Fig. 1, are seen in the load comparison as well. Comparing wc1 loads with wc5 offshore loads, it is seen that these are similar. At low wind speeds, the wc5 loads are slightly higher than the reference loads, and at high wind speeds the situation is opposite. For all wind speeds, the wc 2,3, and 4 loads are larger than the reference loads, and the wc6 loads are largest. These tendencies follow the behavior of the turbulence intensity 1, indicating the high importance of this parameter. The loads are integrated over all load cases, and the results are illustrated as normalized 107 life time equivalent loads, Table 7. Now the difference between wc 2, 3, and 4 becomes clear. Loads for the reference case combined with wind farm operation wc2

Fig. 2 Fatigue equivalent loads for the different wind climates versus wind speed, from top: M flap , M tilt , M tbL

Table 7 Normalized life time fatigue loads for the six different hler curve SN exponent of m 10 is used for blade sites. A Wo loads and a m 4 for all other loads. No. M f lap M edge F thrust M sht M tilt M yaw M tbL M tbT 1 ref 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 2 refwf 1.17 1.00 1.26 1.25 1.24 1.23 1.25 1.18 3 low 1.12 0.97 1.17 1.16 1.15 1.13 1.14 1.06 4 high 1.33 1.06 1.44 1.34 1.43 1.42 1.49 1.42 5 Off. 1.10 1.06 1.15 1.08 1.15 1.14 1.19 1.17 6 Com. 1.58 1.05 1.60 1.59 1.63 1.65 1.64 1.63

Table 8 Main parameters for the IEC61400-1 classes WTGS class Wind speed Weibull A Weibull C I 10.00 11.28 2.00 II 8.50 9.59 2.00 III 7.50 8.46 2.00 IV 6.00 6.77 2.00

are approximately 25% higher than the stand-alone reference case wc1. Reducing the average wind speed wc3 reduces this factor to approximately 15%, and increasing the wind speed wc4 increases the factor to approximately 40%. This illustrates the importance of the mean wind speed distribution to the loads. Moving the wind turbines offshore wc5 increases the mean wind speed, but reduces at the same time the turbulence intensity, due to the reduced roughness length. The resulting loads are approximately 15% larger than the reference case. Comparing the onshore wind farm site wc2 with the offshore site, the loads are lower for the offshore siteeven with a signicantly increased mean wind speed. For the complex terrain case wc6, several wind parameters are different than the reference case, but the same Weibull parameters are assumed. The resulting life time loads are approximately 60% higher than the reference case. Comparing the at terrain wind farm site wc2 with the complex terrain site, the inuence of the terrain is seen to increase the loads approximately 30% from 1.25 to 1.60.

IEC class IIIB loads seem to be very close. The fatigue loads at each wind speed Fig. 4 are almost identical and the IEC life time loads are only slightly higher due to the higher Weibull A parameter 8.46 m/s compared to 8.00 m/s. The loads for the low wind site wc3 are very close to the IEC class IVB loads and the high wind site wc4 loads are seen to agree well with the IEC class IIB loads. The turbulence intensities for the offshore site wc5 are signicantly lower than both IEC classes values. This means that the loads for the wc5 site should be compared with an IEC class with higher mean wind speed than was used for the wc5 loads. In the IEC classication, offshore sites are to be treated separately. The complex terrain site wc6 has higher turbulence intensities than the other sites and the turbulence values compare well with the IEC high turbulence intensity class. Comparing the life time loads, it is seen that the IEC class IIA loads ts well with wc6 loads. The highest wind speed chosen for the six wind climates is lower than the value used in the IEC class I. Due to this, the IEC class IA would be a conservative design basis for all our six sites. If, on the other hand, we consider a complex terrain site with a very high mean wind speed, the loads would be even higher than the IEC class IA. For such a situation, the design wind conditions must be more severe than class IA.

6 Conclusions
This investigation has revealed the differences in power production and loads for six specic sites. The sites are considered to be realistic without being extreme, and differences in energy production up to a factor of 2 are seen between the sites. At the same time, differences in loads of up to 60% are seen. All load components except the edgewise moment, which is mainly gravity seem to change in the same way for the different wind climates. This indicates that the load sensitivity to the most important wind condition parameters is similar for different load signals. The most important wind condition parameters seem to be the mean wind speed distribution and the longitudinal turbulence intensity.

5 Comparison With IEC61400-1

Now that the differences in loads for the different sites have been revealed, it is interesting to see whether the IEC61400-1 classes 5 are representative for the sites chosen by us. The principal wind parameters in the IEC standard are the mean wind speed distribution Table 8 and the turbulence intensity. Four classes exist and an additional class S - special, not considered here. For each of these, two levels of turbulence intensity A: high values, B: low values can be selected. The turbulence intensity depends on the wind speed: I u I 15 15 aU hub / a 1 , U hub (9)

where I 15 0.18 and a 2 for the high turbulence class and I 15 0.16 and a 3 for the low turbulence class. The two IEC levels are compared to the w1-6 intensities in Fig. 3. The design loads for the IEC classes were established using the same method as previously described. For each load case wind speed the loads were calculated with the aeroelastic code and the resulting loads signals were Rainow counted. In the comparison of fatigue equivalent loads at each wind speed, Fig. 4, it is seen that the IEC class B with low turbulence intensities ts very well with the wc2, wc3 and wc4 loads which are the same for these three sites. The IEC A loads are at a level between wc2-4 and wc6 complex terrain. In Table 9, the life time loads for the IEC classes are given. Comparing these values with the life time loads for six different sites Table 7 the IEC turbine classes which apply to the different sites were evaluated. None of the Weibull A parameters for the six sites are exactly equal to the IEC values. It is, however, possible to compare loads for the mean wind distributions that matches best. The wc2 and

Fig. 3 Comparison of turbulence intensities from wind climate 1-6 wc and IEC61400-1 values

Table 9 Normalized life time fatigue loads for the different IEC hler curve classes. Loads are normalized with wc1 loads. A Wo SN exponent of m 10 is used for blade loads and a m 4 for all other loads. No. M f lap M edge F thrust M sht M tilt M yaw M tbL M tbT No. M f lap M edge F thrust M sht M tilt M yaw M tbL M tbT IA 1.56 1.10 1.80 1.60 1.78 1.75 1.90 1.75 IB 1.43 1.08 1.60 1.42 1.60 1.58 1.70 1.59 II A 1.46 1.06 1.65 1.56 1.63 1.61 1.68 1.56 II B 1.32 1.05 1.45 1.38 1.45 1.43 1.49 1.41 III A 1.40 1.03 1.56 1.51 1.53 1.51 1.55 1.43 III B 1.24 1.02 1.36 1.33 1.35 1.33 1.36 1.28 IV A 1.34 0.98 1.41 1.36 1.37 1.34 1.37 1.22 IV B 1.16 0.97 1.23 1.19 1.20 1.18 1.20 1.08

The loads for the six different sites have been compared with loads based on the IEC wind turbine classes. In all casesexcept the offshore class, which according to IEC should be treated separatelythe IEC classes can be chosen in a representative way. Thus, for our six sites, standard IEC class wind turbines can be used without being too conservative. The analysis indicates that wind turbines used in a complex terrain wind farm with a very high annual mean wind speed will be subjected to higher loads than the same turbine subjected to the most severe IEC class IA. For such sites, the classication should be used with care.

Acknowledgments
The present research work has been funded by the European Commission under contract JOR3-CT98-0273 Site-Specic Design Optimization of Wind Turbines based on Numerical Optimization. BONUS Energy A/S has provided details for the aeroelastic modeling of the turbine. The project partners are: Ris National Laboratory, The Netherlands Energy Research Foundation, Bonus Energy A/S, Lagerwey Windturbine B.V., and The University of Sunderland.

References
n, L., Crespo, A., Enevoldsen, P., Go mez-Elvira, R., 1 Frandsen, S. ed., Chaco ndez, J., Hjstrup, J., Manuel, F., and Thomsen, K., 1996, MeasureHerna ments on and Modelling of Offshore Wind Farms, Ris-R-903EN, Ris National Laboratory, Roskilde, Denmark. 2 Sanderhoff, P., 1993, PARK-Users Guide, A PC-program for calculation of wind turbine park performance, Ris-I-668EN, Ris National Laboratory, Roskilde, Denmark. 3 Charnock, H., 1955, Wind stress over a water surface, Quart. J. Roy. Meteor. Soc., 81, 639 640. 4 International Electrotechnical Commission IEC61400-1 Second Edition 199902, 1999, Wind Turbine Generator Systems, Part 1: Safety Requirements, Geneva, Switzerland. 5 Thomsen, K., S.M. Petersen, J.T. Petersen, S. ye, and M. Friedrich, 1996, Terrain Induced Loads on Pitch Regulated Wind Turbines, Ris-R846EN, Ris National Lab., Roskilde, Denmark. 6 Panofsky, H. A., and Dutton, J. A., 1984, Atmospheric Turbulence: Models and Methods for Engineering Applications. Wiley, New York. 7 Fuglsang, P., and Thomsen, K., 1998, Site specic design optimization of wind turbines. In: A Collection of the 1998 ASME Wind Energy Symposium Technical Papers, 36, AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, 12-15 Jan 1998, AIAA, New York, pp. 294 303. 8 Petersen, J. T., 1996, The Aeroelastic Code HawC - Model and Comparisons, in Proc. of 28th IEA Expert Meeting State of the Art of Aeroelastic Codes, Lyngby, Denmark.

Fig. 4 Fatigue equivalent loads for the different wind climates versus wind speed compared with IEC loads, from top: M flap , M tilt , M tbL

Based on the large differences in loads and production, we believe that a very large potential for site-specic design optimization exists. Changing the wind turbine design parameters e.g., blade layout and airfoils will alter the trade-off between loads and production and optimal wind turbine designs can be found for the different sites. As mentioned previously, the results are based on the BONUS 1 MW turbine, which represents the Danish concept type of wind turbines. We believe that the results would be similar for other wind turbines of this concept, but the ndings would not apply directly to other wind turbines.