A method for generating downscaling data using mesoscale data and spatial terrain data

Abstract

The application relates to the field of meteorological data processing, and discloses a method for generating downscaling data by using mesoscale data and spatial terrain data, which comprises the following steps: generating a global scale transformation coefficient alpha according to the data resolution a of the mesoscale data of a specified area and the resolution b of the downscaling data specified by a user, if alpha is greater than or equal to a preset threshold n, then interpolating the mesoscale data by using high-resolution meteorological observation data and the spatial dynamic weight of a grid to obtain intermediate data with a resolution of m, and a < m < b; calculating a terrain complexity index corresponding to each grid of the intermediate data according to a pre-determined terrain evaluation feature group, generating an adaptive mesoscale transformation coefficient beta corresponding to each grid according to the terrain complexity index, the observation station density and the mesoscale data quality; and generating downscaling data corresponding to each grid according to the beta and the intermediate data, determining the installation position of a wind turbine and installing the wind turbine. Therefore, the precision of the downscaling data is improved, and the fine selection of the wind turbine site is met.

Description

Technical Field

[0001] This invention relates to the field of meteorological data processing, and specifically to a method for generating downscaling data using mesoscale data and spatial topographic data. Background Technology

[0002] Weather Research and Forecasting Model (WRF) is widely used as a mesoscale data-based weather forecasting system. However, the computational cost of WRF is increasing exponentially, making it impossible to achieve near real-time data updates and forecasts, and thus failing to meet the needs of the renewable energy wind power sector. In wind energy resource assessment, mesoscale data has low spatial resolution, making it difficult to capture the microscale impacts of complex terrain (such as mountains and coastlines) on wind fields. Early studies in renewable energy wind power typically require resolutions of 1-3 km or even finer (such as 200 m or 100 m) to more accurately assess wind energy resources. In refined applications such as wind turbine site selection and wind resource assessment, low-resolution meteorological data cannot accurately reflect local wind field characteristics, leading to site selection errors and affecting wind energy utilization efficiency. Therefore, effectively improving the spatial resolution of mesoscale data and generating high-precision downscaled meteorological data has become an important research direction.

[0003] Existing downscaling processes often encounter the following technical problems:

[0004] First, mesoscale data has low spatial resolution, making it difficult to capture the microscale impact of complex terrain on wind fields and accurately reflect local wind field characteristics, leading to site selection errors and affecting wind energy utilization efficiency.

[0005] Second, for complex terrain, data mutations are inevitable during the downscaling process, and the computing resources for different regions lack specificity, resulting in poor accuracy of downscaled data and poor accuracy of wind turbine site selection. Summary of the Invention

[0006] The summary section of this invention provides a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] This invention proposes a method for generating downscaling data using mesoscale data and spatial topographic data to solve one or more of the technical problems mentioned in the background section above.

[0008] This invention provides a method for generating downscaling data using mesoscale data and spatial topography data, comprising:

[0009] Based on the data resolution 'a' of the mesoscale data in the specified area and the resolution 'b' of the downscaling data specified by the user, a global scale transformation coefficient 'α' is generated. If 'α' ≥ a preset threshold 'n', then the high-resolution meteorological observation data, combined with the spatial dynamic weights of the grid, is used to interpolate the mesoscale data to obtain intermediate data with a resolution of 'm', where 'a' < 'm' < 'b'.

[0010] Based on a pre-determined set of terrain assessment features, the terrain complexity index corresponding to each grid of the intermediate data is calculated. Based on the terrain complexity index, the density of observation stations, and the quality of mesoscale data, the adaptive mesoscale transformation coefficient β corresponding to each grid is generated.

[0011] Based on β, the downscaled data for each grid is generated using intermediate data.

[0012] Based on the downscaled data, the installation location of the wind turbine was determined and the wind turbine was installed.

[0013] Optional, the first The spatial dynamic weight of each grid The calculation formula is as follows:

[0014]

[0015] in, It is the first The first grid The weights of each factor It is the global weight coefficient of the factor, and the factor is one of the following: elevation, slope, vegetation condition, distance between grid points and observation points. The global weight coefficient of each factor is dynamically adjusted.

[0016] Optionally, based on β, using intermediate data, generate downscaling data for each grid, including:

[0017] Based on the intermediate scale transformation coefficient β, calculate the downscaled grid and the downscaled resolution c corresponding to each grid. If c < b, then correct the downscaled resolution of the grid to b. If c > b, then use the bilinear interpolation algorithm to interpolate the intermediate data to obtain the first-level downscaled grid and the first-level downscaled data corresponding to each grid.

[0018] For each grid in the first-level downscaling grid, second-level downscaling data is generated using the first-level downscaling data and WRF. This includes setting the parent domain resolution to c, generating the initial field and boundary conditions of the parent domain using mesoscale data, running the subdomain simulation, and generating second-level downscaling data with a subdomain resolution of b.

[0019] Optionally, based on β, using intermediate data, the downscaling data corresponding to each grid is generated, which also includes:

[0020] The boundary transition zone of the second-level downscaling data is fused to obtain downscaling data, including: defining the grids with a T-value change > 0.2 / adjacent grids as the abrupt grid boundary region, and using a Laplace smoothing filter to smooth the abrupt grid boundary region.

[0021] Optionally, the method of generating downscaling data using mesoscale data and spatial topography data of the present invention further includes:

[0022] Computing resources are reallocated according to a preset time period. During the reallocation process, double-precision floating-point operations are assigned to grids with T≥0.7, and the total processor utilization of each grid with T≥0.7 is adjusted to the preset utilization rate.

[0023] Optionally, the method of generating downscaling data using mesoscale data and spatial topography data of the present invention further includes:

[0024] Through an online learning mechanism, the difference between the station observations within the grid and the generated downscaled data is calculated periodically. When the difference is greater than twice the standard deviation, a global weight coefficient adjustment strategy is triggered.

[0025] Optionally, in calculating the terrain complexity index for each grid of the intermediate data based on a pre-determined set of terrain assessment features, the method further includes:

[0026] Based on b, a geospatial data cloud with a matching resolution is selected from geospatial data clouds of different resolutions in the specified area as the target geospatial data cloud, which is used to calculate the terrain complexity index corresponding to each grid.

[0027] Optionally, the method of generating downscaling data using mesoscale data and spatial topography data of the present invention further includes:

[0028] After generating downscaled data, multiple test points are selected in the corresponding downscaled grid, and measurements are taken at the test points using a wind tower to obtain actual meteorological data for multiple horizontal planes.

[0029] Calculate the correlation coefficient between the test points and the actual meteorological data.

[0030] The present invention has the following beneficial effects:

[0031] 1. Improved accuracy of downscaling data, meeting the refined requirements of wind turbine site selection. Specifically, the global scale transformation coefficient α is calculated to determine whether high-resolution meteorological observation data should be introduced, and the interpolation process is optimized based on the spatial dynamic weight of the grid; adaptive intermediate scale transformation coefficient β is generated using the terrain complexity index, observation station density, and mesoscale data quality, and adaptive downscaling adjustment is performed; based on β and intermediate data, downscaling data corresponding to each grid is generated, and the wind turbine installation location is determined and installed. Through staged downscaling and differentiated grid division, this grid-level differentiated calculation system can reduce the error of downscaling data for complex terrain to below 1°C (improving accuracy by about 30% compared to traditional methods) while maintaining overall computational efficiency, thereby improving the accuracy of downscaling data and further meeting the refined requirements of wind turbine site selection.

[0032] 2. Improve the downscaling accuracy of mesoscale data to make it more suitable for complex terrain regions, and optimize computational resource allocation and online learning capabilities to enhance the accuracy of weather forecasting and wind energy assessment. Improve the accuracy of downscaling data through grid-based dynamic weight calculation, interpolation methods, and WRF simulation to better reflect the actual conditions of complex terrain regions. Employ a terrain complexity index β and dynamically adjust weight coefficients to ensure that downscaling data more accurately reflects meteorological changes under different terrain conditions. Use a Laplace smoothing filter to handle abrupt grid boundaries, resolving data abrupt changes caused by interpolation or simulation, and improving the continuity and stability of downscaling data. Implement a computational resource reallocation strategy, allocating double-precision floating-point computing resources to areas with high computational demands (T≥0.7) to improve computational efficiency and reduce redundant calculations. Employ an online learning mechanism to periodically calculate the error between downscaled data and measured data; when the error exceeds a threshold, adjust global weights to improve the real-time adaptability and accuracy of downscaling data. Based on the user-specified downscaling resolution b, select the most matching data from geospatial data clouds of different resolutions to improve the accuracy of terrain complexity index calculation. The reliability and applicability of the downscaling method are evaluated by acquiring measured meteorological data through a wind tower and calculating its correlation coefficient with the downscaled data. Attached Figure Description

[0033] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0034] Figure 1 This is a flowchart of a method for generating downscaling data using mesoscale data and spatial topographic data according to the present invention;

[0035] Figure 2This is a grid partitioning map corresponding to the resolution of intermediate data in a method for generating downscaled data using mesoscale data and spatial terrain data according to the present invention.

[0036] Figure 3 This invention relates to a method for generating downscaled data using mesoscale data and spatial terrain data, which is a grid-based map of the target data spatial resolution. Detailed Implementation

[0037] The invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the drawings and embodiments of the invention are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0038] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0039] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0040] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0041] The names of messages or information exchanged between the various devices of this invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0042] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] like Figure 1 The diagram illustrates a flowchart of a method for generating downscaling data using mesoscale data and spatial topography data according to the present invention, specifically including the following steps:

[0044] Step 101: Based on the data resolution a of the mesoscale data in the specified area and the resolution b of the downscaling data specified by the user, generate the global scale transformation coefficient α. If α ≥ the preset threshold n, then use the high-resolution meteorological observation data and combine it with the spatial dynamic weight of the grid to interpolate the mesoscale data to obtain intermediate data with a resolution of m, where a < m < b.

[0045] In some embodiments, the execution entity of the method for generating downscaling data using mesoscale data and spatial topographic data according to the present invention can be a backend server. Mesoscale data provides city-level and regional-level predictions and reconstructions of meteorological variables such as temperature, humidity, wind speed, and precipitation; however, mesoscale data is still insufficient to meet the refined requirements such as wind turbine site selection. Based on this, the designated area refers to the target area where mesoscale data needs to be downscaled. The designated area can be a wind turbine installation planning area. Data resolution 'a' refers to the original spatial resolution of the mesoscale data, i.e., the size of a data grid cell. Spatial resolution is the smallest detail or information that can be captured within a unit area. Specifically, spatial resolution is usually used to describe the grid cell size of a dataset, representing the area range that the data can reflect. For example, the data resolution 'a' of mesoscale data is (approximately 5km × 5km) latitude and longitude grids, where 'a' represents the spatial resolution of the original meteorological data. When the user wants to obtain higher resolution data (such as 1km × 1km), downscaling is required. The resolution 'b' of the user-specified downscaling data refers to the user's desired target data spatial resolution, i.e., the size of the data grid after downscaling. For example, if b = 1km, it means the user wants to upgrade the original resolution of the mesoscale data to a resolution of 1km × 1km. The goal of the downscaling process is to transform the mesoscale data from resolution a to resolution b. Based on this, the global scale transformation coefficient α is obtained by dividing a by b. The global scale transformation coefficient α is a key indicator for determining whether staged downscaling is needed. α represents the ratio between the grid size a of the original data and the resolution b of the user-specified downscaling data. α is used to determine whether interpolation downscaling is necessary. In practice, directly processing large-scale data may be computationally too expensive, while α helps to select suitable interpolation strategies, improving the efficiency and accuracy of downscaling calculations. High-resolution meteorological observation data refers to meteorological data with higher spatial accuracy than mesoscale data. High-resolution meteorological observation data is data obtained from meteorological stations or wind towers, and the data content can include measured ground wind speed, wind direction, etc. The spatial dynamic weight of the grid refers to assigning different weights to grids at different locations during the downscaling interpolation process. Based on this, if α ≥ the preset threshold n, a finer grid system than α is constructed within the specified area at an intermediate resolution m. These grids are new data grids that need to be interpolated from the mesoscale data and filled with them. For each new grid, the influence of multiple geographical factors on the interpolation is considered, and the spatial dynamic weights of the grids are calculated. A weighted interpolation method (such as weighted inverse distance) is used to fuse high-resolution meteorological observation data with mesoscale data; the core idea of ​​interpolation is that the closer the distance and the more similar the terrain, the higher the weight. Finally, intermediate data with a resolution of m is obtained, where a < m < b. Here, n is the preset threshold, representing the preset scale transformation factor.As an example, if a is 5km and b is 1km, then α is 5, and n can be 3 for example. When α ≥ the preset threshold n, and the specified area is a complex terrain, the mesoscale data is first corrected using high-resolution meteorological observation data, and then spatial dynamic weights are applied. After adjusting the error magnitude, interpolation is performed on the mesoscale data to obtain intermediate data with a resolution of m, where m can be 3 km, for example... Figure 2 As shown in the figure, after interpolation, a 5km × 5km grid is divided into four 3km × 3km grids, designated A, B, C, and D. When α ≥ the preset threshold n, it indicates that the original mesoscale data has a large discrepancy in accuracy compared to the user's requirements, and the simple interpolation error is too large, necessitating the introduction of more real observation data and terrain correction information. The high-resolution meteorological observation data for the specified area is obtained from ground-based meteorological stations. Interpolation refers to estimating the value of unknown data points among known data points. Spatial dynamic weighting refers to dynamically adjusting the weight of each data point in the interpolation calculation based on the spatial relationships between data points.

[0046] Step 102: Calculate the terrain complexity index for each grid of the intermediate data based on the pre-determined terrain assessment feature group. Generate the adaptive intermediate scale transformation coefficient β for each grid based on the terrain complexity index, observation station density, and mesoscale data quality.

[0047] In some embodiments, after generating intermediate data with a resolution of m, a three-dimensional evaluation matrix containing eight terrain features is constructed by calling the ALOSWorld 3D digital elevation model with a precision of 30m. In addition to the basic elevation standard deviation (σ≥200m triggers terrain correction) and aspect consistency index (Circular Variance <0.4 is considered uniform aspect), derived parameters such as terrain location index (TPI values ​​outside the ±50m range are marked as anomalous terrain) and curvature variability (Curvature CV>30% is judged as complex micro-terrain) are added. The terrain complexity index T is generated by extracting the principal component PC1 (variance contribution rate 82%) through principal component analysis. The specific steps include: firstly, feature standardization preprocessing is performed, and the eight terrain feature parameters (such as slope variability, elevation standard deviation, etc.) are Z-score standardized.

[0048]

[0049] in, For the first Grid Network Original values ​​of the parameters, and The mean and standard deviation of each parameter are given. Then, the covariance matrix is ​​constructed using the following formula to calculate the 8x8 feature covariance matrix:

[0050]

[0051] in, Let covariance matrix be the variance matrix. For the centralized data matrix, for The transpose of the matrix, The sample size (number of observations) is specifically the number of rows in the matrix. The eigenvalue covariance matrix is ​​used to reveal the intrinsic correlations between topographic parameters (e.g., aspect consistency and elevation standard deviation are often negatively correlated). Principal component extraction is then performed, using the Jacobi iteration method to solve for the eigenvalues ​​and corresponding eigenvectors of the covariance matrix, sorted in descending order of eigenvalues. As an example, the largest eigenvalue for PC1 is 6.56 (total variance is 8), and the variance contribution rate is calculated: the contribution rate is the largest eigenvalue divided by the total variance, yielding 82%. Next, principal component scores are calculated, projecting the standardized data onto the PC1 direction.

[0052]

[0053] in, For the first Principal component scores for each sample. The PC1 eigenvector components (e.g., slope variability weight 0.41, elevation standard deviation weight 0.38). For the first The first sample Several standardized variables were then used. Finally, nonlinear normalization was performed, and a modified Sigmoid function was used to map the PC1 score to the [0,1] interval.

[0054]

[0055] in, For the first The terrain complexity index for each sample, parameters Control the distribution pattern (usually taken as 2.5 to 3.5), The median score for PC1.

[0056] In some embodiments, the adaptive particle swarm optimization algorithm is activated when T ≥ 0.7, and the value of β is dynamically adjusted within the grid: flat areas (T < 0.3), observation station density ≥ 2 stations / 10km 2 The mesoscale data quality is rated as excellent, with β=0.2. The terrain is of medium quality (0.3≤T<0.7), and the observation station density is 1 station / 10km. 2For mesoscale data with a medium quality rating, β=ln(T+1) is used for transition. For complex terrain (T≥0.7) and observation station density of 0, the mesoscale data quality rating is poor, so β=1-T. Thus, by adjusting the value of β, targeted grid partitioning is achieved under different terrain conditions, observation station density conditions, and mesoscale data quality conditions. This is essentially the second stage of downscaling, meaning that differential scaling is achieved during the second stage of downscaling. For cases where no specific value for β is specified, an appropriate value can be selected based on actual needs.

[0057] Step 103: Based on β, use the intermediate data to generate downscaling data for each grid.

[0058] In some embodiments, the downscaled grid and the downscale resolution c corresponding to each grid are calculated based on the intermediate scale transformation coefficient β. The downscale resolution is calculated using the formula c = β × m. Downscaled data refers to the specific data values ​​(such as temperature, wind speed, etc.) calculated after downscaling. The downscaled data provides specific values ​​within each grid cell, representing the numerical values ​​of a specific physical quantity on the refined grid, reflecting the meteorological or environmental characteristics of the region after downscaling. If c < b, the downscale resolution of the corresponding grid is corrected to b; if c > b, the intermediate data is interpolated using a bilinear interpolation algorithm to obtain the first-level downscaled grid and first-level downscaled data for each grid. For each grid in the first-level downscaled grid, second-level downscaled data is generated using the first-level downscaled data and WRF. The boundary transition zones of the second-level downscaled data are then fused to obtain the final downscaled data.

[0059] Step 104: Determine the installation location of the fan based on the downscaling data and install the fan.

[0060] In some embodiments, a comprehensive score corresponding to the downscaled data of each grid is calculated and sorted to obtain a downscaled data sequence. The grid with the highest comprehensive score in the downscaled data sequence is selected as the wind turbine installation location. In practice, geological information can also be queried from a pre-stored geological information database corresponding to the grid to determine whether the area corresponding to the grid meets the installation requirements. If not, the grid with the second-highest comprehensive score is selected as the wind turbine installation location. After determining the installation location, installation instructions are sent to the workers at the work site so that they can install the wind turbine according to the instructions.

[0061] These embodiments improve the accuracy of downscaling data, meeting the refined requirements of wind turbine site selection. Specifically, a global scale transformation coefficient α is calculated to determine whether high-resolution meteorological observation data should be introduced, and the interpolation process is optimized based on the spatial dynamic weight of the grid. Adaptive intermediate scale transformation coefficients β are generated using the terrain complexity index, observation station density, and mesoscale data quality, and adaptive downscaling adjustments are performed. Based on β and the intermediate data, downscaling data corresponding to each grid is generated, and the wind turbine installation location is determined and installed. Through phased downscaling and differentiated grid partitioning, this grid-level differentiated computing system can reduce the error of downscaling data for complex terrain to below 1°C (improving accuracy by approximately 30% compared to traditional methods) while maintaining overall computational efficiency, thereby improving the accuracy of downscaling data and further meeting the refined requirements of wind turbine site selection.

[0062] In some embodiments, to further address the second technical problem described in the background section, namely, "For complex terrain, data mutations are inevitable during downscaling, and computational resources for different regions lack specificity, resulting in poor accuracy of downscaled data and poor accuracy of wind turbine site selection," in some embodiments of the present invention, the first... Spatial dynamic weights of each grid The calculation formula is as follows:

[0063]

[0064] in, It is the first The first grid The weights of each factor It is the global weight coefficient of the factor, and the factor is one of the following: elevation, slope, vegetation condition, distance between grid points and observation points. The global weight coefficient of each factor is dynamically adjusted.

[0065] In some embodiments, wherein This represents the total number of influencing factors considered.

[0066] In practice, we need to calculate the spatial dynamic weights of the 5th grid, considering three factors: elevation, slope, and vegetation condition. The weights of these three factors in the 5th grid are 0.3, 0.3, and 0.2, respectively. The corresponding global weight coefficients are: elevation weight 0.5, slope weight 0.3, and vegetation condition weight 0.2. Then, we calculate... The value is 0.28. The global weight coefficient for each factor is dynamically adjusted. Vegetation conditions are used to describe the type, coverage, and growth status of surface vegetation. In spatial interpolation, the distance between grid points and observation points is the spatial distance between these two locations. Distance is a key weighting factor in spatial interpolation. Generally, the closer a grid point is to an observation point, the greater the influence of that observation point on its estimated value.

[0067] Specifically, based on β, downscaling data for each grid is generated using intermediate data, including:

[0068] Step 1: Calculate the downscaled grid and downscaled resolution c corresponding to each grid based on the intermediate scale transformation coefficient β. If c < b, then correct the downscaled resolution of the grid to b. If c > b, then use the bilinear interpolation algorithm to interpolate the intermediate data to obtain the first-level downscaled grid and first-level downscaled data corresponding to each grid.

[0069] In some embodiments, the intermediate scale transformation coefficient β reflects the combined impact of terrain complexity, observation station density, and mesoscale data quality for each grid, determining how to adjust the data resolution. The downscaling grid and its corresponding downscaling resolution c are calculated using the formula c=β×m. During downscaling, each grid uses data units at different scales. Different downscaling grids may have different resolutions c. If c < b, it indicates over-refinement and needs to be adjusted to b. If c > b, it indicates that downscaling is not yet complete and further interpolation using bilinear interpolation is needed to obtain the first-level downscaling grid and first-level downscaling data for each grid. As an example, if c is 3km and b is 5km, c < b, so b = 5km is directly assigned to c = b = 5km. When c is 3km and b is 1km, c > b. For each grid, the intermediate data is interpolated using bilinear interpolation to generate the corresponding downscaling data. For example, ... Figure 3As shown in the figure, each of the four 3km×3km grids is divided into nine 1km×1km grids, resulting in downscaled data with a resolution of 1km. The first-level downscaled grid is the initial downscaled grid, and its resolution may still be greater than b, requiring further processing. The first-level downscaled data is the first-stage downscaled data calculated through bilinear interpolation and still needs optimization. Downscaled data refers to the specific data values ​​(such as temperature, wind speed, etc.) calculated after downscaling. Downscaled data provides specific values ​​within each grid cell based on the processed grid, representing the numerical value of a specific physical quantity on the refined grid, reflecting the meteorological or environmental characteristics of the area after downscaling. For example, in a 1km×1km downscaled grid, the temperature, humidity, or wind speed of each grid cell is the downscaled data, calculated through interpolation or simulation.

[0070] Step 2: For each grid in the first-level downscaling grid, generate second-level downscaling data using the first-level downscaling data and WRF. This includes setting the parent domain resolution to c, generating the initial field and boundary conditions of the parent domain using mesoscale data, running the subdomain simulation, and generating second-level downscaling data with a subdomain resolution of b.

[0071] In some embodiments, the WRF (Weather Research and Forecasting) model is a high-resolution numerical weather prediction model used to simulate changes in weather conditions. Here, the WRF model is used for second-level downscaling, further refining the data based on the first-level downscaling grid. Based on this, the WRF model is used to simulate on the parent domain, generating higher-resolution second-level downscaled data for the subdomain. Specifically, the parent domain resolution is set to c. The initial field and boundary conditions of the parent domain are extracted from mesoscale data. Based on the parent domain, a smaller region (i.e., the subdomain) is defined, and the WRF model is run in this subdomain. The simulation yields higher-resolution data (e.g., 1km × 1km), called second-level downscaled data. This data reflects more refined meteorological or environmental characteristics within the subdomain. Here, the subdomain resolution is b, meaning the data within the subdomain will have a finer spatial resolution than the parent domain, for example, reduced from 3km to 1km. This allows us to obtain more accurate meteorological or environmental data, especially in areas requiring high resolution (e.g., wind energy resource assessment). The parent domain refers to a larger region, typically the region of the first-level downscaling data.

[0072] This includes generating downscaling data for each grid based on β and intermediate data, and also includes:

[0073] The boundary transition zone of the second-level downscaling data is fused to obtain downscaling data, including: defining the grids with a T-value change > 0.2 / adjacent grids as the abrupt grid boundary region, and using a Laplace smoothing filter to smooth the abrupt grid boundary region.

[0074] In some embodiments, the boundary transition zone refers to the transition region between data of different resolutions that typically occurs during downscaling. For example, when downscaling from the resolution of a parent domain (e.g., 3 km) to the resolution of a child domain (e.g., 1 km), the boundary region may exhibit discontinuous or uneven transitions. This transition zone can cause abrupt changes in the downscaled data near the boundary, especially at the interface between different resolution regions, where the data may change drastically, contradicting the actual natural transition pattern. Therefore, for the boundary transition zone of second-level downscaling data, if the abrupt change in the terrain complexity index between two adjacent grids exceeds a preset threshold (i.e., a T-value change > 0.2), these two grids are defined as the boundary region of abrupt grids. A Laplace smoothing filter is then used to adjust the values ​​in the abrupt region, making the changes in these regions smoother. The Laplace smoothing filter uses a Gaussian kernel to smooth the data. As an example, for each data point in a 3×3 window, a new smoothed data point is calculated by convolving with the Gaussian kernel.

[0075] The method for generating downscaling data using mesoscale data and spatial terrain data of the present invention further includes:

[0076] Computing resources are reallocated according to a preset time period. During the reallocation process, double-precision floating-point operations are assigned to grids with T≥0.7, and the total processor utilization of each grid with T≥0.7 is adjusted to the preset utilization rate.

[0077] In some embodiments, computational resource reallocation involves dynamically adjusting the allocation of computational resources based on the processing needs of different grids. This is typically done to optimize computational efficiency, allocating more resources to areas requiring more computational resources (such as high-complexity areas) and reducing resource allocation to areas with lower computational needs. During mesoscale data downscaling, some grids may require more computational resources (e.g., higher precision or more processing power). Therefore, reallocating computational resources can improve overall processing speed and accuracy, avoiding resource waste. T represents a specific computational complexity index, used to quantify the computational needs of different grids. The T value can be based on the grid's terrain complexity index; here, grids with T ≥ 0.7 refer to grid regions with a high terrain complexity index. If a grid has T ≥ 0.7, double-precision floating-point operations are allocated to process the data in that grid, ensuring sufficient precision during computation. If T < 0.7, single-precision floating-point operations may be allocated. Double-precision floating-point operations provide higher computational precision. They use 64 bits to represent values, which can represent a larger range and more precision compared to single-precision (32 bits). Total processor utilization refers to the proportion of computer processor resources used by a particular task (or grid) during computation. It is typically a parameter in a computational resource scheduling system, representing the computational power required by the task. Based on this, for grids with T ≥ 0.7, due to their high computational demands, the computational resource management system adjusts their processor utilization to ensure these grids can complete computations within a reasonable timeframe. For example, if a grid has high computational demands, multiple CPU cores are allocated to compute that grid simultaneously. For grids with T ≥ 0.7, 60% of the computational resources are allocated.

[0078] The method for generating downscaling data using mesoscale data and spatial terrain data of the present invention further includes:

[0079] Through an online learning mechanism, the difference between the station observations within the grid and the generated downscaled data is calculated periodically. When the difference is greater than twice the standard deviation, a global weight coefficient adjustment strategy is triggered.

[0080] In some embodiments, the online learning mechanism is a machine learning method that dynamically adjusts model parameters. It can process data streams in real time and continuously optimize the model, meaning that model parameters are continuously updated during model operation to optimize them as the data changes. Here, online learning is used to dynamically adjust the computational strategy for downscaling data. Grid-based station observations refer to the actual values ​​(such as temperature, humidity, wind speed, etc.) measured by actual meteorological observation stations within each grid. Based on this, subtraction is used at fixed time intervals to calculate the error between the grid-based station observations and the generated downscaled data. The mean and standard deviation of the error are calculated over a period of time. If the error is greater than twice the standard deviation, the backend server queries the corresponding global weight coefficient from the global weight coefficient adjustment strategy library. The global weight coefficient adjustment strategy library includes ranges greater than twice the standard deviation and their corresponding global weight coefficients.

[0081] The calculation of the terrain complexity index for each grid in the intermediate data, based on a pre-determined set of terrain assessment features, also includes:

[0082] Based on b, a geospatial data cloud with a matching resolution is selected from geospatial data clouds of different resolutions in the specified area as the target geospatial data cloud, which is used to calculate the terrain complexity index corresponding to each grid.

[0083] In some embodiments, a geospatial data cloud refers to a dataset storing geographic information at different resolutions, including information such as elevation, slope, and vegetation. The geospatial data cloud is pre-stored on a local server. The process involves selecting a geospatial data cloud with a matching resolution from the geospatial data clouds of a specified area, using the parameter 'b'. The target geospatial data cloud refers to the geographic data that best matches 'b' from a dataset of multiple resolutions, serving as the data source for calculating the terrain complexity index. For example, assuming 'b' = 1km × 1km, then terrain data with a 1km × 1km resolution would be selected from the data cloud.

[0084] The method for generating downscaling data using mesoscale data and spatial terrain data of the present invention further includes:

[0085] After generating downscaled data, multiple test points are selected in the corresponding downscaled grid, and measurements are taken at the test points using a wind tower to obtain actual meteorological data for multiple horizontal planes.

[0086] Calculate the correlation coefficient between the test points and the actual meteorological data.

[0087] In some embodiments, test points refer to representative locations selected from the downscaled grid for actual measurements. Test points can be areas with complex terrain, high wind speeds, or areas with significant wind resource potential. Based on this, after generating downscaled data, multiple test points are selected from the downscaled grid, and actual meteorological data at multiple horizontal planes are measured at these test points using anemometer towers. For example, a high wind speed area and an area with significant wind resource potential are selected as test points, and wind speed, direction, and other data are measured at heights of 10m, 50m, and 100m in these two areas using anemometer towers. The correlation coefficient between the actual meteorological data and the downscaled data corresponding to the test points is calculated. The correlation coefficient can be the Pearson correlation coefficient, used to measure the linear correlation between two variables. For example, if the correlation coefficient is 1, it indicates that the downscaled wind speed data perfectly matches the anemometer tower data. If the correlation coefficient is less than 0.6, it indicates that the downscaled wind speed data has a large error compared to the anemometer tower data, and the model needs optimization.

[0088] In these embodiments, the downscaling accuracy of mesoscale data is improved to make it more suitable for complex terrain regions, and computational resource allocation and online learning capabilities are optimized to enhance the accuracy of weather forecasting and wind energy assessment. The accuracy of downscaling data is improved through grid-based dynamic weight calculation, interpolation methods, and WRF simulation to better reflect the actual conditions of complex terrain regions. A terrain complexity index β is adopted, and weight coefficients are dynamically adjusted to ensure that downscaling data more accurately reflects meteorological changes under different terrain conditions. A Laplace smoothing filter is used to process abrupt grid boundaries, resolving data abrupt changes caused by interpolation or simulation, and improving the continuity and stability of downscaling data. A computational resource reallocation strategy allocates double-precision floating-point computing resources to areas with high computational demands (T≥0.7), improving computational efficiency and reducing redundant calculations. An online learning mechanism is employed to periodically calculate the error between downscaling data and measured data; when the error exceeds a threshold, global weights are adjusted to improve the real-time adaptability and accuracy of downscaling data. Based on the user-specified downscaling resolution b, the most matching data is selected from geospatial data clouds of different resolutions to improve the accuracy of terrain complexity index calculation. Measured meteorological data are obtained by using a wind measurement tower, and the correlation coefficient between the data and the downscaling data is calculated to evaluate the reliability and applicability of the downscaling method.

[0089] The above description is merely a selection of preferred embodiments of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to specific combinations of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A method for generating downscaling data using mesoscale data and spatial topographic data, characterized in that, include: Based on the data resolution 'a' of the mesoscale data in the specified area and the resolution 'b' of the downscaling data specified by the user, a global scale transformation coefficient 'α' is generated. If 'α' ≥ a preset threshold 'n', then the high-resolution meteorological observation data, combined with the spatial dynamic weights of the grid, is used to interpolate the mesoscale data to obtain intermediate data with a resolution of 'm', where 'a' < 'm' < 'b'. Based on a pre-determined set of terrain assessment features, the terrain complexity index corresponding to each grid of the intermediate data is calculated. Based on the terrain complexity index, the density of observation stations, and the quality of mesoscale data, the adaptive mesoscale transformation coefficient β corresponding to each grid is generated. Based on β, the downscaled data for each grid is generated using intermediate data. Based on the downscaled data, the installation location of the wind turbine was determined and the wind turbine was installed. No. The spatial dynamic weight of each grid The calculation formula is as follows: in, It is the first The first grid The weights of each factor This is the global weight coefficient of the factor, which is one of the following: elevation, slope, vegetation condition, distance between grid points and observation points. The global weight coefficient of each factor is dynamically adjusted. The step of generating downscaling data for each grid based on β and using intermediate data includes: Based on the intermediate scale transformation coefficient β, calculate the downscaled grid and the downscaled resolution c corresponding to each grid. If c < b, then correct the downscaled resolution of the grid to b. If c > b, then use the bilinear interpolation algorithm to interpolate the intermediate data to obtain the first-level downscaled grid and the first-level downscaled data corresponding to each grid. For each grid in the first-level downscaling grid, second-level downscaling data is generated using the first-level downscaling data and WRF. This includes setting the parent domain resolution to c, generating the initial field and boundary conditions of the parent domain using mesoscale data, running the subdomain simulation, and generating second-level downscaling data with a subdomain resolution of b.

2. The method for generating downscaling data using mesoscale data and spatial topographic data according to claim 1, characterized in that, The step of generating downscaling data for each grid based on β and intermediate data also includes: The boundary transition zone of the second-level downscaling data is fused to obtain downscaling data, including: defining the grids with a sudden change in T value > 0.2 as the boundary zone of the abrupt grids, and using a Laplace smoothing filter to smooth the boundary zone of the abrupt grids.

3. The method for generating downscaling data using mesoscale data and spatial topographic data according to claim 1, characterized in that, Also includes: Computing resources are reallocated according to a preset time period. During the reallocation process, double-precision floating-point operations are assigned to grids with T≥0.7, and the total processor utilization of each grid with T≥0.7 is adjusted to the preset utilization rate.

4. The method for generating downscaling data using mesoscale data and spatial topographic data according to claim 1, characterized in that, Also includes: Through an online learning mechanism, the difference between the station observations within the grid and the generated downscaled data is calculated periodically. When the difference is greater than twice the standard deviation, a global weight coefficient adjustment strategy is triggered.

5. The method for generating downscaling data using mesoscale data and spatial topographic data according to claim 1, characterized in that, The calculation of the terrain complexity index for each grid in the intermediate data, based on a pre-defined set of terrain assessment features, also includes: Based on b, a geospatial data cloud with a matching resolution is selected from geospatial data clouds of different resolutions in the specified area as the target geospatial data cloud, which is used to calculate the terrain complexity index corresponding to each grid.

6. The method for generating downscaling data using mesoscale data and spatial topographic data according to claim 1, characterized in that, Also includes: After generating downscaled data, multiple test points are selected in the corresponding downscaled grid, and measurements are taken at the test points using a wind tower to obtain actual meteorological data for multiple horizontal planes. Calculate the correlation coefficient between the test points and the actual meteorological data.

Images