Applicability of WorldCover in Wind Power Engineering: Application Research of Coupled Wake Model Based on Practical Project

Abstract

This paper discusses how the incorporation of high-resolution ground coverage dataset ESA WorldCover into a wind flow field and wake simulation calculation, as well as the use of the coupled wake model for wind farm output simulation, can improve the accuracy of wind resource assessment using engineering examples. In the actual case of grid-connected wind farms in central China, SCADA wind speed data is reconstructed to the free flow wind speed in front of the wind turbine impeller using the transfer function of the nacelle, and the wind farm is modeled using OpenWind software, simulating the wind speed at the height of each wind turbine hub and each wind turbine output. The results show that when other initial data are consistent, using ESA’s high-precision land cover dataset WorldCover 10 m to make roughness lengths which improves the wind farm output simulation accuracy by 8.91%, showing that it is worth trying to apply WorldCover 10 m to the wind farm simulation design. At the same time, this case is used to compare and analyze the application of the Eddy-Viscosity wake model and the two coupled wake models based on the Eddy-Viscosity wake model. The results show that the coupled wake model will have higher accuracy than the Deep Array Eddy Viscosity wake model and it is 1.24% more accurate than the Eddy Viscosity wake model, and the ASM Eddy Viscosity wake model is 5.21% more accurate than the Eddy Viscosity wake model.

1. Introduction

As people become more concerned about environmental protection and green energy, the wind power industry continues to thrive. Wind energy consumption increased by 88% between 2002 and 2017 in the world’s 17 major wind energy-consuming countries, including China. During this time, China’s renewable energy consumption increased by 39%, India by 23%, Ireland by 21%, and the United Kingdom by 20% [1]. China commissioned 17 GW offshore wind power capacity in 2021, accounting for 80% of the global offshore growth, making China’s total offshore wind capacity 27.7 GW [2], rapidly surpassing the UK and Germany to take the lead globally in terms of offshore wind energy capacity with a 44.8% market share [3]. China promised in 2020 that it would achieve carbon neutrality by 2060. Wind energy’s carbon dioxide emission intensity in China is more than 98% lower than that of traditional fossil fuels, according to studies, and the emission reduction effect can reach 84–98% [4]. The world’s three major economies, the European Union, China, and the United States, have all implemented effective incentives to promote the development of the wind power industry and increase the proportion of green power electricity [5].

The evaluation of Chinese wind resources has gradually progressed to fine assessment, and a more accurate and reliable assessment of resources will help to develop wind resources more reasonably, save costs, and increase output. Some Chinese researchers have emphasized that, despite becoming the world’s largest wind power installer in the last two decades, the quality of installed capacity has been subpar [6]. As a result, it is both meaningful and necessary to improve the accuracy and reliability of resource assessment. Engineers must use advanced technology and resources to obtain high-resolution and high-precision wind resource maps in wind farms to support the construction of wind power projects.

The surface cover’s most direct influence on near-surface airflow is to change the near-surface flow field, thereby changing the near-surface wind speed profile. Many scholars have already conducted extensive research in this area. Neff et al. (1998) discovered that densely vegetated ridges and hills can cause nonlinear interactions [7]. Jacobs et al. (1995) measured the vertical and horizontal changes in wind speed in the maize canopy and discovered a 20–30% difference between the average wind speed and the horizontal average [8]. This research is extremely practical. Corn is a popular economic crop in northern China. According to the findings by Fu et al. (2020), the influence of changes in soil geology on surface wind is limited, changes in the surface cover have a significant impact on simulated wind speed, and roughness length is the most important factor influencing wind speed [9]. Wen et al. (2014) demonstrate that improving the accuracy of ground cover can improve the accuracy of wind speed simulation [10]. According to the findings of Frank Baier et al. (2022), the higher the resolution of the ground cover data used, the smaller the error of wind speed simulation, especially near the surface [11]. Matthew J.C. et al. (2012) discussed the effect of surface roughness on wind turbine wake turbulence at a given atmospheric stability in their study [12]. D.L. Elliott and J.C. Barnard (1990) investigated the variability of wind speed and turbulence intensity as a function of wind direction and surface roughness across the site [13].

According to the findings, the surface cover has a significant impact on wind resources and wind turbine wake simulation, and changes in the surface cover and data resolution are also important factors. To improve the accuracy and reliability of the simulation, high-resolution surface cover data should be utilized in wind resource assessment.

With the rapid development of wind power projects in recent years, researchers have not stopped researching and developing wake models, shifting from a single wake model to a coupled wake model [14]. Another question that wind resource engineers have is whether the coupled wake model is more exact and can help improve wind resource assessment accuracy.

At the moment, the most commonly used land cover products in wind resource assessment are NASA GeoCoverLC90m resolution dataset (2009) [15], European Space Agency (ESA) Copernicus Global Land Service (CGLS) land cover 100 m data set (2015–2019) [16], and China National basic Geographic Information Center global surface cover data product GlobeLand30 (2020) [17]. This paper will use the ESA’s 10 m resolution global surface cover data released in 2020 to make roughness length as the initial data for wind farm software simulation design, and use different coupled wake models to calculate the wind power output. The experimental project selects a grid-connected wind farm in central China, and the experimental results are compared to the actual data. The study of this paper is expected to provide a reference for accurate wind resource assessment.

2. Background

2.1. Data Source

The surface cover data used in this paper are from the European Space Agency’s WorldCover project. The main outcome of this project is the release in 2020 of a free global land cover (GLC) product with a resolution of 10 m based on Sentinel-1 and Sentinel-2 data. It includes 11 different types of land cover and has been independently verified with a global accuracy of 74.4%. In Asia, the overall accuracy is 80.7%. The accuracy of the categories of tree cover, snow and ice, farmland, water, and bare/sparse vegetation is higher, exceeding 80%. The accuracy of grassland and architecture is moderate, while shrubs, wetlands, and moss/lichens are less accurate. Moss, lichens, and grasslands are overestimated, while trees are underestimated [18].

The 10 m resolution surface cover data products, which are also based on Sentinel satellite data, include Google’s DynamicWorld product coverage [19]. Water, trees, cultivated land, and other coverage types are divided into nine categories, with an overall accuracy of 71.3% Landcover coverage [20]. With an overall accuracy of 85%, the coverage type includes nine categories, such as water, grassland, ice, and snow. Although LandCover appears to have higher overall accuracy, studies have shown that WorldCover has higher accuracy and LandCover accuracy is lower at lower cell resolution (100 m²) [21,22,23].

The National Basic Geographic Information Center of China’s land cover data GlobeLand30 (2020) is widely used in the current wind resource assessment work. GlobeLand30 products cover an area of 80° S–80° N with a resolution of 30 m. Ground cover is classified into ten types, including cultivated land, forest, water, and artificial cover. Among them, the minimum classification accuracy of six types, including cultivated land and forest, is controlled at or above 70%, while the lowest classification accuracy of four types, including water bodies and artificial cover, is controlled at or above 80%. In China, the precision of the data set is 80.4% [17,24,25].

2.2. The Context of the Experimental Project

The experimental project Yingchen wind farm is located in Qingling Mountain in central China, in the north temperate zone, and forest coverage in this area can reach 37.1%. The natural woodland is dominated by deciduous broad-leaved forests such as Robinia pseudoacacia, willow, Toona sinensis, paulownia, and so on, while the economic woodland is dominated by walnut, persimmon, peach, and other fruit forests, and the crops are primarily wheat, corn, sweet potato, and so on. The wind farm has a capacity of 48 MW and employs 146–3000 wind turbines from a Chinese manufacturer. The wheels are 90 m tall and were connected to the grid in 2020.

At the project site, obtain WorldCover (2020), LandCover (2020), and GlobeLand30 (2020) ground coverage data. Figure 1 and Figure 2 depict the surface coverage of wind power sites in the area and its surrounding areas.

As shown in Figure 2, WorldCover, LandCover, and GlobeLand30 are basically the same in terms of overall surface cover classification, but LandCover data lacks woodland cover data; WorldCover and GlobeLand have a better approximation in the distribution of all types of surface cover, while WorldCover shows more ground details, which is confirmed by the GoogleEarth satellite map.

2.3. Making the Roughness Length

To use the ESA WorldCover surface cover data set in wind resource assessment, the mapping relationship between surface cover data and surface roughness must be established. The mapping relationship between ground coverage and roughness length is established using the European Wind Atlas [26] and the New European Wind Atlas [27], and the roughness length is adjusted based on the local actual situation (

Table 1. Roughness length.

Code	Land Use Land Cover Class	EWA Roughness Classification	EWA Roughness Length (m)	Modified Roughness Length (m)
10	Tree cover	2.5	0.65	0.7
20	Shrubland	1.5	0.15	0.2
30	Grassland	1	0.03	0.05
40	Cropland	1	0.05	0.03
50	Built-up	3	1	0.55
60	Bare	1	0.01	0.01
70	Snow and ice	0	0.001	0.001
80	Water	0	0.0001	0.0001
90	Wetland	1	0.03	0.02
95	Mangroves	2	0.15	0.15
100	Moss and lichen	1	0.01	0.01

) [28,29,30,31,32]. Expand the 10 km range from the site location to the east, west, north, and south to account for the impact of upstream surface cover on the experiment [33]. When the wind blows over the top of the mountain, a pressure difference exists between the windward and leeward sides of the mountain, increasing friction loss and effective roughness length [34]. Because there is no dense vegetation near the wind measurement site, displacement height is not required. The EWA roughness length value and the improved roughness length value were used to assign the land cover data to generate the roughness length map that the program could read (Figure 3).

3. Verification Mode

Accurate energy calculations are critical in wind power engineering. Improving the precision of output estimation can help to improve the precision of investment return expectations. Wind resource engineers must select a suitable wake model based on the characteristics of different wind farm terrain environments when calculating energy, and when the wake model performs simulation calculations of energy, ground roughness is a very dependent parameter. As a result, in the same simulation environment, a high-precision land cover map can improve the wake model’s calculation accuracy.

The Yingchen wind farm, which has been connected to the grid, is chosen as the actual measurement project in this paper. According to this project’s feasibility study report (FSR), the Openwind [35] wind farm design software is also used for modeling, the WindMap [36] wind flow model is used to calculate the wind flow field, and the Eddy-Viscosity wake model is used to calculate energy. The initial data uses the same wind measurement data, surveying and measured topographic map, and the same uncertainty table. On this basis, the roughness length is calculated using EAS land cover data, and the energy is calculated using the Deep Array EV wake model and the ASM-EV wake model. It should be noted that the wake model’s general parameters use the same settings, while the non-general parameters use the default parameters without modification. The experiment’s energy data is compared to the wind farm feasibility study report data and the wind farm’s actual operation data. On one hand, it confirms that WorldCover data can be used in the wind power industry. On the other hand, it compares the errors of different wake models. The flowchart process will be used for verification in this paper (Figure 4).

3.1. Wake Model

The Eddy-Viscosity wake model (EV) [37] is a linear CFD calculation that calculates the wake of an axisymmetric structure wind turbine using the time-averaged Navier–Stokes equations (RANS). The model employs cylindrical coordinates and assumes incompressible airflow.

The Eddy-Viscosity wake model automatically observes mass and momentum conservation in the wake. The average Eddy-Viscosity of each downstream wake segment is used to calculate the wind-turbulence interaction. The wake in the Eddy-Viscosity model has a Gaussian distribution, and its recovery depends on the turbulence intensity. The more turbulence there is, the more the wake wind mixes with the free wind around it, and the wake travels and recovers faster [33]. Its equation for calculation is as follows:

$$1-\frac{v}{v_0}=D_{M }exp\left\{-3.56{\left(\frac{r}{b}\right)}^2\right\}$$

where $v_0$ represents the average free stream wind speed, v represents the wind speed at a distance r from the wake centerline, $D_{M }$ represents the initial wind speed attenuation at the wake centerline, and b represents the wake width parameter, and the equation is as follows:

$$b=\sqrt{\frac{3.56C_T}{8D_M\left(1-0.5D_M\right)}}$$

where $D_M$ is related to thrust coefficient $C_T$ and turbulence intensity, the expression is as follows:

$$D_M=C_T-0.05-\left(16C_T-0.5\right)\frac{I_A}{1000}$$

where $I_A$ is the ambient turbulence intensity (%).

The Eddy-Viscosity wake model was used to calculate the energy in the feasibility study. Figure 5 depicts the Eddy-Viscosity wake model simulating the wake behind a single wind turbine, which employs a Gaussian distribution to improve the prediction of the wake velocity deficit, and is run by Openwind. The X-axis in the figure is a multiple of the wind turbine impeller’s diameter(D), and the background free wind speed is 10 m/s, which is displayed in red. The legend assumes that the background wind speed is 1(100%), and different colors correspond to different multiples of that speed.

The Deep Array Eddy-Viscosity wake model (DAWM-EV) [38] is a modification and improvement of the Eddy-Viscosity wake mode, a coupling model based on Sten Frandsen’s theory. Deep Array and Eddy-Viscosity are separately modeled in the DAWM model and combined by taking the maximum value of roughness influence and basic wake influence (background roughness is obtained by reading the imported rough map).

An infinite array of wind turbines is represented as a uniform region of high surface roughness in Frandsen’s theory. The roughness drags on the atmosphere, causing changes in the structure of the planetary boundary layer (PBL) downstream, most notably a decrease in free-stream wind speed at turbine hub height. According to this theory, the equivalent roughness of the wind farm is:

$$z_{00}=h_{H}exp\left(-\frac{k}{\sqrt{c_t+{\left(k/\ln \left(h_H/z_0\right)\right)}^2}}\right)$$

where $h_H$ is the hub height, k is the von Karman constant (about 0.4), $z_0$ is the ambient roughness between turbines, and $c_t$ is the distributed thrust coefficient,

$$c_t=\frac{\pi }{8S_d{S}_c}{C}_T$$

where $C_T$ is the thrust coefficient, $S_d$ is the mean downwind spacings in rotor diameters and $S_c$ is the mean crosswind spacings in rotor diameters.

Assume that each wind turbine occupies a distinct area that contributes to increased surface roughness. Increased roughness creates the internal boundary layer as the wind reaches the wind turbine. The wind width line or shear in this internal boundary layer is defined by the wind turbine roughness rather than the ambient roughness, and the velocity at the top of the internal boundary layer must immediately match the velocity above it. After the wind has passed through the wind turbine, a second internal boundary layer will form to indicate the transition back to the environment’s surface. Both internal boundary layers grow with downstream distance, according to the following equation.

$$h_{ibl}\left[\ln \left(\frac{h_{ibl}}{z_0}\right)-1\right]=\left(\frac{x}{z_0}-1\right)z_0$$

In this equation, $h_{ibl}$ represents the height of the internal boundary layer, x represents the distance generated by the internal boundary layer, and $z_0$ represents the downstream roughness (wind turbine roughness for the first internal boundary layer and ambient roughness for the second internal boundary layer). The downstream wind turbine produces its own internal boundary layer under the upstream wind turbine.

Once the equivalent roughness is defined, the meteorological theory is used to estimate the impact on the hub-height wind speed deep within the array (i.e., where the PBL has reached equilibrium with the array roughness) under the assumption of a constant geostrophic wind speed G and a neutral logarithmic profile throughout the PBL. Taking only the first turbine’s IBL pair and assuming that both IBLs have grown to exceed hub height, the equation for hub-height speed is as follows:

$$\frac{v_H^{\prime }}{v_H}=\frac{\ln \left(h_1/z_0\right)\ln \left(h_2/z_{00}\right)}{\ln \left(h_2/z_0\right)\ln \left(h_1/z_{00}\right)}$$

$v_H^{\prime }$ is the wind speed of the wind turbine hub height downstream of the wind farm, $v_H$ is the wind speed of the upstream wind turbine hub height, h₁ and h₂ are the heights of the first and second internal boundary layers, respectively, and $z_0$ and $z_{00}$ are the roughness of the wind turbine and the environment (background).

Figure 6 depicts the wake behind a single wind turbine as simulated by the Deep Array EV wake model, which predicts wake velocity deficits using a Gaussian distribution in the near-wake region and a linear extension in the far-wake region, and it is run by Openwind. The X-axis in the figure is a multiple of the wind turbine impeller’s diameter(D), and the background-free wind speed is 10 m/s, which is displayed in red. The legend assumes that the background wind speed is 1(100%), and different colors correspond to different multiples of that speed.

The Area Slowdown Eddy-Viscosity wake model (ASM-EV) [39] considers the wind farm as a whole. The wind farm is related to the inversion layer at the top of the atmospheric boundary layer as additional surface roughness and as a gravity wave generator, which means that even though the wind speed of the atmospheric boundary layer above the wind field is uniform, it is very sensitive to the decrease of the lower wind speed and the decrease of energy, so there will be changes in the pressure gradient and the generation of gravity waves. The model calculates the interaction force between the atmosphere and wind farm based on the conservation of kinetic energy and the above-influencing factors, which is a top-down simulation method of fluid kinetic energy. The basic equation is:

$$\frac{v\left(t\right)}{v_0}=1+\left(\frac{v}{v_0}-1\right)exp\left(-\alpha t\right)$$

where $v_0$ is the free wind speed of the upstream wind turbine at hub height, v is the wind speed of the downstream wind turbine at hub height.

$$\alpha t=k{u}_{*}z/\Delta z^2$$

where z is PBL’s height, h is wind turbine hub height, $\Delta z$= z − h, $u_{*}$ is the friction velocity.

The ASM EV wake model’s single wake wind speed is depicted in the figure below (Figure 7), which ran by Openwind. The X-axis in the figure is a multiple of the wind turbine impeller’s diameter(D), and the background free wind speed is 10 m/s, which is displayed in red. The legend assumes that the background wind speed is 1 (100%), and different colors correspond to different multiples of that speed.

3.2. Nacelle Transfer Function

Wind farm operational data are derived from the SCADA (Supervisory control and data acquisition) system. The wind speed data in SCADA is the nacelle anemometer measurement data. The anemometer is easily affected by the wind rotor and the nacelle because it is installed behind the wind rotor on the top of the nacelle. As a result, the nacelle transfer function must be used to correct it to the free stream wind speed at hub height in front of the impeller. The nacelle transfer function establishes the relationship between the nacelle wind speed and the wind speed measured in the wind measuring tower [40].

In this paper, the wind speed of the anemometer tower and the wind speed of the nacelle are divided into 0.5 m/s intervals in this study [41], and the wind speed in front of the nacelle impeller is calculated using the following equation [42]:

$$v_H^{\prime }=\frac{v_{H,i+1}-v_{H,i}}{V_{N,i+1}-v_{N,i}}\times \left(v_N-v_{N,i}\right)+v_{H,i}$$

where $v_H^{\prime }$ is the wind speed at the hub height of the wind turbine, $v_{H,i}$, and $v_{H,i+1}$ is the bin-average wind speeds in bin i and bin i + 1, $v_{N,i},$ and $v_{N,i+1}$ is the wind speed measured by the nacelle anemometer in the bin i and bin i + 1.

3.3. The Sector Affected by the Wake

According to the calculation formula of the influence sector given in the IEC 61400-12-1: 2017 standard, determine the influence sector range of the first row of the wind turbine on the rear row of wind turbines [43].

The calculation equation is as follows:

$$\theta =1.3arctan\left(\frac{2.5D_n}{l_n}+0.15\right)+10$$

where $D_n$ is the diameter of the wind turbine impeller, $l_n$ is the distance between wind turbines.

4. Results and Analysis

The table below shows the results of the estimated energy in the Yingchen wind farm feasibility study report, the estimated energy in this experiment, and the actual energy of the wind farm (

Table 2. Comparison of actual energy and calculated energy.

	FSR	EV	DAWM-EV	ASM-EV	Actual
Mean Speed [m/s]	5.496	5.265	5.265	5.265	5.291
Gross Energy [GWh]	127.644	125.611	125.611	125.611	123.644
Waked Energy [GWh]	114.107	105.322	104.097	97.119	96.411
Wake Loss [%]	10.63	16.13	17.10	22.66	22.03
Net Energy [GWh]	91.445	84.405	83.423	77.831	79.057

). Except for the roughness length data, which is made using WorldCover 10 m, all initial input data in this verification use data from the feasibility study stage, and the values of the uncertainty parameters are all set by the feasibility study stage. The Eddy-Viscosity wake model uses the same settings as the feasibility study stage. According to the results in Table 2, the deviation between the calculated and actual power generated by the Eddy-Viscosity wake model in this verification is 6.76%, which is 8.91% higher than the feasibility study EV model. As a result, the high-accuracy land cover dataset, WorldCover 10 m can be considered to improve measurement accuracy.

In this verification, the energy is calculated using the Eddy-Viscosity wake model and two basic coupled models of the model. According to the results, the deviation between the Deep Array Eddy-Viscosity wake model and actual energy is 5.52%, which is 10.15% greater than the feasibility study report (FSR) and 1.24% greater than the Eddy-Viscosity wake model. The difference between the ASM Eddy-Viscosity wake model and actual energy is 1.55%, which is 14.12% greater than the feasibility study report and 5.21% greater than the Eddy-Viscosity wake model.

To further clarify the differences between the three wake models used in this experiment, a ridge in the wind farm area is chosen and the wake attenuation is analyzed one by one. Although the arrangement is more regular and the number of wind turbines is greater on the southwest side, the influence of the topography is greater, which is not conducive to the analysis of the influence of the wake, and the elevation drop of the ridge on the northeast side of the wind farm is limited. The wind farm ridge wind turbines on the northeast side are used in this experiment to study wake attenuation.

The wind turbines on the northeast ridge are 3# to 8#. Because the 3# wind turbines are not on the same side of the mountain as the other wind turbines, and their position is relatively offset to the east relative to other wind turbines on the same ridge, they are not included in this analysis. Figure 8 depicts the ranking positions of 4#–8# wind turbines. The wake of the upstream wind turbine will affect the downstream wind turbine, and the influence range of the upstream wind turbine on downstream wind turbines can be calculated using the wake influence sector equation, and the north is 0°.

Table 3. The influence sector of upstream wind turbine on downstream wind turbine.

	Down	5#	6#	7#	8#
Up
4#		103.3°~173.3°	115.5°~168.1°	123.5°~167.9°	130.3°~169.3°
5#			101.7°~177.7°	120.3°~176.5°	129.3°~174.5°
6#				118.1°~196.9°	131.7°~186.7°
7#					123.6°~196.6°

shows the sector range of the downstream wind turbine affected by the wake of the upstream fan when 4# is the first upstream wind turbine, using the formula given above with reference to the IEC 61400-12-1.

This study reviews the sectors where the downstream wind turbines are impacted by the upstream 4# wind turbine. On 3 August, 2020, 0:45 to 2:45 lasted 2 h, and the wind direction is 135° to 160°, which is covered by the range in Table 3, and it is relatively heavily affected by the wake.

Table 4. Wind speed comparison table.

	4#	5#	6#	7#	8#
Actual Waked Speed [m/s]	8.403	5.587	4.815	5.294	6.967
EV Waked Speed [m/s]	8.379	5.274	5.134	5.177	7.092
DAWM-EV Waked Speed [m/s]	8.379	5.274	4.519	4.338	7.034
ASM-EV Waked Speed [m/s]	8.379	5.494	4.868	4.822	6.741

displays the average wind speed calculated by different wake models for each wind turbine in this experiment during the above time period and wind direction. According to the trend, the free wind loses a lot of energy after passing through the 4# wind turbine, and the wind speed is greatly reduced when it reaches the 5# wind turbine. The wind blows through the 5# wind turbine, losing some of its energy. The wind is supplemented with energy from the surrounding environment as it passes through the 6# wind turbine and loses some of its energy to reach 7#. At the same time, because 7# is not completely parallel to the first three wind turbines, the wake flow received by the 7# wind turbine is reduced. In addition, the impact is minor. The 8# wind turbine has a greater offset to the east, and the distance from the upstream 7# wind turbine is 367.5 m, while the distance from the upstream 4# wind turbine has increased to 1267.5 m, indicating that the wind has recovered enough energy from the environment.

The wind blows from the 4# wind turbine to the subsequent 5#–8# wind turbine. Combining with Figure 9, it is also discovered that the altitude of the 8# wind turbine has a large drop from the wind turbine arranged in front of it, and the altitude difference from the 7# is 51.38 m, the altitude difference with the 4# wind turbine is 91.96 m, and the wind will eventually reach the 8# wind turbine through continuous climbing. If the free flow wind passes through the 4# position and reaches the 8# position, the wind speed should be continuously increased by the terrain.

As a result of the environment, terrain, and wind turbine arrangement, wind speeds of 4#, 5#, and 6# wind turbines decreased row by row, while wind speeds of 7# and 8# wind turbines recovered noticeably. Table 4 displays the results.

The turbulence at each wind turbine shown in

Table 5. Simulated turbulence table.

	4#	5#	6#	7#	8#
Ambient TI [%]	6.66	6.52	6.51	6.27	5.77
EV Total TI [%]	6.66	21.16	22.28	21.48	15.68
DAWM-EV Total TI [%]	6.66	20.15	25.86	27.03	16.66
ASM-EV Total TI [%]	15.80	34.89	43.34	48.38	35.17

, when combined with the turbulence tables generated by the wake model, confirms the energy loss and recovery at the wind turbine hub.

The power generation calculated by the ASM Eddy-Viscosity wake model is the closest to the actual energy, as shown by the summary of the calculation results in

Table 6. Array yield comparison.

	4#	5#	6#	7#	8#
Actual Array Yield [kWh]	4320.583	1566.864	1011.900	1240.540	2786.203
EV Array Yield [kWh]	4619.868	1372.439	1214.295	1482.740	3143.623
DAWM-EV Array Yield [kWh]	4619.868	1372.439	806.209	902.434	3071.128
ASM-EV Array Yield [kWh]	4070.235	1534.638	1040.528	1155.667	2803.804

. In the third row, the difference between the Deep Array wake model and the Eddy-Viscosity wake model becomes apparent. The Eddy-Viscosity wake model’s simulated wake flow has begun to recover greater than the loss at the fourth wind turbine 7#, and power generation has begun to rise. 8# The wind turbines’ comprehensive geographical location and simulation calculation of the wake flow show that the power generation has a higher recovery than the previous three wind turbines 5#, 6#, and 7#. The Deep Array wake model simulation also shows energy recovery for the 7# wind turbine. Greater than the loss, but recovery takes longer. The ASM Eddy-Viscosity wake model simulation data trend is consistent with the previous two models. Because of its top-down simulation method, the energy lost in the wind farm can only be supplemented by the turbulent kinetic energy above, reducing wind resistance and increasing turbulence. reflected more clearly. The energy of the 5# and 6# wind turbines is decreasing, the energy of the 7# wind turbine is being restored, the energy of the 8# wind turbine is being less affected by the wake, and the energy supplemented from the environment is becoming more sufficient. Figure 10 depicts the average wind speed and wind turbine output power during the analysis period.

5. Conclusions

The following conclusions can be drawn from the preceding analysis:

When other initial data are consistent, using ESA’s high-precision land cover dataset WorldCover 10 m to make roughness lengths improve simulation accuracy by 8.91%, indicating that it is worthwhile to try to apply WorldCover 10 m to wind farm simulation design. When conducting wind resource assessments, it is recommended to use a higher-resolution ground cover dataset and to adjust the roughness length appropriately based on the actual local vegetation coverage.

The coupling wake model outperforms the basic wake model in terms of accuracy. The Deep Array Eddy-Viscosity wake model is 1.24% more accurate than the Eddy-Viscosity wake model, and the ASM Eddy-Viscosity wake model is 5.21% more accurate than the Eddy-Viscosity wake model. Encourage the use of coupled wakes in wind resource assessment work and help improve wake models.

Although the Deep Array Eddy-Viscosity wake model performed less well than the ASM Eddy-Viscosity wake model in this verification, one reason is that it is a wake model designed for large wind farms, whereas the wind farm used in this verification was on a smaller scale. Follow-up studies are needed to determine which of the two performs better in large-scale wind farms.

We use the nacelle in transfer function in this paper to reconstruct the wind speed of the wind turbine, and we also advocate the construction of production wind measuring towers in wind farms to improve the accuracy of wind speed reconstruction, and thus the accuracy of wind power prediction.

For verification, this study employs an actual grid-connected wind farm. Although a relatively ideal experimental result was obtained, more experimental items and data must be used to form a statistically significant result that can be used to fine-tune wind power projects. Wind resource evaluation provides new ideas to help improve wind resource evaluation accuracy, allowing for more accurate budgets, more valuable investment suggestions, and more convincing returns.