A numerical model based cross-scale wind farm wind resource assessment method

Abstract

The present application relates to the field of wind power generation technology, and discloses a cross-scale wind farm wind resource evaluation method based on a numerical model, comprising the following steps: step 1, obtaining multi-source observation data of a target area; step 2, screening a WRF optimal configuration scheme suitable for the terrain of the target area; step 3, generating annual-scale wind resource data as initial wind resource data through a WRF dynamic downscaling method; step 4, carrying out partition correction on the initial wind resource data to obtain wind measurement data; step 5, constructing a WRF-WFP model coupling a WRF model and a wind farm parameterization model; step 6, optimizing the lower boundary condition of the WRF; step 7, simulating the atmospheric flow of the target area and quantifying the wake evolution characteristics of the wind farm; step 8, constructing a wake analysis model; and step 9, formulating a coordinated capacity planning of the wind farm group. The present application can accurately predict the power generation capacity of a to-be-developed wind farm, and provide a reliable basis for scientific decision-making of the resource distribution development and planning of a million-kilowatt wind farm.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and more specifically to a method for assessing wind resources in wind farms across scales based on numerical models. Background Technology

[0002] Wind resource assessment is one of the most important references in both the macro and micro site selection stages of wind farms. Only by accurately assessing wind resources can the power generation potential of wind farms in target areas be determined, the investment risks of wind farm construction be quantified, and the economic benefits of wind farms be guaranteed. With the rapid development of the wind power industry, the development of conventional wind-rich areas is approaching saturation. This is driving the wind power industry to expand into areas such as deserts or offshore areas, which are vast but difficult to develop and have low economic utilization rates.

[0003] Wind characteristics are time-varying (diurnal, seasonal, and intergenerational). To assess the wind resource status of candidate wind farms, it is generally necessary to collect long-term historical wind speed observation data to reduce the uncertainty caused by wind speed variations. In practical engineering, the common practice is to use the Measure-Correlate-Predict (MCP) algorithm to assess wind resources when long-term wind farm observation data is lacking. The basic idea of ​​the MCP algorithm is to first set up a wind measurement tower at the candidate wind farm site to measure wind speed, requiring at least one year of observation data. The location of the wind measurement tower must be representative, representing the wind resource status of the entire wind farm. Since one year of observation data from the wind measurement tower only represents the short-term wind resource status of the wind farm, long-term observation data from long-term stations such as reference meteorological stations near the wind farm is needed to obtain the long-term wind resource status. This data is then used to correct the short-term wind measurement data from the wind measurement tower into a set of wind speed data that can reflect the long-term characteristics of the wind farm. However, in the Shagohuang area of ​​Alxa League, due to the rapid development of wind power construction, and the large area and complex terrain of the Shagohuang region, representative observational data is often unavailable. In the traditional MCP method, the first step of obtaining observational data is impossible, making the wind resource assessment process ineffective. Therefore, it is necessary to use numerical simulation methods to obtain wind resource simulation datasets to replace observational data.

[0004] However, the application of current numerical simulation methods faces multiple technical challenges. At the physical mechanism level, firstly, real onshore wind farms are mostly located in complex mountainous or hilly terrains and exhibit large-scale, diverse layouts. Although existing studies have proposed parameterized models for wind farms, their universality under different scales, atmospheric environments, and geographical conditions still needs systematic verification. Secondly, current research mainly focuses on the wake effect of a single wind turbine or independent wind farm, with a serious lack of understanding of the mechanisms of mutual interference between wind farm clusters, which is a key factor in the planning of large-scale wind power bases. As the single-unit capacity, rotor diameter, and installed scale of wind turbines continue to increase, accurately and efficiently assessing the wake characteristics and power output of large-scale wind farms has become a core challenge for industrial development and a fundamental prerequisite for exploring the atmospheric feedback effect of wind farms.

[0005] In summary, there is an urgent need to develop a cross-scale numerical assessment system that integrates high-resolution terrain analysis, wind turbine wake coupling, and atmospheric environmental feedback to address the three major industry needs: feasibility of assessment in areas without data, scientific planning of large-scale bases, and controllability of environmental effects. Summary of the Invention

[0006] The present invention aims to provide a cross-scale wind farm wind resource assessment method based on numerical models, which can accurately predict the power generation capacity of wind farms to be developed, effectively reduce the risks of wind power development, and provide a reliable basis for scientific decision-making in the resource planning of large-scale wind farms.

[0007] The basic solution provided by this invention is: a cross-scale wind farm wind resource assessment method based on numerical models, comprising the following steps:

[0008] Step 1: Acquire multi-source observation data for the target area;

[0009] Step 2: Based on the meteorological conditions and underlying surface characteristics of the target area, select the optimal WRF configuration scheme that is suitable for the terrain of the area;

[0010] Step 3: For the target area, combine digital elevation data and surface roughness data, and use the WRF dynamic downscaling method to generate annual-scale wind resource data with adjustable spatiotemporal resolution as the initial wind resource data.

[0011] Step 4: Perform partition correction on the initial wind resource data to obtain wind measurement data;

[0012] Step 5: Based on wind measurement data, wind farm planning information, and regional topographic and distribution characteristics, construct a WRF-WFP model that couples the WRF model with the wind farm parameterization model.

[0013] Step 6: Optimize the lower boundary conditions of WRF based on the terrain database of the target area;

[0014] Step 7: Using the optimal WRF configuration scheme in Step 2, combined with the WRF-WFP model constructed in Step 5, and the lower boundary condition of WRF in Step 7, the atmospheric flow in the target area is simulated, and the wake evolution characteristics of the wind farm are quantified.

[0015] Step 8: Construct wake analysis models for wind turbine level and wind farm level based on wake evolution characteristics;

[0016] Step 9: Use machine learning automatic optimization method, combined with wake analysis model to optimize wind farm deployment scheme, analyze wake interference effect of nearby wind farms, and formulate coordinated capacity planning for wind farm group.

[0017] The working principle and advantages of this invention are as follows:

[0018] This invention presents a cross-scale wind farm wind resource assessment method based on numerical models. This method can accurately predict the power generation capacity of wind farms under development, effectively reduce the risks of wind power development, and provide a reliable basis for scientific decision-making in large-scale wind farm resource planning. The key points are:

[0019] First, steps 1-3 employ multi-source observational data (satellite, reanalysis, etc.) to drive WRF dynamic downscaling, directly addressing the failure of traditional MCP due to the scarcity of observational data in the desert region. This generates spatiotemporally adjustable annual-scale wind field data, circumventing the engineering bottleneck of long-term on-site anemometer tower deployment. This process completely avoids dependence on on-site anemometer towers, replacing physical observations with numerical simulation, providing fundamental input that meets engineering accuracy requirements for data-deficient areas. Compared to traditional MCP methods, this approach not only covers the assessment blind spots in the desert region but also significantly reduces the time-varying uncertainty of wind speed caused by insufficient representativeness of short-term observations, providing a universal solution for scenarios such as offshore wind power that also face observational difficulties.

[0020] Secondly, addressing the issues of insufficient accuracy in complex terrain and lack of parameterized verification for wind farms, steps 5-7 construct a WRF-WFP coupled model. On one hand, this approach integrates wind farm planning information into the parameterized model, utilizing wind farm planning information (turbine location coordinates, turbine parameters) to calculate the contribution of turbine thrust to the source term of the atmospheric momentum equation in real time, specifically quantifying the turbine-atmosphere interaction. This tight coupling method overcomes the limitation of existing studies that only focus on the wake of independent wind farms, enabling the simulation of dynamic interactions of wind farms of different sizes (from single farms to wind farm clusters) during the evolution of the atmospheric boundary layer. On the other hand, this approach improves the accuracy of flow simulation under complex terrain by optimizing the lower boundary conditions of the terrain database in step 6, overcoming the local flow field distortion caused by terrain smoothing in traditional models; this helps to improve the simulation accuracy and realism of complex terrain.

[0021] Finally, steps 8-9 establish a multi-level wake model (single turbine → wind farm → wind farm cluster) and introduce machine learning to optimize the layout. This enables the systematic quantification of wake interference effects between wind farm clusters within a unified framework, filling a key gap in the background technology for large-scale base planning and providing decision-making suggestions for the large-scale development of desert areas. Attached Figure Description

[0022] Figure 1 This is a first schematic diagram of the method flow of an embodiment of the cross-scale wind farm wind resource assessment method based on numerical model of the present invention;

[0023] Figure 2 This is a second schematic diagram of the method flow of an embodiment of the present invention, which is a cross-scale wind farm wind resource assessment method based on numerical model.

[0024] Figure 3 This is a schematic diagram of the height layer setting in an embodiment of the cross-scale wind farm wind resource assessment method based on numerical models of the present invention. Detailed Implementation

[0025] The following detailed explanation illustrates the specific implementation methods:

[0026] Example 1

[0027] The basic implementation examples are as follows: Figure 1 and Figure 2 As shown: A cross-scale wind farm wind resource assessment method based on numerical models, comprising the following steps:

[0028] Step 1: Obtain multi-source observation data for the target area.

[0029] The multi-source observation data includes data from wind measurement towers, meteorological stations, lidar, and wind power towers; this data is characterized by high wind speed accuracy but limited spatial coverage. After acquiring the multi-source observation data, it is further processed and cleaned according to geographical location, different altitude levels, and available time periods.

[0030] The data processing and cleaning process includes outlier removal, time alignment, and data imputation, and reconstructs missing data for the desert region. The data imputation includes imputation based on wind shear curves for short-term missing values ​​and imputation based on Markov methods for long-term missing values.

[0031] Optionally, multiple reanalysis datasets (such as ERA5, ERA-Interim, CMFD, MERRA-2, JAR-55, and GLDAS) and the LiGA database can be collected simultaneously to further improve data completeness. This part of the data is characterized by wide spatial coverage but relatively low accuracy.

[0032] In this embodiment, the selected target area is the desert region (such as the desert region of the Ordos base); and two reanalysis data are selected as driving data: FNL (i.e., global analysis data released by the U.S. National Center for Environmental Prediction) and ERA5 (i.e., fifth-generation global atmospheric reanalysis data released by the European Centre for Medium-Range Weather Forecasts) to drive the WRF model to perform trial calculations in the desert region. Based on the simulation results and the comparison with nearby meteorological observation stations, the reanalysis data with higher accuracy is selected as the final driving data.

[0033] Step 2: Based on the meteorological conditions and underlying surface characteristics of the target area, select the optimal WRF configuration scheme that is suitable for the terrain of the area.

[0034] The WRF model consists of radiative transfer schemes (longwave and shortwave radiation schemes), cumulus parameterization schemes, microphysics models, land surface models, atmospheric boundary layer parameterization schemes, and surface parameterization schemes. Different parameterization schemes have significantly different simulation effects for different regions and different weather processes.

[0035] When selecting the optimal WRF configuration scheme for the terrain of the suitable area, multiple physical parameterization schemes are first set up. Then, based on the reanalysis data, sensitivity analysis is performed on different physical parameterization schemes. From these, the physical parameterization schemes that meet the preset standards are selected as the optimal configuration scheme for the WRF mode.

[0036] The various physical parameterization schemes include: four combinations of physical parameterization schemes consisting of PBL parameterization schemes -- MYJ, MYNN2.5, QNSE, YSU, and their corresponding SL schemes -- ETA, MYNN, QNSE, MM5; as shown in Table 1.

[0037] Among them, the other physical parameterization settings for the four physical parameterization schemes are as follows: the RRTM scheme is used for long-wave radiation, the Dudhia scheme is used for short-wave radiation, the WSM6 scheme is used for microphysical processes, the Noah scheme is used for the land surface model, and the Kain-Fritsch scheme is used for the cumulus scheme. The cumulus parameterization scheme is not used for areas with a resolution of less than 3 km (because the vertical flux caused by updrafts and downdrafts and the compensation motion outside the cloud can be clearly resolved at a grid size of about 5 to 10 km).

[0038] Table 1 Sensitivity Test Table for Parameterization Scheme

[0039]

[0040] The analytical indicators include mean deviation (MBE), root mean square error (RMSE), relative root mean square error (RRMSE), and consistency (IA); the preset standards include: , .

[0041] In this embodiment, sensitivity analysis of both horizontal and vertical grids is also performed simultaneously:

[0042] Nested domains were set up for the target area—in this embodiment, three nested domains were set up with horizontal resolutions of 9km, 3km, and 1km, corresponding to horizontal grid layers of 180*180, 180*180, and 180*180, respectively. The innermost 1km domain (dx03) covered the area centered on the simulated wind farm. The vertical resolution was defined as 10m (dz10) within 300m of the ground surface, and then vertically stretched. Then, the horizontal resolution (dx) and vertical resolution (dz) were changed to form sensitivity tests for horizontal and vertical grids. For example, the innermost high-resolution nested grid was set to 1km, 3km, and 9km (using a 3x nesting ratio) to test the sensitivity of the mode to the horizontal grid resolution; further, the vertical grid resolution within 300m of the ground surface was increased from 10m to 20m and 30m respectively to test the sensitivity of the mode to the vertical grid resolution. Furthermore, based on the results of the sensitivity analysis, the most suitable grid resolution is selected. For example, when comparing the errors of key meteorological elements (such as wind speed, turbulent kinetic energy, and wind direction frequency distribution) simulated at different resolutions (9km / 3km / 1km), if the error improvement rate is less than 3% when the resolution is increased to 1km, then the next higher resolution (3km) is selected as the optimal solution for the cost-effectiveness of the horizontal grid.

[0043] Comparing the simulation accuracy of boundary layer height and vertical wind speed shear at resolutions of 10m / 20m / 30m, if the R² (i.e., wind speed profile fitting degree) decreases by more than 5% when the resolution is reduced to a certain level (e.g., 20m), and the R² of the next level (10m) is ≥0.95, then the next level (10m) is determined to be the optimal solution for the vertical grid.

[0044] Step 3: For the target area, combine digital elevation data and surface roughness data, and use the WRF dynamic downscaling method to generate annual-scale wind resource data with adjustable spatiotemporal resolution as the initial wind resource data.

[0045] Specifically, in this embodiment, the WRF dynamic downscaling method used includes the ndown one-way nested downscaling method and the weather forecast model spatial downscaling method based on the terrain classification super-resolution model. Through the above methods, mesoscale wind resource data is downscaled to obtain high-precision simulated wind resource data with different spatiotemporal resolutions and different height layers.

[0046] The ndown unidirectional nested downscaling method operates by first running a simulation on a coarse-resolution grid, then using the simulation results as initial fields and boundary conditions to input a simulation on a fine-resolution grid, and finally running the simulation on the fine-resolution grid again. This method is applicable to multi-nested grid downscaling, meets the requirement of the finest downscaling in this embodiment, and has a short execution time and high computational efficiency. Furthermore, to ensure the accuracy of the simulation data, WRF-LES data representing a portion of the time period are used to verify the ndown downscaling results.

[0047] The spatial downscaling method for weather forecast models based on terrain classification super-resolution models includes the following sub-steps:

[0048] S3.1 Under m different terrains, the meteorological element samples under m terrains are obtained by performing n (n>1) nested dynamic downscaling calculations through weather forecast models. Among them, the area represented by the data of the 1st, 2nd, ..., nth layers gradually decreases while the resolution gradually increases.

[0049] S3.2 Extract meteorological element data samples of different resolutions from the nth layer region under the same terrain in S3.1, select two sets of data of different resolutions, and divide them into training set, validation set and test set according to the time dimension;

[0050] S3.3, using different meteorological element data from the training and validation sets in S3.2, the super-resolution model is trained. High-resolution data samples are used as labels for the super-resolution model, and low-resolution data are interpolated to have the same resolution as the labels as inputs to the super-resolution model. The model is then tested on the test set to obtain m super-resolution models for different terrains.

[0051] S3.4 Obtain terrain data for m types of terrain, pass through the m-classification model to obtain a terrain classification model. When using this terrain classification model, select the similarity probability between the predicted terrain and the m types of terrain that pass through the softmax layer as the weight for fusing the multi-terrain prediction results.

[0052] S3.5: The meteorological data to be predicted is predicted separately using the super-resolution models for m different terrain types obtained in S3.3; The terrain data of the area to be predicted is used to output the similarity probability between the area to be predicted and the m different terrain types using the terrain classification model obtained in S3.4; The similarity probability between the m terrain types is used to weight and sum the m prediction results of the super-resolution models to obtain the fusion prediction result of the m super-resolution models.

[0053] The fusion prediction results are the final predicted values ​​of high-resolution meteorological elements (such as wind speed and temperature) for the target area.

[0054] The above methods enable intelligent, efficient replacement and enhancement of dynamic downscaling. First, the super-resolution model itself is trained using high-resolution results generated by nested dynamic downscaling (such as WRF multi-layer computation) as "labels," thus learning and internalizing the physical laws captured by the downscaling process (such as topographic-atmospheric interactions). Second, in the application phase, this fusion model can directly and quickly "super-upgrade" lower-resolution meteorological input data (such as global model output) to the required high resolution (such as 1km) without rerunning the time-consuming complete nested dynamic downscaling calculation, thereby significantly improving efficiency. Simultaneously, terrain-adaptive fusion maintains the physical rationality and regional adaptability of the downscaling; it significantly improves the resolution and accuracy of meteorological forecasts in complex terrain areas (through terrain-specific models), greatly reduces computational costs and time (replacing repeated downscaling), and achieves smoothness in terrain transition zone predictions (through probabilistic fusion weights).

[0055] Notably, during training, the super-resolution model not only learns about resolution enhancement but also implicitly captures and quantifies the physical influence of specific terrain features (such as sand dunes and valleys) on local meteorological elements (such as turbulence and wind shear). These patterns can even be inverted and reused to form a valuable "terrain-climate response" knowledge base, transcending the scope of simple image super-resolution and providing a new approach to understanding regional climate. Furthermore, the probability weights output by the terrain classification model can also serve as effective indicators for quantifying the complexity and mixing degree of regional terrain.

[0056] Step 4: Perform partition correction on the initial wind resource data to obtain wind measurement data.

[0057] The partition correction includes: a first-level correction based on machine learning, comprising the following sub-steps:

[0058] S4.1 Determine the wind speed at the height level that needs to be corrected based on the numerical simulation and statistical downscaling verification results; in this embodiment, the wind speeds are selected as 10m and 30m.

[0059] S4.2, acquire wind measurement data and simulation data corresponding to the height layer to be corrected, and extract a 24-hour time series from the original data with a step size of 1h to generate a dataset;

[0060] S4.3 uses random forest simulation to train a machine learning model and applies it to the site to be corrected to check the correction effect.

[0061] Secondary correction based on vertical extrapolation of wind speed includes the following sub-steps:

[0062] S4.4, using the high-altitude wind speed as the reference point for vertical extrapolation, a wind profile is fitted; in this embodiment, the wind speed at a height of 100m is selected; if the 100m height data is missing, the wind data closest to the 100m height can be selected for extrapolation.

[0063] S4.5, using this wind profile as the correction benchmark, if the relative error of the simulation result is greater than 3%, the correction factor is set to the wind profile value / simulated value; if the relative error of the simulation result is less than 3%, the correction factor is set to 1.

[0064] S4.6 multiplies the correction factor by the time series data at the corresponding height of the corresponding station to correct the initial wind resource data.

[0065] Step 5: Based on wind measurement data and wind farm planning information (such as the layout of wind turbines in the wind farm to be developed, the planned installed capacity, and the power and thrust curves of the wind turbines), combined with regional topographic features and regional distribution features, construct a WRF-WFP model that couples the WRF model with the wind farm parameterized model (WFD).

[0066] The coupled WRF mode and wind farm parameterization model include: for different wind turbine layouts and installed capacities, using the fitted power curve and fitted thrust curve of the wind turbine, parameterizing the wind turbine in the form of a momentum sink into the WRF mode.

[0067] The coupled WRF mode and wind farm parameterization model also include: the coupled aerodynamic characteristics of the ultra-flexible blade wind turbine.

[0068] Furthermore, in this embodiment, the coupling between the WRF mode and the wind farm parameterization model is implemented based on Fortran code;

[0069] During the coupling process, the dragforce subroutine in Fortran code is called to calculate the interaction between the wind turbine and the atmosphere, including the following steps:

[0070] Design a vertical analysis scheme for the wind turbine's operational layers: By iteratively adjusting the vertical coordinate system of the model, ensure that at least 25 vertical layers cover the wind turbine's rotor sweep height range to capture the rotor-atmosphere momentum exchange process. Specifically, when setting the vertical layers, first set a group of near-surface densified height layers and test run them in WRF mode. Then, export the ground-level height of different vertical layers (obtained by subtracting terrain height from altitude), compare the ground-level height with the height swept by the wind turbine rotor, and adjust the set eta_levels accordingly. Then, repeat the above process until there are enough vertical layers within the height swept by the rotor, such as... Figure 3As shown. Among them, eta_levels is a parameter used to define the vertical layering of the pattern; the WRF pattern defines the number of vertical grids by setting the eta_levels parameter (such as 35 layers, 50 layers, etc.), and each layer corresponds to a specific η value; η is a dimensionless vertical coordinate, and its value range is usually from 0 to 1.

[0071] The more vertical layers the wind turbine rotor passes through, the more accurate the analysis of the interaction between the wind turbine and the atmosphere, and the more precise the calculated wind power and wake. In this embodiment, a total of 65 height layers were set, with 30 height layers below 300m to capture the corresponding wind parameter changes. Furthermore, 25 vertical layers were selected within the sweep range of the rotor diameter to achieve a relatively balanced accuracy and computational efficiency.

[0072] Calculate the wind speed at the height of the wind turbine hub. In the specific calculation, first find the two height layers closest to the height of the wind turbine hub, calculate the wind speeds for each layer, and then obtain the wind speed at the hub height through linear interpolation.

[0073] The dragcof subroutine in Fortran code is used to calculate the turbulent kinetic energy coefficient, power coefficient, and thrust coefficient of the wind turbine. First, the program calculates the area swept by the turbine blades by calculating the area. Then, it calculates the thrust coefficient segmented by cut-in and cut-out wind speeds. When the wind speed is less than the cut-in speed or greater than the cut-out speed, a static thrust coefficient value is assigned; when the wind speed is within the operating wind speed range, the thrust coefficient is calculated by interpolation. For the power coefficient, the same segmented method is used: 0 is assigned to the non-operating wind speed range, and interpolation is performed in the operating wind speed range using the input data. Finally, the turbulent kinetic energy coefficient is calculated from the difference between the thrust coefficient and the power coefficient.

[0074] Calculate the power generated by the wind turbine:

[0075] The power generation capacity of the wind turbines within a grid (i,j) is:

[0076] ;

[0077] in, Standard air density ; Power coefficient; Wind speed at wheel hub height; The rotating area of ​​the wind turbine; Number of wind turbines within a grid.

[0078] Next, the impact of the wind turbine on turbulent kinetic energy and horizontal momentum is calculated. Specifically, based on the calculated turbulent kinetic energy coefficient, power coefficient, and thrust coefficient of the wind turbine, the momentum and turbulent action terms of the wind turbine on the atmosphere (i.e., the impact of the wind turbine on turbulent kinetic energy and horizontal momentum) in the WRF grid cell are calculated with reference to the following formula:

[0079] The momentum balance equation is: ;

[0080] If we ignore the change in momentum in the vertical direction, then: ;

[0081] In the formula, u and v represent the horizontal velocity air volume.

[0082] Similarly, the turbulent kinetic energy balance equation is: ;

[0083] ;

[0084] in, Here is the TKE coefficient, where TKE is the turbulent kinetic energy. , For power coefficient, This is the thrust coefficient; that is, the portion that is not converted into electrical energy is converted into turbulent kinetic energy.

[0085] (i,j) represents the grid coordinates of the wind turbine; is the average wind speed of grid (i,j); k is the number of vertical layers; z is the height of the vertical layer; The area of ​​the wind turbine is the area intercepted at the k-th and k+1-th layers in the vertical direction.

[0086] This completes the calculation of the wind turbine parameters and couples them into the WRF mode.

[0087] In calculating the impact of wind turbines on turbulent kinetic energy, the horizontal turbulent kinetic energy transfer mechanism is activated in the planetary boundary layer scheme, enabling the wind farm wake turbulence to spread laterally between grid cells, thereby configuring the horizontal advection function of turbulent kinetic energy.

[0088] The turbulent kinetic energy horizontal advection function is achieved in the following ways:

[0089] Choose the MYNN2 boundary layer scheme that includes TKE prognostic variables;

[0090] The TKE scalarization processing mechanism is enabled to convert the turbulent kinetic energy field into a scalar field capable of horizontal convection and diffusion;

[0091] By linking the surface parameterization scheme with the boundary layer scheme, the conservation of turbulent energy can be ensured.

[0092] By configuring the above settings, the turbulent kinetic energy (TKE) horizontal advection function can be activated during program calculations, thereby significantly improving the simulation accuracy of wind farm wake turbulence. Compared to the traditional MYNN2 boundary layer scheme, which predicts the evolution of TKE in the vertical direction for each node using a one-dimensional TKE equation that only relates to the vertical coordinates, the traditional MYNN2 scheme does not further design for TKE transport schemes (i.e., horizontal convection, vertical convection, and diffusion). Therefore, there is no 1TKE horizontal advection from one horizontal node to another, resulting in an underestimation of turbulence intensity in the wake region and failing to reflect the true characteristics of wind farm wakes.

[0093] This scheme introduces a horizontal transport mechanism (corresponding to the TKE scalarization processing mechanism) to enable turbulence energy to spread laterally between grids, accurately reproducing the wake sector expansion effect and improving simulation accuracy and realism. Secondly, employing the MYNN2 boundary layer scheme (including TKE prognostic variables) and associating it with a surface parameterization scheme ensures strict conservation of turbulent energy in both horizontal and vertical directions, avoiding numerical dissipation and reducing prediction errors for abrupt turbulence changes in complex terrain (such as turbulence enhancement on the leeward slope of sand dunes).

[0094] Step 6: Optimize the lower boundary conditions of WRF based on the terrain database of the target area.

[0095] Specifically, in this embodiment, when optimizing the lower boundary conditions of WRF, publicly available high-precision satellite terrain databases, such as STER, MODIS, and SRTM, are used. The WRF preprocessing tool WPS (i.e., WRF Preprocessing System) is used to filter out satellite data noise and retain the real terrain fluctuations through the geogrid program of WPS. Then, the filtered terrain field is coupled with reanalysis data (such as ERA5) to generate a thermodynamic-dynamic co-current lower boundary, thereby constructing the fine lower boundary terrain conditions of WRF.

[0096] Preferably, the WRF-WFP model in step 5 is used to draw a high-precision wind resource map of the target area under different atmospheric stability, wind speed and wind direction, taking into account the changes in atmospheric flow in the field caused by the operation of wind turbines, so as to provide a direct analysis of the wind field evolution law.

[0097] The wake characteristics include turbulence intensity, wake recovery coefficient, wake influence range, etc. In this embodiment, when drawing wind resource maps, existing drawing software can be used to convert physical quantities into maps.

[0098] Furthermore, when optimizing the lower boundary conditions of the WRF, a subgrid terrain drag force parameterization scheme is simultaneously integrated, including:

[0099] The turbulent deformation drag force induced by the terrain is transformed into an explicit terrain stress term; and the turbulent deformation drag force is coupled as a stress profile to the atmospheric motion equation to correct the momentum deficit caused by the terrain-related flow.

[0100] Specifically, based on the terrain gradient tensor Calculate the turbulent deformation drag force:

[0101] ;

[0102] in, For turbulent deformation drag force; This is the terrain drag coefficient, calibrated according to the surface roughness classification. air density; Horizontal wind speed;

[0103] The turbulent deformation drag term is directly embedded into the momentum equation source term of WRF, explicitly characterizing the obstruction effect of terrain on airflow.

[0104] By using a terrain stress-wind speed feedback mechanism, the systematic bias in near-surface wind resource simulation can be reduced.

[0105] Specifically, the stress transfer function is constructed in the MYNN2 boundary layer scheme:

[0106] ;

[0107] in, The topographic stress turbulence diffusion coefficient is used to vertically diffuse topographic stress to the entire boundary layer (up to 2 km) through nonlinear iteration, thus correcting the momentum loss caused by topographic flow around the surface.

[0108] Furthermore, a terrain stress-wind speed feedback mechanism is introduced, including:

[0109] Introducing a dynamic terrain drag coefficient : .

[0110] in, For feedback gain factor; and These are observed wind speed and simulated wind speed, respectively. In this embodiment, the measured data is assimilated every 6 hours to reduce the systematic bias in the near-surface wind resource simulation in real time.

[0111] Through the above settings, the terrain stress term can be explicitly quantified and vertically coupled to the atmospheric motion equations, directly correcting the momentum deficit caused by terrain-induced flow around the terrain, thus overcoming the limitation of insufficient terrain drag parameterization in traditional WRF models. In particular, this scheme establishes a dynamic feedback mechanism between terrain stress and wind speed, using real-time observation data to dynamically calibrate the drag coefficient. This systematically reduces the near-surface wind speed simulation bias in complex terrain areas (such as steep slopes and canyons), significantly improving the reliability of power generation predictions for desert / mountain wind farms, and providing more reliable wind shear input for the anti-turbulence design of ultra-flexible blades.

[0112] Secondly, the vertical diffusion of topographic stress terms can induce gravity wave breaking effects, significantly enhancing the turbulent mixing intensity at the top of the boundary layer (300-500m height) at night, a phenomenon completely absent in traditional parameterization schemes. Furthermore, the stress-wind speed feedback mechanism helps to capture the periodic oscillation characteristics of topographic wake vortices, providing a new frequency domain analysis dimension for the assessment of turbulent fatigue loads in wind farms.

[0113] Step 7: Using the optimal WRF configuration scheme in Step 2, combined with the WRF-WFP model constructed in Step 5, and the lower boundary conditions of WRF in Step 7, the atmospheric flow in the target area is simulated under different wind turbine layouts and different wind turbine specifications. This allows for the acquisition of wind flow characteristics in different sectors of the wind turbine, such as wind speed attenuation distance and attenuation magnitude, and the quantification of wind farm wake evolution characteristics.

[0114] The wake evolution characteristics of the wind farm include: wake velocity deficit rate, turbulence enhancement index, wake deflection angle, and wake recovery distance.

[0115] Step 8: Construct wake analysis models for wind turbine level and wind farm level based on wake evolution characteristics.

[0116] The wake analysis models include the Jensen wake model for wind turbines and the Gaussian wake model.

[0117] Specifically, in this embodiment, based on the atmospheric flow field simulated in step 7 and the wind farm wake evolution characteristics extracted therefrom, the downstream main wind direction profile of the target wind turbine is extracted, and the wind speed evolution law downstream of the wind turbine and the relationship between wind speed and distance are obtained. Then, the Jensen wake model and Gaussian wake model of the wind turbine for the target area are fitted.

[0118] Step 9: Use machine learning automatic optimization method, combined with wake analysis model to optimize wind farm deployment scheme, analyze wake interference effect of nearby wind farms, and formulate coordinated capacity planning for wind farm group.

[0119] Specifically, in this embodiment, the optimal wind turbine layout scheme is automatically optimized using a wake analysis model and a machine learning method based on a genetic algorithm. The wake influence area of ​​the wind farm is determined, its impact on the power output of adjacent wind farms is analyzed, and an engineering scheme is provided on how to further plan the wind farm's wind energy output to achieve higher overall wind power system efficiency.

[0120] This embodiment provides a cross-scale wind farm wind resource assessment method based on numerical models, which can accurately predict the power generation capacity of wind farms to be developed, effectively reduce the risks of wind power development, and provide a reliable basis for scientific decision-making in large-scale wind farm resource planning.

[0121] Example 2

[0122] A cross-scale wind farm wind resource assessment method based on numerical models has been modified as follows, based on Example 1.

[0123] In step 3, after generating the initial wind resources, the following steps are also performed:

[0124] Step 3-a: Collect historical (e.g., the last 10 years) measured wind speed data (wind measurement tower / SCADA data) for the target area, and calculate the systematic deviation between the simulated and measured values ​​(e.g., wind speed distribution, turbulence intensity).

[0125] In this embodiment, the measured wind speed data specifically includes: wind tower data (such as wind speed, wind direction, turbulence intensity, etc. at different heights); and wind turbine SCADA operation data (such as wind speed at hub height, power curve, etc.).

[0126] Step 3-b: Based on the deviation distribution inversion, retrieve the sensitivity parameters of the WRF physical parameterization scheme (such as the mixing length and surface roughness scaling factor in the boundary layer scheme), and update the parameter set using the ensemble Kalman filter method.

[0127] Step 7 also includes an iterative verification step, which includes:

[0128] The historical wind field is resimulated using the updated WRF-WFP model, and the new simulated values ​​are compared with the measured values ​​a second time. If the error exceeds the threshold, return to step 3-b to re-adjust the parameters until convergence.

[0129] Specifically, for existing wind farms: access the historical database of the wind farm (in this embodiment, actual operating data from the past 3 years is selected), and extract the following key parameters as verification benchmarks: average wind speed (based on the meteorological tower), wind speed profile, turbulence intensity, power output (actual power generation curve of the wind turbine), etc.

[0130] Based on the measured values ​​of key parameters, calculate the deviation between the measured values ​​and the new simulated values. If the deviation is higher than the threshold, return to step 3-b to adjust the parameters again until convergence.

[0131] In this embodiment, the deviation exceeding the threshold includes: wind speed RMSE > 1.5 m / s and power MAE > 6%.

[0132] The convergence conditions are: wind speed RMSE ≤ 0.5 m / s and power MAE ≤ 3%.

[0133] For the wind farm to be planned: call the historical wind measurement database of the target area and extract the following key parameters as verification benchmarks: average wind speed, wind speed profile, turbulence intensity, etc.

[0134] Using key parameters as measured values, calculate the deviation between the measured and simulated values. If the deviation exceeds a threshold, return to step 3-b to re-adjust parameters until convergence. The deviation exceeding the threshold includes: wind speed RMSE > 1.5 m / s, and the convergence condition is: wind speed RMSE ≤ 0.5 m / s.

[0135] This embodiment provides a cross-scale wind farm wind resource assessment method based on numerical models, which can form a closed loop of "initial simulation → measured comparison → parameter correction → secondary simulation". Through this closed-loop iterative simulation correction, the accuracy of resource assessment can be effectively improved.

[0136] Example 3

[0137] A cross-scale wind farm wind resource assessment method based on numerical models has been modified as follows, based on Example 1.

[0138] In step 1, geographic information and multi-source observation data around the target area are collected simultaneously. In this embodiment, information within 100km of the target area is selected for collection.

[0139] The system iterates through the geographic information surrounding the target area. If there are existing wind farms, it further collects basic information about the existing wind farms (including the coordinates of the wind farm, the model of the wind turbine, the hub height, etc.), historical wind measurement databases, and historical operation databases. If there are no existing wind farms, no further processing is performed.

[0140] To reduce computational load, data can be collected specifically within a specified time period for subsequent verification and analysis.

[0141] Step 7 also includes an edge verification step, comprising:

[0142] (a) Extracting simulation data: Extract the time-series wind speed at the coordinates of the wind measurement tower in the target area from the WRF-WFP simulation field output in step 7; extract the simulated wind speed at the hub height at the GPS coordinates of the surrounding wind farms. If there are no existing wind farms in the surrounding area, randomly select several surrounding coordinate points and extract their simulated wind speeds at different heights.

[0143] (b) Calculate fit:

[0144] For the target region, calculate the correlation coefficient R² between the simulated and measured values, and the absolute error |Δk| of the Weibull shape parameter k.

[0145] For the surrounding wind farms, the relative error RE between the simulated and measured values ​​is calculated and statistically analyzed by wind direction sector.

[0146] For surrounding areas where no existing wind farms exist, the correlation coefficient R² between simulated and measured values ​​is calculated and statistically analyzed by coordinate point.

[0147] (c) Perform threshold determination:

[0148] If R² < 0.99 or |Δk| > 0.1 for the target area, terrain parameter correction is triggered.

[0149] If RE > 5% in any sector of the surrounding wind farm, wake parameter correction will be triggered.

[0150] If R² < 0.95 in the surrounding area, wind field initialization correction is triggered.

[0151] Specifically, in the terrain parameter correction, return to step 6 to correct the terrain parameters of the target area. In this embodiment, the corrected terrain parameter is the terrain dynamic roughness length (z0). The larger the z0 value, the rougher the surface (e.g., mountains, forests, urban building complexes), the stronger the wind resistance, and the more obvious the near-surface wind speed attenuation; conversely, the smaller the z0 value (e.g., flat grasslands, sea surfaces), the smaller the influence of the surface on the wind, and the more uniform the wind speed profile.

[0152] During the correction, the target area is divided into 1km×1km grids, and the measured-simulated wind speed deviation of each grid is calculated. A regression model of this deviation and terrain parameters (slope, curvature) is established. Based on the regression coefficients, z0 is corrected in reverse, and the lower boundary conditions of the terrain database are updated.

[0153] In the wake parameter correction, the wake attenuation coefficient in the WRF-WFP model is corrected. (Corresponding to the Jensen model) and turbulence intensity correction factor (Corresponding Fitch source item).

[0154] For wind sectors where RE exceeds the limit, extract the measured wake loss value downstream of the prevailing wind direction of that sector. And calculate its value compared with the simulated wake loss value. The relative error, then according to the error proportion Adjust the parameters. That is: ; , This represents the turbulence intensity deviation value.

[0155] Then, the parameters of the WRF-WFP model are updated.

[0156] In wind field initialization, the correction target is the initial boundary field of WRF (such as ERA5 reanalysis data). Specifically, the deviation between simulated and measured values ​​in the wind direction frequency distribution is calculated, forming a multidimensional wind direction frequency difference vector. This vector is then converted into a trend term in the Nudging algorithm to relax and adjust the initial wind field (e.g., converting it into adjustment requirements for wind field components; for example, if the measured frequency of a certain wind direction sector is higher than the simulated value, the wind field intensity in that direction needs to be increased).

[0157] This embodiment provides a cross-scale wind farm wind resource assessment method based on numerical models. Compared with Embodiment 1, it additionally introduces collaborative simulation analysis of the surrounding areas of the target area. This enables the simulation data obtained by the model of this scheme to not only match the wind farm operation data of the target area by more than 99%, but also match the data of the surrounding wind farm by more than 95%, which has extremely high simulation accuracy and can meet the needs of engineering implementation.

[0158] Example 4

[0159] A cross-scale wind farm wind resource assessment method based on numerical models has been modified as follows, based on Example 1.

[0160] If there are existing wind farms in the target area, the following adjustment steps shall be introduced.

[0161] Before step 5, insert the following steps: construct a "wind farm-free" underlying surface scenario, run a control experiment based on the same WRF configuration, and generate background wind resource data that is unaffected by wind power development.

[0162] In step 7, simulations are performed under two scenario conditions, including:

[0163] Scenario A: Load the actual wind farm layout (including existing / new sites) and run the WRF-WFP model.

[0164] Load wind farm parameters, including: wind turbine geometry parameters—collect detailed parameters of existing and new wind turbines, including hub height, blade radius, tower height and diameter, etc., ensuring parameter accuracy. Wind turbine aerodynamic parameters—obtain thrust coefficient curves, power curves, and other data for each wind turbine, which can be obtained from technical manuals provided by the wind turbine manufacturer or measured data. For new wind turbines, if measured data is lacking, typical parameters of the same model can be referenced. Layout parameters—determine the precise coordinates (latitude and longitude or projected coordinates) and arrangement angles of existing and new wind turbines. Then run the WRF-WFP model.

[0165] Scenario B: Load only the existing wind farm layout and run the WRF-WFP model.

[0166] Data related to existing wind farms, including turbine parameters and layout information, were extracted from the wind farm parameters of Scenario A, excluding newly built wind farms. The same WRF-WFP model configuration and operation settings as Scenario A were used to ensure consistent simulation conditions for both scenarios, with the only difference being the wind farm layout.

[0167] The wind speed difference between Scenario A and the control experiment (i.e., baseline wind resource data) is calculated to quantify the cumulative impact of new wind farms on the target area. This difference quantifies the additional wind speed attenuation caused by new wind farms on top of existing farms, helping to assess the incremental impact of new planned projects on regional wind resources. .

[0168] Calculate the difference between Scenario B and the control experiment (i.e., background wind resource data) to quantify the existing impact of built wind farms: .

[0169] And the cross-effects between the two: Cross-influence can be used to capture the synergistic effects of wind farm clusters. When When this occurs, it indicates that the wakes of existing and newly built wind farms overlap, leading to a greater reduction in wind speed; when At that time, there may be a partial cancellation effect of wake interference.

[0170] The wind speed is for scenarios including both existing and newly built wind farms (Scenario A). The wind speed is for the scenario that only includes existing wind farms (Scenario B). The wind speed is the baseline wind resource data (i.e., the wind speed in the case of no wind farm).

[0171] Alternatively, it can also be based on and ,draw Distribution map, adding icons to the locations of new wind farm sites and marking key areas (such as grid points with attenuation exceeding 5%); and drawing... Contour maps, with annotations added to areas of dense contour lines (indicating significant cross-effects), can help analyze the causes of synergistic effects (such as wake superposition direction and terrain acceleration / deceleration effects) in conjunction with wind farm layout.

[0172] The above analysis confirms the incremental impact of newly built wind farms, enabling precise quantification of the complex impacts on wind farm clusters and helping to optimize the layout scheme of newly built wind farms.

[0173] In step 9, during the automatic optimization process using machine learning, a threshold constraint for wind resource changes is also set, such as a wind speed attenuation rate of ≤5%, to ensure that the newly planned wind farm cluster does not significantly deteriorate the operating environment of the existing wind farm.

[0174] In this embodiment, a machine learning method based on genetic algorithm is used to automatically find the optimal solution, with the total power generation of the wind farm group as the optimization target, and a wind speed attenuation rate penalty term is set.

[0175] This embodiment provides a cross-scale wind farm wind resource assessment method based on numerical models, suitable for assessing wind farm clusters. This solution specifically decouples the impact of wind farms into the independent impact of existing farms, the superimposed impact of newly built farms, and the cross-impact impact, enabling precise quantification of each component's effect and overcoming the shortcomings of traditional assessment methods that fail to adequately consider the synergistic effects of wind farm clusters.

[0176] Example 5

[0177] A cross-scale wind farm wind resource assessment method based on numerical models has been modified as follows, based on Example 1.

[0178] The coupled WRF mode and wind farm parameterization model also include: the coupled aerodynamic characteristics of the ultra-flexible blade wind turbine.

[0179] In this embodiment, an ultra-flexible blade solver is constructed.

[0180] Specifically, the ultra-flexible blade solver incorporates aerodynamic-structural coupling equations:

[0181] ;

[0182] Where EI is the bending stiffness and y is the lateral displacement of the blade. Relative wind speed, It is the flexibility index (blade tip displacement / impeller diameter). For real-time aerodynamic twist angle; This refers to the dynamic lift coefficient. It is aerodynamic.

[0183] x refers to the coordinate along the blade span (the length direction from the leaf root to the leaf tip, similar to a one-dimensional coordinate "from the petiole to the leaf tip"), used to describe the mechanical response of the blade at different span positions (e.g., x=0 at the leaf root, x=R at the leaf tip, where R is half the blade length). m is the mass per unit span of the blade.

[0184] This is the initial angle of attack of the blade; The effective angle of attack of the blade; air density; and These are the first time derivative of the blade's lateral displacement (i.e., lateral vibration velocity) and the second time derivative of the blade's lateral displacement (i.e., lateral acceleration), respectively.

[0185] In practical applications, wind speed at grid points is transmitted. The solution is then applied to the ultra-flexible blade solver. Based on the wind speed at the grid points, and combined with the aerodynamic-structural coupling equations, the aerodynamic forces are solved. . This represents the wind speed value at the horizontal position (x, y), the height level (z), and time t.

[0186] Will The WRF momentum equation is injected using a weighted average of the impeller swept area to update the atmospheric flow field.

[0187] Dynamic roughness correction is performed, and the equivalent roughness is calculated. This is to update the underlying surface of the WRF, thereby completing the coupling of the aerodynamic characteristics of the ultra-flexible blade wind turbine.

[0188] in, ;

[0189] For equivalent roughness, For terrain dynamic roughness length; denoted as the average displacement of the impeller plane, k is an empirical coefficient (default value is 0.15), and D is the diameter of the fan impeller.

[0190] This embodiment provides a cross-scale wind farm wind resource assessment method based on numerical models, which, compared with Embodiment 1, can more realistically reflect the wind farm flow field and achieve more accurate wind resource assessment.

[0191] Specifically, ultra-flexible blades undergo significant elastic deformation under airflow, resulting in a constantly changing aerodynamic shape. Coupled with this characteristic into the model, it allows for a more accurate simulation of airflow within a wind farm. For example, traditional rigid blade models assume a fixed blade shape, while ultra-flexible blades may bend significantly in strong winds, altering the direction and velocity distribution of the incoming flow. By considering this characteristic, the model can more accurately predict wind speed and direction distribution within the wind farm, including the range and intensity of the wake effect, thus making the simulation results closer to the actual operating conditions of a wind farm.

[0192] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A numerical model based cross-scale wind farm wind resource assessment method, characterized in that, The method comprises the following steps: Step 1, obtaining multi-source observation data of a target area; Step 2, screening a WRF optimal configuration scheme suitable for the terrain of the target area based on the meteorological conditions and the features of the underlying surface of the target area; Step 3, generating annual-scale wind resource data with adjustable spatial and temporal resolutions for the target area by a WRF dynamic downscaling method in combination with digital elevation data and surface roughness data, as initial wind resource data; Step 4, performing zonal correction on the initial wind resource data to obtain wind measurement data; Step 5, based on the wind measurement data, constructing a WRF-WFP model coupled with a WRF model and a wind farm parameterization model according to wind farm planning information in combination with regional terrain features and regional distribution features; The coupled WRF model and wind farm parameterization model comprise: for different wind turbine layouts and installed capacities, wind turbines are parameterized into the WRF model in the form of momentum sinks by using wind turbine fitting power curves and fitting thrust curves; The coupled WRF model and wind farm parameterization model further comprise: coupling the aerodynamic characteristics of super-flexible blade wind wheels; Step 6, optimizing the lower boundary condition of WRF based on the terrain database of the target area; Step 7, simulating the atmospheric flow of the target area and quantifying the wake evolution characteristics of the wind farm by using the WRF optimal configuration scheme in step 2, the WRF-WFP model constructed in step 5, and the WRF lower boundary condition in step 7; Step 8, constructing a wind turbine level and wind farm level wake analysis model based on the wake evolution characteristics; Step 9, optimizing the wind farm machine layout scheme by using a machine learning automatic optimization method in combination with the wake analysis model, analyzing the wake interference effect of the adjacent wind farm, and formulating a coordinated capacity planning of the wind farm cluster.

2. The method of claim 1, wherein, In step 1, the multi-source observation data comprises wind measurement tower data, meteorological station data, laser radar data, and wind power tower data; after obtaining the multi-source observation data, the multi-source observation data is sorted and cleaned according to geographical location, different height layers, and data available time periods; In the sorting and cleaning process, abnormal value elimination, time alignment, and data interpolation are included, and missing data in the Shaguo barren area is reconstructed; the data interpolation comprises interpolation based on wind shear curves for short-time missing values and interpolation based on a Markov method for long-time missing values.

3. The method of claim 1, wherein, In step 2, when screening the WRF optimal configuration scheme suitable for the terrain of the target area, a plurality of physical parameterization schemes are set, and then sensitivity analysis is performed on different physical parameterization schemes to screen out a physical parameterization scheme with an analysis index meeting a preset standard as the WRF model optimal configuration scheme; The plurality of physical parameterization schemes comprise four groups of physical parameterization schemes composed of PBL parameterization schemes: MYJ, MYNN2.5, QNSE, YSU, and corresponding SL schemes: ETA, MYNN, QNSE, MM5; The analysis indexes include mean bias error (MBE), root mean square error (RMSE), relative root mean square error (RRMSE) and agreement index (IA); and the preset standard includes: , .

4. The method of claim 1, wherein, In step 4, the zonal correction comprises a first-level correction based on machine learning, including the following substeps: S4.1, determining the height layer wind speed that needs to be corrected according to the numerical simulation and statistical downscaling verification results; S4.2, Obtain the wind data and simulation data corresponding to the height layer to be corrected, and extract a 24-hour time series from the original data with a step size of 1h to generate a data set; S4.3, Train the machine learning model using random forest simulation, and apply it to the site to be corrected to check the correction effect; The second-level correction based on wind speed vertical extrapolation includes the following sub-steps: S4.4, Take the high-layer wind speed as the reference point for vertical extrapolation, and fit the wind profile; S4.5, Take the wind profile as the reference for correction. If the relative error of the simulation result is greater than 3%, set the correction coefficient as the wind profile value / simulation value; if the relative error of the simulation result is less than 3%, set the correction coefficient as 1; S4.6, Multiply the time series data of the corresponding height of the corresponding station by the correction coefficient to realize the correction of the initial wind resource data.

5. The method of claim 4, wherein, The WRF dynamic downscaling method includes: The weather forecast model spatial downscaling method based on terrain classification super-resolution model includes the following sub-steps: S3.1, Under m different terrains, perform n-layer nested dynamic downscaling calculation through a weather forecast model to obtain meteorological element samples under m terrains, wherein the regions represented by the 1st, 2nd, …, n layers of data gradually decrease, and the resolution gradually increases; S3.2, Extract different meteorological element data samples of different resolutions of the n-layer region under the same terrain in S3.1, select two groups of data with different resolutions, and divide them into a training set, a validation set, and a test set according to the time dimension; S3.3, Use the training set and the validation set in S3.2 to train the super-resolution model with different meteorological element data, wherein the high-resolution data sample is used as the label of the super-resolution model, the low-resolution data is interpolated into data with the same resolution as the label as the input of the super-resolution model, and the test set is used for verification, and finally the super-resolution model under m different terrains is obtained; S3.4, Obtain terrain data under m terrains, and obtain a terrain classification model through m classification models. When using the terrain classification model, select the similarity probability between the output predicted terrain and the m terrains through the softmax layer as the weight for fusing the prediction results of multiple terrains; S3.5, Use the m super-resolution models under different terrains obtained in S3.3 to predict the meteorological data; use the terrain classification model obtained in S3.4 to output the similarity probability between the predicted region and the m different terrains; and use the similarity probability between the m terrains to weight and sum the m prediction results of the super-resolution model to obtain the fused prediction result of the m super-resolution models.

6. The method of claim 1, wherein, In step 5, the coupling of the WRF model and the wind farm parameterization model is realized based on Fortran code; In the coupling process, the dragforce subroutine of the Fortran code is called to realize the calculation of the interaction between the wind turbine and the atmosphere, including the following steps: Design a vertical analytical scheme for the wind turbine action layer: adjust the vertical coordinate system of the model through iteration to ensure that at least 25 vertical layers cover the wind turbine swept height range to capture the blade-turbulence momentum exchange process; The wind speed of the hub height of the wind turbine is calculated; The dragcof subroutine is called to calculate the turbulent kinetic energy coefficient, power coefficient and thrust coefficient of the wind turbine; The power generated by the wind turbine is calculated; the influence of the wind turbine on the turbulent kinetic energy and horizontal momentum is calculated, and thus the calculation of the parameters of the wind turbine is completed, and is coupled into the WRF model; In the calculation of the influence of the wind turbine on the turbulent kinetic energy, the horizontal transport mechanism of the turbulent kinetic energy is activated in the planetary boundary layer scheme, so that the wake turbulence of the wind farm can be diffused horizontally between grid cells, and the horizontal advection function of the turbulent kinetic energy is configured.

7. The method of claim 6, wherein, The horizontal advection function of the turbulent kinetic energy is realized by the following way: Select the MYNN2 boundary layer scheme containing the TKE prognostic variable; Enable the TKE scalarization processing mechanism to convert the turbulent kinetic energy field into a scalar field that can be horizontally convected and diffused; Correlate the surface parameterization scheme with the boundary layer scheme to ensure the conservation of turbulent kinetic energy.

8. The method of claim 1, wherein, In step 6, when optimizing the lower boundary condition of WRF, the subgrid terrain drag parameterization scheme is integrated synchronously, including: The terrain-induced turbulent deformation drag force is parameterized as an explicit terrain stress term, and the turbulent deformation drag force is coupled to the atmospheric motion equation as a stress profile to correct the momentum loss caused by terrain flow; Through the terrain stress-wind speed feedback mechanism, the systematic deviation of the simulation of near-surface wind resources is reduced.

Images