A wind resource prediction method, device, equipment and medium for meso-scale terrain

Abstract

The present application provides a method, device, equipment and medium for wind resource prediction of meso-scale terrain, including: extracting surface roughness data for the target terrain, and performing hydrodynamic numerical simulation processing on the target terrain after grid division to determine three-dimensional flow field data; performing data preprocessing on the grid coordinates, three-dimensional flow field data and surface roughness data of the target terrain to determine a fixed-dimensional terrain elevation matrix and a fixed-dimensional surface roughness matrix; calculating the blocking coefficient in the oncoming flow direction for the fixed-dimensional terrain elevation matrix to determine the wind field blocking characteristics of the target terrain; inputting the fixed-dimensional terrain elevation matrix, wind field blocking characteristics and fixed-dimensional surface roughness matrix into a wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling and feature fusion to predict the wind resource result map of the target terrain. It can quickly and accurately predict the wind resource results of any terrain.

Description

Technical Field

[0001] The present application relates to the technical field of wind resource prediction, and in particular, to a method, device, equipment and medium for wind resource prediction of meso - micro scale terrain. Background Art

[0002] Current wind resource prediction mainly relies on numerical weather prediction models, statistical analysis methods, and traditional machine learning algorithms. Numerical weather prediction models simulate and predict future wind speed and direction by solving atmospheric physical equations, combining terrain data and real - time weather reports. However, the existing wind resource prediction technologies have the following main defects in data collection, model performance, and production deployment: (1) Difficult data collection: The collection of wind resource data mainly relies on limited measuring points that are discretely distributed in space and time. Although these measuring points can obtain wind resource information within a certain range, from the perspective of the overall wind field, they are only local samples and cannot comprehensively and continuously cover the entire wind field space. (2) Insufficient model performance: Traditional methods such as numerical simulation methods and machine learning methods have problems such as low accuracy and slow operation speed. For example, numerical simulation requires a large amount of computing resources and has a slow prediction speed; while machine learning methods lack accuracy, especially under complex terrain and irregular wind direction conditions, the prediction effect is limited. (3) Low accuracy of wind resource prediction. Therefore, how to improve the accuracy of wind resource prediction has become a technical problem that cannot be underestimated. Summary of the Invention

[0003] In view of this, the purpose of the present application is to provide a method, device, equipment and medium for wind resource prediction of meso - micro scale terrain. Through the wind resource prediction model, the wind resource results of any terrain of meso - micro scale can be quickly and accurately predicted, while enhancing the adaptability and generality of the model and reducing the dependence on a large amount of historical data and high - performance hardware.

[0004] The embodiment of the present application provides a method for wind resource prediction of meso - micro scale terrain. The wind resource prediction method includes:

[0005] Extract the surface roughness data of the target terrain, and perform hydrodynamic numerical simulation processing on the meshed target terrain to determine the three - dimensional flow field data of the target terrain;

[0006] Perform data pre - processing on the grid coordinates, the three - dimensional flow field data, and the surface roughness data of the target terrain to determine a fixed - dimension terrain elevation matrix and a fixed - dimension surface roughness matrix;

[0007] Calculate the blocking coefficient of the fixed - dimension terrain elevation matrix in the oncoming flow direction to determine the wind field blocking characteristics of the target terrain;

[0008] Input the fixed - dimension terrain elevation matrix, the wind - field barrier feature, and the fixed - dimension surface roughness matrix into the wind resource prediction model for convolutional downsampling, non - linear transformation, self - attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a fixed - dimension flow - field matrix at the target height as the training target, and the initial network model is a UNet - architecture convolutional neural network model with multiple Transformer layers added.

[0009] In a possible implementation manner, the data pre - processing of the grid coordinates of the target terrain, the three - dimensional flow - field data, and the surface roughness data to determine the fixed - dimension terrain elevation matrix and the fixed - dimension surface roughness matrix includes:

[0010] Perform linear interpolation on the underlying surface grid coordinates in the grid coordinates of the target terrain to a fixed - dimension two - dimensional terrain matrix, and normalize the original terrain height of each terrain data point based on the lowest height in the two - dimensional terrain matrix to determine the fixed - dimension terrain elevation matrix;

[0011] Based on the linear interpolation, perform conversion processing, outlier removal processing, and maximum - minimum normalization processing on the surface roughness data of each terrain data point to determine the fixed - dimension surface roughness matrix.

[0012] In a possible implementation manner, the calculation of the barrier coefficient in the oncoming - flow direction for the fixed - dimension terrain elevation matrix to determine the wind - field barrier feature of the target terrain includes:

[0013] For each terrain data point, starting from this terrain data point, judge point - by - point along the straight line against the wind direction to determine whether the terrain elevation data of the upwind terrain data points on the straight line against the wind direction is greater than the terrain elevation data of this terrain data point;

[0014] If so, based on the height difference between the terrain elevation data of the upwind terrain data point and the terrain elevation data of the terrain data point, and the straight - line distance between the upwind terrain data point and the terrain data point, determine the barrier coefficient of the upwind terrain data point to this terrain data point;

[0015] Accumulate the barrier coefficients of each terrain data point to determine the comprehensive barrier coefficient of each terrain data point;

[0016] Perform maximum - minimum normalization processing on multiple comprehensive barrier coefficients to determine the wind - field barrier feature of the target terrain.

[0017] In a possible implementation, the blocking coefficient of the upwind terrain data point to the terrain data point is determined through the following steps:

[0018]

[0019]

[0020] Wherein, is the blocking coefficient of the terrain data point ; is the upwind terrain data point on the upwind straight line, The value range of is , z is the terrain elevation data, t is the height difference, L is the terrain length of the fixed-dimension terrain elevation data, is the sign function.

[0021] In a possible implementation, inputting the fixed-dimension terrain elevation matrix, the wind field blocking feature, and the fixed-dimension surface roughness matrix into the wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain includes:

[0022] Based on multiple encoding layers of the wind resource prediction model, convolutional downsampling feature extraction is sequentially performed on the fixed-dimension terrain elevation matrix, the wind field blocking feature, and the fixed-dimension surface roughness matrix to determine encoding feature maps of different scales;

[0023] Based on multiple Transformer layers, non-linear transformation and self-attention mechanism processing are performed on the deep encoding feature map output by the deepest encoding layer to extract implicit features including local information and global information of the deep encoding feature map;

[0024] Based on multiple decoding layers of the wind resource prediction model, convolutional upsampling processing is performed on the implicit features and multiple encoding feature maps, and during the decoding process, feature fusion is performed on the decoding feature map output by the decoding layer and the corresponding encoding feature map in a skip connection manner;

[0025] Based on the convolutional layer of the wind resource prediction model, convolutional processing is performed on the decoding feature map output by the last decoding layer to generate the wind resource result map; wherein, the wind speed modulus under each terrain data point is displayed in the wind resource result map.

[0026] In a possible implementation, the wind resource prediction model is determined through the following steps:

[0027] Input the sample fixed - dimension terrain elevation matrices, sample wind field barrier characteristics, and sample fixed - dimension surface roughness matrices of multiple sample terrains into the initial network model to determine the predicted wind resource result map for each of the sample terrains;

[0028] Based on the linear interpolation algorithm, perform transformation processing, outlier removal processing, and maximum - minimum value processing on the sample three - dimensional flow field data of each sample terrain, and extract the wind speed component of each sample terrain data point at the target height from the ground;

[0029] Calculate the actual wind speed modulus of each sample terrain at the target height based on the wind speed components of each sample terrain data point;

[0030] Iteratively train the initial network model based on the mean square error and mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map to determine the wind resource prediction model.

[0031] In a possible implementation manner, the iterative training of the initial network model based on the mean square error and mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map to determine the wind resource prediction model includes:

[0032] Detect whether the mean square error and the mean absolute error are less than or equal to a preset threshold;

[0033] If either is no, perform hyperparameter adjustment and network parameter adjustment on the initial network model, and continue the iterative training of the initial network model after parameter adjustment;

[0034] If both are yes, use the initial network model as the wind resource prediction model.

[0035] The embodiment of the present application also provides a wind resource prediction device for mesoscale terrain. The wind resource prediction device includes:

[0036] An acquisition module, configured to extract surface roughness data for the target terrain, and perform computational fluid dynamics numerical simulation processing on the meshed target terrain to determine the three - dimensional flow field data of the target terrain;

[0037] A pre - processing module, configured to perform data pre - processing on the grid coordinates, the three - dimensional flow field data, and the surface roughness data of the target terrain to determine a fixed - dimension terrain elevation matrix and a fixed - dimension surface roughness matrix;

[0038] A wind field blocking feature construction module, configured to calculate a blocking coefficient in the oncoming flow direction for the fixed-dimension terrain elevation matrix, and determine the wind field blocking feature of the target terrain;

[0039] A wind resource prediction module, configured to input the fixed-dimension terrain elevation matrix, the wind field blocking feature, and the fixed-dimension surface roughness matrix into a wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion, and predict a wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a target height fixed-dimension flow field matrix as the training target, and the initial network model is a UNet architecture convolutional neural network model added with multiple Transformer layers.

[0040] An embodiment of the present application further provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the wind resource prediction method for the mesoscale terrain as described above are executed.

[0041] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of the wind resource prediction method for the mesoscale terrain as described above are executed.

[0042] A method, device, equipment and medium for wind resource prediction of meso - micro scale terrain provided by an embodiment of the present application. The wind resource prediction method includes: extracting surface roughness data for a target terrain, and performing hydrodynamic numerical simulation processing on the target terrain after grid division to determine three - dimensional flow field data of the target terrain; performing data pre - processing on the grid coordinates, the three - dimensional flow field data and the surface roughness data of the target terrain to determine a fixed - dimension terrain elevation matrix and a fixed - dimension surface roughness matrix; calculating a blocking coefficient in the oncoming flow direction for the fixed - dimension terrain elevation matrix to determine the wind field blocking characteristics of the target terrain; inputting the fixed - dimension terrain elevation matrix, the wind field blocking characteristics and the fixed - dimension surface roughness matrix into a wind resource prediction model for convolutional down - sampling, non - linear transformation, self - attention mechanism, convolutional up - sampling and feature fusion to predict a wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a fixed - dimension flow field matrix at a target height as the training target, and the initial network model is a UNet - architecture convolutional neural network model with multiple Transformer layers added. The beneficial effect of the present application is that: through the wind resource prediction model, the wind resource results of any meso - micro scale terrain can be predicted quickly and accurately, while enhancing the adaptability and generality of the model, and reducing the dependence on a large amount of historical data and high - performance hardware.

[0043] To make the above - mentioned objects, features and advantages of the present application more obvious and understandable, the following specific preferred embodiments are given, and in conjunction with the accompanying drawings, the detailed description is as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1 A flowchart of a method for wind resource prediction of meso - micro scale terrain provided by an embodiment of the present application;

[0046] Figure 2 A schematic diagram of the UI interface display of the wind resource result map provided by an embodiment of the present application;

[0047] Figure 3 A schematic diagram of the structure of a device for wind resource prediction of meso - micro scale terrain provided by an embodiment of the present application;

[0048] Figure 4The second structural schematic diagram of a wind resource prediction device for meso - microscale terrain provided by an embodiment of the present application;

[0049] Figure 5 The structural schematic diagram of an electronic device provided by an embodiment of the present application. Specific implementation manners

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Usually, the components of the embodiments of the present application described and illustrated herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by those skilled in the art without creative efforts belongs to the scope of protection of the present application.

[0051] First, the applicable application scenarios of the present application are introduced. The present application can be applied to the field of wind resource prediction technology.

[0052] Through research, it is found that current wind resource prediction mainly relies on numerical weather prediction models, statistical analysis methods, and traditional machine learning algorithms. The numerical weather prediction model solves the atmospheric physical equations, combines terrain data and real - time weather reports, and simulates and predicts future wind speed and direction. However, the existing wind resource prediction technologies have the following main defects in data collection, model performance, and production deployment: (1) Difficult data collection: The collection of wind resource data mainly relies on a limited number of measuring points that are discretely distributed in space and time. Although these measuring points can obtain wind resource information within a certain range, from the perspective of the overall wind field, they are only local samples and cannot comprehensively and continuously cover the entire wind field space. (2) Insufficient model performance: Traditional methods such as numerical simulation methods and machine learning methods have problems such as low accuracy and slow operation speed. For example, numerical simulation requires a large amount of computing resources and has a slow prediction speed; while machine learning methods have deficiencies in accuracy, especially under complex terrain and irregular wind direction conditions, the prediction effect is limited. (3) Low accuracy of wind resource prediction. Therefore, how to improve the accuracy of wind resource prediction has become a technical problem that cannot be underestimated.

[0053] Based on this, the embodiments of the present application provide a wind resource prediction method for meso - microscale terrain. Through the wind resource prediction model, the wind resource results of any terrain in the meso - microscale can be quickly and accurately predicted, while enhancing the adaptability and generality of the model and reducing the dependence on a large amount of historical data and high - performance hardware.

[0054] Please refer to Figure 1 , Figure 1 , which is a flowchart of a wind resource prediction method for mesoscale terrain provided by an embodiment of the present application. As Figure 1 shown in

[0055] S101: Extract the surface roughness data of the target terrain, and perform hydrodynamic numerical simulation processing on the meshed target terrain to determine the three-dimensional flow field data of the target terrain.

[0056] In this step, first extract the surface roughness data from the target terrain obtained from the terrain database, then perform meshing processing on the target terrain, and perform hydrodynamic numerical simulation processing on the meshed target terrain to determine the three-dimensional flow field data of the target terrain.

[0057] Here, a rectangular area of size L×L is cropped from the terrain database according to the given longitude and latitude coordinates, and then the area is discretized into grids at a specified resolution. Then, according to the set target inflow and boundary conditions, hydrodynamic numerical simulation processing is performed on the target terrain to determine the three-dimensional flow field data of the target terrain. In addition, the surface roughness data of the target terrain in the selected area is extracted and saved.

[0058] In a specific embodiment, according to the predetermined longitude and latitude coordinates, a rectangular area with a size of L×L at the target point is cropped from the terrain database, and the cropped rectangular area is meshed at a resolution of 25m to obtain the grid data of the target terrain . For the target calculation area, set the boundary inflow conditions. Further, set the inflow direction to east wind, and set the inflow wind speed profile to the wind speed profile with a thermal stability of 5. Set the bottom surface of the calculation domain as a no-slip boundary, and the rest as free boundary conditions. According to the initial boundary conditions, use the RANS model to perform numerical simulation on the target calculation domain. After the flow is stable, select the three-dimensional flow field result at the final fixed time step as the three-dimensional flow field data corresponding to the target area, and extract and output the roughness data of the target area.

[0059] S102: Perform data preprocessing on the grid coordinates, the three-dimensional flow field data, and the surface roughness data of the target terrain to determine a fixed-dimensional terrain elevation matrix and a fixed-dimensional surface roughness matrix.

[0060] In this step, perform data preprocessing on the grid coordinates, the three-dimensional flow field data, and the surface roughness data of the target terrain to determine a fixed-dimensional terrain elevation matrix and a fixed-dimensional surface roughness matrix.

[0061] Here, data preprocessing includes terrain data format conversion processing, data cleaning, and data normalization processing, etc.

[0062] In a possible implementation manner, preprocessing the grid coordinates of the target terrain, the three-dimensional flow field data, and the surface roughness data to determine a fixed-dimension terrain elevation matrix and a fixed-dimension surface roughness matrix includes:

[0063] A: Linearly interpolate the underlying surface grid coordinates in the grid coordinates of the target terrain to a two-dimensional terrain matrix with a fixed dimension, and normalize the original terrain height of each terrain data point based on the lowest height in the two-dimensional terrain matrix to determine the fixed-dimension terrain elevation matrix.

[0064] Here, linearly interpolate the underlying surface grid coordinates in the grid coordinates of the target terrain to a two-dimensional terrain matrix with a fixed dimension, normalize the original terrain height of each terrain data point according to the lowest height in the two-dimensional terrain matrix, and determine the fixed-dimension terrain elevation matrix.

[0065] Among them, from the three-dimensional flow field data obtained by CFD calculation, first extract the grid coordinate information of the underlying surface as the initial terrain data. Then, taking the lowest point in the area as the reference, normalize the height of the original terrain data to obtain unified relative height information. Subsequently, through linear interpolation, convert the processed terrain elevation data into a matrix with a fixed dimension to ensure that it can adapt to the input requirements of the model.

[0066] B: Based on the linear interpolation, perform conversion processing, outlier removal processing, and maximum-minimum normalization processing on the surface roughness data of each terrain data point to determine the fixed-dimension surface roughness matrix.

[0067] Here, perform conversion processing, outlier removal processing, and maximum-minimum normalization processing on the surface roughness data of each terrain data point according to the linear interpolation to determine the fixed-dimension surface roughness matrix.

[0068] Among them, eliminate the cases where the CFD calculation does not converge through outlier identification to ensure data quality. Statistically calculate the maximum and minimum values of the terrain elevation data and the surface roughness data respectively, subtract the minimum value of the current terrain data from each terrain data point to unify the relative height reference. Perform maximum-minimum normalization on all data types (terrain elevation data and surface roughness data), as shown in the following formula:

[0069]

[0070] Among them, is the value after normalization, is the original data value, is the minimum value of this data type, is the maximum value of this data type.

[0071] Here, according to different wind direction conditions, taking the negative direction of the X-axis of the wind direction as the reference, the terrain data is relatively rotated so that the rotated terrain uniformly takes the negative direction of the X-axis as the oncoming flow direction, which is convenient for the subsequent use of the network model with a unified input form.

[0072] S103: Calculate the blocking coefficient in the oncoming flow direction for the fixed-dimensional terrain elevation matrix to determine the wind field blocking characteristics of the target terrain.

[0073] In this step, calculate the blocking coefficient in the oncoming flow direction for the terrain elevation data in the fixed-dimensional terrain elevation matrix to determine the wind field blocking characteristics of the target terrain, that is, numerically visualize the interaction relationship between the terrain and the oncoming flow.

[0074] In a possible implementation manner, the calculating the blocking coefficient in the oncoming flow direction for the fixed-dimensional terrain elevation matrix to determine the wind field blocking characteristics of the target terrain includes:

[0075] a: For each terrain data point, starting from this terrain data point, judge point by point along the straight line against the wind direction to determine whether the terrain elevation data of the upwind terrain data points on the straight line against the wind direction is greater than the terrain elevation data of this terrain data point.

[0076] Here, for each terrain data point, starting from this terrain data point, judge point by point along the straight line against the wind direction to determine whether the terrain elevation data of the upwind terrain data points on the straight line against the wind direction is greater than the terrain elevation data of this terrain data point.

[0077] b: If so, based on the height difference between the terrain elevation data of the upwind terrain data point and the terrain elevation data of the terrain data point and the straight-line distance between the upwind terrain data point and the terrain data point, determine the blocking coefficient of the upwind terrain data point to this terrain data point.

[0078] Here, if so, according to the height difference between the terrain elevation data of the upwind terrain data point and the terrain elevation data of the terrain data point and the straight-line distance between the upwind terrain data point and the terrain data point, determine the blocking coefficient of the upwind terrain data point to the terrain data point.

[0079] In a possible implementation manner, the blocking coefficient of the upwind terrain data point to the terrain data point is determined through the following steps:

[0080]

[0081]

[0082] Among them, is the terrain data point of the barrier coefficient, is the upwind terrain data point on the upwind straight line, The value range of is z is the terrain elevation data, t is the height difference, L is the terrain length of the fixed-dimensional terrain elevation data, is the sign function.

[0083] Here, L can be set to 256, The role of is to judge when is greater than

[0084] Among them, the wind field barrier characteristics can be disassembled into two parts: height difference and upwind straight-line distance. Taking , the height difference as an example, the set of points on the upwind straight line is , The closer the distance between and is, the larger the value of the barrier coefficient, indicating that the upwind straight-line point has a greater oncoming flow barrier to the terrain data point ; as gradually increases, the distance from the terrain data point

[0085] c: Accumulate the barrier coefficients of each of the terrain data points to determine the comprehensive barrier coefficient of each of the terrain data points.

[0086] Here, accumulate the barrier coefficients of each terrain data point to determine the comprehensive barrier coefficient of each terrain data point.

[0087] c: Perform maximum-minimum normalization processing on the multiple comprehensive barrier coefficients to determine the wind field barrier characteristics of the target terrain.

[0088] Here, perform maximum-minimum normalization processing on the multiple comprehensive barrier coefficients to determine the wind field barrier characteristics of the target terrain.

[0089] S104: Input the fixed - dimensional terrain elevation matrix, the wind field barrier feature, and the fixed - dimensional surface roughness matrix into the wind resource prediction model for convolutional downsampling, non - linear transformation, self - attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain.

[0090] In this step, the fixed - dimensional terrain elevation matrix, the wind field barrier feature, and the fixed - dimensional surface roughness matrix are input into the wind resource prediction model for convolutional downsampling, non - linear transformation, self - attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain.

[0091] Among them, the wind resource result map of the target terrain shows the wind speed modulus of each terrain data point.

[0092] Among them, the wind resource prediction model is obtained by training the initial network model with the fixed - dimensional flow field matrix at the target height as the training target. The initial network model is obtained by adding multiple Transformer layers to the UNet - architecture convolutional neural network model.

[0093] Among them, the fixed - dimensional flow field matrix at the target height is to analyze and load the original three - dimensional flow field data file (.vtk format), extract the flow field information at a height of 100 meters above the ground. The nearest - neighbor interpolation algorithm is used to accurately obtain the wind speed component data (u, v, w) at this height level from the three - dimensional flow field. Subsequently, based on the wind speed component data, the wind speed modulus U is calculated through vector synthesis. Finally, using linear interpolation, the wind speed modulus data in the original coordinate system is resampled into a fixed - dimensional matrix to ensure its compatibility with the model input format.

[0094] Among them, in this application, in addition to using the UNet - architecture convolutional neural network model to construct the initial network model, it can also be constructed according to other variant architectures such as TransUNet, and this part is not specifically limited.

[0095] In a possible implementation manner, the step of inputting the fixed - dimensional terrain elevation matrix, the wind field barrier feature, and the fixed - dimensional surface roughness matrix into the wind resource prediction model for convolutional downsampling, non - linear transformation, self - attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain includes:

[0096] (1): Based on multiple encoding layers of the wind resource prediction model, sequentially perform convolutional downsampling feature extraction on the fixed - dimensional terrain elevation matrix, the wind field barrier feature, and the fixed - dimensional surface roughness matrix to determine encoding feature maps of different scales.

[0097] Here, the encoding layer uses convolutional kernels to extract features from the wind field obstruction features, the fixed-dimension terrain elevation matrix, and the fixed-dimension surface roughness matrix. During the encoding process, the size of the feature map gradually decreases, while the number of channels gradually increases. And downsampling operations are performed after each encoding layer to halve the spatial size of the feature map and increase the number of channels, thereby obtaining encoded feature maps of different scales.

[0098] (2): Based on multiple Transformer layers, perform non-linear transformation and self-attention mechanism processing on the deep encoded feature map output by the deepest encoding layer, and extract the implicit features including local information and global information of the deep encoded feature map.

[0099] Here, the Transformer layer is different from the classical UNet. The model proposed in this solution introduces multiple Transformer layers in the deep encoding stage, so as to be able to convert and serialize the input image, and then learn to extract the implicit features including local and global information in the deep encoded feature map through multiple Transformer layers. These features are then reshaped to match the size of the deepest encoding layer.

[0100] (3): Based on multiple decoding layers of the wind resource prediction model, perform convolutional upsampling processing on the implicit features and multiple encoded feature maps, and perform feature fusion on the decoded feature map output by the decoding layer and the corresponding encoded feature map in the decoding process by means of skip connections.

[0101] Here, according to multiple decoding layers of the wind resource prediction model, perform convolutional upsampling processing on the implicit features and multiple encoded feature maps, and perform feature fusion on the decoded feature map output by the decoding layer and the corresponding encoded feature map in the decoding process by means of skip connections.

[0102] Among them, in the decoding stage, the model gradually restores the spatial resolution of the feature map through transposed convolutional layers, while reducing the number of channels. This process reconstructs the spatial details of the image and fuses feature maps of different scales to enhance the model's ability to recognize complex structures. The model adopts a full-scale skip connection method, allowing the combination of feature maps of the encoding layer and the decoding layer during the decoding process to achieve full-scale feature fusion.

[0103] (4): The convolutional layer of the wind resource prediction model performs convolution processing on the decoded feature map output by the last decoding layer to generate the wind resource result map; wherein, the wind speed modulus under each terrain data point is shown in the wind resource result map.

[0104] Here, the convolutional layer performs convolution processing on the decoded feature map output by the last decoding layer to generate the wind resource result map.

[0105] Among them, the functions and capabilities of the wind resource prediction model are as follows: Interactive interface: provides a simple and efficient UI, and the wind resource prediction results can be obtained by clicking operations; Data output: supports the export of prediction results, providing various formats such as npy and vtk, which is convenient for subsequent analysis and application; Real-time prediction: combines the latest meteorological and terrain data to dynamically update the prediction results to ensure the timeliness of the output.

[0106] Further, please refer to Figure 2 , Figure 2 which is a schematic diagram of the UI interface display of the wind resource result map provided by the embodiment of the present application. As Figure 2 shown, the wind speed modulus at any terrain data point of the target terrain is displayed on the output interface of the wind resource prediction. The color streamlines shown in Figure 2 represent different color displays corresponding to different wind speed moduli, so as to facilitate users to distinguish the distribution of wind resource aggregation under the target terrain.

[0107] In a specific embodiment, according to the wind field data acquisition module, a rectangular area of size L×L is cropped from the terrain database according to the given longitude and latitude coordinates, and then the area is discretized into grids at a specified resolution. Then, according to the set target incoming flow and boundary conditions, a CFD numerical simulation is performed on the target terrain to obtain the three-dimensional flow field of the target terrain, and the elevation data and roughness data of the selected area are also extracted and saved. The wind field data acquisition module is connected to an external data source through an interface, and the collected surface roughness data and three-dimensional flow field data are transmitted to the wind field data processing module for processing to obtain a fixed-dimensional terrain elevation matrix and a fixed-dimensional surface roughness matrix. The fixed-dimensional terrain elevation matrix is sent to the wind field barrier feature construction module to calculate the barrier coefficient of the terrain elevation data in the fixed-dimensional flow field matrix in the incoming flow direction, and the wind field barrier feature of the target terrain is determined. The wind field barrier feature, the fixed-dimensional terrain elevation matrix, and the fixed-dimensional surface roughness matrix are input into the wind resource prediction model to predict the wind resource result map of the target terrain.

[0108] In a possible implementation manner, the wind resource prediction model is determined through the following steps:

[0109] I: Input the sample fixed-dimensional terrain elevation matrices, sample wind field barrier features, and sample fixed-dimensional surface roughness matrices of multiple sample terrains into the initial network model to determine the predicted wind resource result maps of each sample terrain.

[0110] Here, the sample wind field barrier features, sample fixed-dimensional terrain elevation matrices, and sample fixed-dimensional surface roughness matrices of multiple sample terrains are input into the initial network model to determine the predicted wind resource result maps of each sample terrain.

[0111] Among them, the processing process of the predicted wind resource result map of the sample terrain is the same as that of the wind resource result map of the target terrain, and this part will not be elaborated here.

[0112] II: Based on the linear interpolation algorithm, perform conversion processing, outlier removal processing, and maximum and minimum value processing on the sample three-dimensional flow field data of each sample terrain, and extract the wind speed component of each sample terrain data point at the target height from the ground in the processed sample three-dimensional flow field data.

[0113] Here, perform data processing on the sample three-dimensional flow field data of each sample terrain, and extract the wind speed component of each sample terrain data point at the target height from the ground in the processed sample three-dimensional flow field data.

[0114] III: Calculate the actual wind speed modulus of each sample terrain based on the wind speed component of each sample terrain data point.

[0115] Here, calculate the actual wind speed modulus of each sample terrain according to the wind speed component of each sample terrain data point.

[0116] IV: Iteratively train the initial network model based on the mean square error and the mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map, and determine the wind resource prediction model.

[0117] Here, iteratively train the initial network model according to the mean square error and the mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map, and determine the wind resource prediction model.

[0118] In a possible implementation manner, the iteratively training the initial network model based on the mean square error and the mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map to determine the wind resource prediction model includes:

[0119] i: Detect whether the mean square error and the mean absolute error are less than or equal to a preset threshold.

[0120] Here, detect whether the mean square error and the mean absolute error are less than or equal to a preset threshold.

[0121] ii: If either is no, perform hyperparameter adjustment and network parameter adjustment on the initial network model, and continue to iteratively train the initial network model after parameter adjustment.

[0122] Here, if any of them is false, hyperparameter adjustment and network parameter adjustment are performed on the initial network model, and the iterative training of the initial network model after parameter adjustment is continued.

[0123] iii: If all are true, the initial network model is used as the wind resource prediction model.

[0124] In specific embodiments, terrain data and CFD wind resource simulation data are obtained, a data format conversion and normalization algorithm is configured for the obtained terrain data and CFD wind resource simulation data to ensure data quality, and a data storage and management system supporting large-scale data processing is built. A wind field barrier feature extraction algorithm is configured to expand the data feature dimension. The terrain height, wind field barrier, and roughness features are integrated. Based on the UNet network, a new type of initial network model is designed, and Transformer layers are embedded between the deep encoders to improve the model's feature learning ability. Sufficient historical data is prepared for model training and verification, and high-performance computing devices such as GPUs are configured to accelerate the training. Techniques such as mean square error, mean absolute error, and hyperparameter tuning are used to optimize the model's generalization performance and prediction accuracy, and a model regular update mechanism is established to adapt to data and environmental changes to obtain the final wind resource prediction model.

[0125] The wind resource prediction method for mesoscale terrain implemented in this application has the following specific advantages: By adopting advanced CFD technology and selecting representative terrain data across the country, high-precision simulation results are obtained, significantly improving data quality. Innovatively integrating the UNet and Transformer architectures enables second-level response and meets the real-time prediction requirements. High accuracy is maintained under complex terrains and variable wind directions. The wind resource prediction model has strong feature extraction capabilities, excellent generalization performance, can adapt to different regions and wind resource conditions, and has high robustness, ensuring stable and reliable prediction results, and the predicted wind resource result map can be intuitively displayed on the UI interface.

[0126] A wind resource prediction method for meso - micro scale terrain provided by an embodiment of the present application, the wind resource prediction method includes: extracting surface roughness data for a target terrain, and performing a computational fluid dynamics numerical simulation process on the meshed target terrain to determine three - dimensional flow field data of the target terrain; performing data pre - processing on the grid coordinates of the target terrain, the three - dimensional flow field data, and the surface roughness data to determine a fixed - dimension terrain elevation matrix and a fixed - dimension surface roughness matrix; calculating a blocking coefficient in the on - coming flow direction for the fixed - dimension terrain elevation matrix to determine the wind field blocking characteristics of the target terrain; inputting the fixed - dimension terrain elevation matrix, the wind field blocking characteristics, and the fixed - dimension surface roughness matrix into a wind resource prediction model for convolutional down - sampling, non - linear transformation, self - attention mechanism, convolutional up - sampling, and feature fusion to predict a wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a fixed - dimension flow field matrix at a target height as the training target, and the initial network model is a UNet architecture convolutional neural network model with multiple Transformer layers added. Through the wind resource prediction model, the wind resource results of any meso - micro scale terrain can be predicted quickly and accurately, while enhancing the adaptability and generality of the model, and reducing the dependence on a large amount of historical data and high - performance hardware.

[0127] Please refer to Figure 3 、 Figure 4 , Figure 3 which is one of the structural schematic diagrams of a wind resource prediction device for meso - micro scale terrain provided by an embodiment of the present application; Figure 4 which is the second structural schematic diagram of a wind resource prediction device for meso - micro scale terrain provided by an embodiment of the present application. As Figure 3 shown in

[0128] The wind resource prediction device 300 for meso - micro scale terrain includes:

[0129] An acquisition module 310, configured to extract surface roughness data for a target terrain, and perform a computational fluid dynamics numerical simulation process on the meshed target terrain to determine three - dimensional flow field data of the target terrain;

[0130] A pre - processing module 320, configured to perform data pre - processing on the grid coordinates of the target terrain, the three - dimensional flow field data, and the surface roughness data to determine a fixed - dimension terrain elevation matrix and a fixed - dimension surface roughness matrix;

[0131] The wind resource prediction module 340 is used to input the fixed-dimension terrain elevation matrix, the wind field barrier characteristics, and the fixed-dimension surface roughness matrix into a wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion, and predict the wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a target height fixed-dimension flow field matrix as the training target, and the initial network model is a UNet architecture convolutional neural network model with multiple Transformer layers added.

[0132] Further, when the preprocessing module 320 is used to perform data preprocessing on the grid coordinates of the target terrain, the three-dimensional flow field data, and the surface roughness data to determine the fixed-dimension terrain elevation matrix and the fixed-dimension surface roughness matrix, the preprocessing module 320 specifically is used for:

[0133] Perform linear interpolation on the underlying surface grid coordinates in the grid coordinates of the target terrain to a fixed-dimension two-dimensional terrain matrix, and normalize the original terrain height of each terrain data point based on the lowest height in the two-dimensional terrain matrix to determine the fixed-dimension terrain elevation matrix;

[0134] Based on the linear interpolation, perform conversion processing, outlier removal processing, and maximum-minimum normalization processing on the surface roughness data of each terrain data point to determine the fixed-dimension surface roughness matrix.

[0135] Further, when the wind field barrier feature construction module 330 is used to calculate the barrier coefficient in the oncoming flow direction for the fixed-dimension terrain elevation matrix to determine the wind field barrier characteristics of the target terrain, the wind field barrier feature construction module 330 specifically is used for:

[0136] For each terrain data point, starting from this terrain data point, linearly judge point by point along the reverse wind direction line to determine whether the terrain elevation data of the upwind terrain data points on the reverse wind direction line is greater than the terrain elevation data of this terrain data point;

[0137] If so, based on the height difference between the terrain elevation data of the upwind terrain data point and the terrain elevation data of the terrain data point and the straight-line distance between the upwind terrain data point and the terrain data point, determine the barrier coefficient of the upwind terrain data point to this terrain data point;

[0138] Accumulate the barrier coefficients of each terrain data point to determine the comprehensive barrier coefficient of each terrain data point;

[0139] Perform min-max normalization on multiple comprehensive barrier coefficients to determine the wind field barrier characteristics of the target terrain.

[0140] Further, the wind field barrier feature construction module 330 determines the barrier coefficient of the upwind terrain data point to the terrain data point through the following steps:

[0141]

[0142]

[0143] Among them, is the barrier coefficient of the terrain data point is the upwind terrain data point on the upwind straight line, The value range of is , z is the terrain elevation data, t is the height difference, L is the terrain length of the fixed-dimensional terrain elevation data, is the sign function.

[0144] Further, when the wind resource prediction module 340 is used to input the fixed-dimensional terrain elevation matrix, the wind field barrier characteristics, and the fixed-dimensional surface roughness matrix into the wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain, the wind resource prediction module 340 is specifically used for:

[0145] Based on multiple encoding layers of the wind resource prediction model, perform convolutional downsampling feature extraction on the fixed-dimensional terrain elevation matrix, the wind field barrier characteristics, and the fixed-dimensional surface roughness matrix in sequence to determine encoding feature maps of different scales;

[0146] Based on multiple Transformer layers, perform non-linear transformation and self-attention mechanism processing on the deep encoding feature map output by the deepest encoding layer to extract implicit features of the deep encoding feature map containing local information and global information;

[0147] Based on multiple decoding layers of the wind resource prediction model, perform convolutional upsampling processing on the implicit features and multiple encoding feature maps, and perform feature fusion on the decoding feature map output by the decoding layer and the corresponding encoding feature map in the decoding process through skip connection;

[0148] The convolutional layer based on the wind resource prediction model performs convolutional processing on the decoded feature map output by the last decoding layer to generate the wind resource result map; wherein, the wind speed modulus under each terrain data point is shown in the wind resource result map.

[0149] Further, as Figure 4 shown, the mesoscale terrain wind resource prediction device 300 further includes a model training module 350, and the model training module 350 determines the wind resource prediction model through the following steps:

[0150] Input the sample fixed-dimension terrain elevation matrix, sample wind field barrier characteristics, and sample fixed-dimension surface roughness matrix of multiple sample terrains into the initial network model to determine the predicted wind resource result map of each sample terrain;

[0151] Based on the linear interpolation algorithm, perform conversion processing, outlier removal processing, and maximum and minimum value processing on the sample three-dimensional flow field data of each sample terrain, and extract the wind speed component of each sample terrain data point at the target height from the ground;

[0152] Calculate the actual wind speed modulus of each sample terrain at the target height based on the wind speed component of each sample terrain data point;

[0153] Based on the mean square error and mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map, perform iterative training on the initial network model to determine the wind resource prediction model.

[0154] Further, when the model training module 350 is used to perform iterative training on the initial network model based on the mean square error and mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map to determine the wind resource prediction model, the model training module 350 is specifically used for:

[0155] Detect whether the mean square error and the mean absolute error are less than or equal to a preset threshold;

[0156] If either is no, perform hyperparameter adjustment and network parameter adjustment on the initial network model, and continue to perform iterative training on the initial network model after parameter adjustment;

[0157] If both are yes, use the initial network model as the wind resource prediction model.

[0158] A wind resource prediction device for meso - micro scale terrain provided by an embodiment of the present application, the wind resource prediction device includes: a collection module, configured to extract surface roughness data for a target terrain, and perform hydrodynamic numerical simulation processing on the target terrain after grid division to determine three - dimensional flow field data of the target terrain; a pre - processing module, configured to perform data pre - processing on the grid coordinates of the target terrain, the three - dimensional flow field data, and the surface roughness data to determine a fixed - dimension terrain elevation matrix and a fixed - dimension surface roughness matrix; a wind field barrier feature construction module, configured to calculate a barrier coefficient in the oncoming flow direction for the fixed - dimension terrain elevation matrix to determine the wind field barrier feature of the target terrain; a wind resource prediction module, configured to input the fixed - dimension terrain elevation matrix, the wind field barrier feature, and the fixed - dimension surface roughness matrix into a wind resource prediction model for convolutional down - sampling, non - linear transformation, self - attention mechanism, convolutional up - sampling, and feature fusion to predict a wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a target - height fixed - dimension flow field matrix as the training target, and the initial network model is a UNet - architecture convolutional neural network model with multiple Transformer layers added. Through the wind resource prediction model, the wind resource results of any meso - micro scale terrain can be predicted quickly and accurately, while enhancing the adaptability and generality of the model, and reducing the dependence on a large amount of historical data and high - performance hardware.

[0159] Please refer to Figure 5 , Figure 5 which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 5 shown in

[0160] the electronic device 500 includes a processor 510, a memory 520, and a bus 530. Figure 1 The memory 520 stores machine - readable instructions executable by the processor 510. When the electronic device 500 runs, the processor 510 communicates with the memory 520 through the bus 530. When the machine - readable instructions are executed by the processor 510, the steps of the wind resource prediction method for meso - micro scale terrain in the method embodiment as shown above can be executed. The specific implementation manner can refer to the method embodiment and will not be elaborated here.

[0161] An embodiment of the present application also provides a computer - readable storage medium. A computer program is stored on the computer - readable storage medium. When the computer program is run by a processor, the steps of the wind resource prediction method for meso - micro scale terrain in the method embodiment as shown above can be executed. The specific implementation manner can refer to the method embodiment and will not be elaborated here. Figure 1

[0162] ​Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0163] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some communication interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0164] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0165] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0166] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs that can store program codes.

[0167] Finally, it should be noted that the above-described embodiments are only specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, rather than limiting it. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any person skilled in the art within the technical scope disclosed by the present application can still modify the technical solutions described in the foregoing embodiments or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A wind resource prediction method for meso-scale terrain, characterized in that, The wind resource prediction method includes: Extracting surface roughness data for the target terrain, and performing hydrodynamic numerical simulation processing on the target terrain after grid division to determine the three-dimensional flow field data of the target terrain; Performing data preprocessing on the grid coordinates of the target terrain, the three-dimensional flow field data, and the surface roughness data to determine a fixed-dimension terrain elevation matrix and a fixed-dimension surface roughness matrix; Calculating the blocking coefficient in the oncoming flow direction for the fixed-dimension terrain elevation matrix to determine the wind field blocking characteristics of the target terrain; Inputting the fixed-dimension terrain elevation matrix, the wind field blocking characteristics, and the fixed-dimension surface roughness matrix into a wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a fixed-dimension flow field matrix at the target height as the training target, and the initial network model is a UNet architecture convolutional neural network model with multiple Transformer layers added; The calculating the blocking coefficient in the oncoming flow direction for the fixed-dimension terrain elevation matrix to determine the wind field blocking characteristics of the target terrain includes: For each terrain data point, starting from this terrain data point, linearly judging point by point along the reverse wind direction to determine whether the terrain elevation data of the upwind terrain data points on the reverse wind direction line is greater than the terrain elevation data of this terrain data point; If so, based on the height difference between the terrain elevation data of the upwind terrain data point and the terrain elevation data of the terrain data point, and the straight-line distance between the upwind terrain data point and the terrain data point, determining the blocking coefficient of the upwind terrain data point to the terrain data point; Performing an accumulation process on the blocking coefficients of each terrain data point to determine the comprehensive blocking coefficient of each terrain data point; Performing maximum-minimum normalization processing on multiple comprehensive blocking coefficients to determine the wind field blocking characteristics of the target terrain.

2. The wind resource prediction method according to claim 1, characterized in that The performing data preprocessing on the grid coordinates of the target terrain, the three-dimensional flow field data, and the surface roughness data to determine a fixed-dimension terrain elevation matrix and a fixed-dimension surface roughness matrix includes: Performing linear interpolation on the underlying surface grid coordinates in the grid coordinates of the target terrain to a fixed-dimension two-dimensional terrain matrix, and normalizing the original terrain height of each terrain data point based on the lowest height in the two-dimensional terrain matrix to determine the fixed-dimension terrain elevation matrix; Based on the linear interpolation, performing conversion processing, outlier removal processing, and maximum-minimum normalization processing on the surface roughness data of each terrain data point to determine the fixed-dimension surface roughness matrix.

3. The wind resource prediction method according to claim 1, wherein Determining the blocking coefficient of the upwind terrain data point to the terrain data point through the following steps: Among them, is the terrain data point of the barrier coefficient, is the upwind terrain data point on the upwind straight line, The value range of , z is the terrain elevation data, t is the height difference, L is the terrain length of the fixed-dimension terrain elevation data, is the sign function.

4. The wind resource prediction method according to claim 1, wherein, Inputting the fixed-dimension terrain elevation matrix, the wind field barrier characteristics, and the fixed-dimension surface roughness matrix into a wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion to predict the wind resource result map of the target terrain, including: Successively performing convolutional downsampling feature extraction on the fixed-dimension terrain elevation matrix, the wind field barrier characteristics, and the fixed-dimension surface roughness matrix based on multiple encoding layers of the wind resource prediction model to determine encoded feature maps of different scales; Performing non-linear transformation and self-attention mechanism processing on the deep encoded feature map output by the deepest encoding layer based on multiple Transformer layers to extract implicit features of the deep encoded feature map containing local information and global information; Performing convolutional upsampling processing on the implicit features and multiple encoded feature maps based on multiple decoding layers of the wind resource prediction model, and performing feature fusion on the decoded feature map output by the decoding layer and the corresponding encoded feature map in the decoding process through skip connection; Performing convolution on the decoded feature map output by the last decoding layer based on the convolution layer of the wind resource prediction model to generate the wind resource result map; wherein, the wind speed modulus under each terrain data point is displayed in the wind resource result map.

5. The wind resource prediction method according to claim 1, wherein Determine the wind resource prediction model through the following steps: Input the sample fixed-dimension terrain elevation matrix, sample wind field barrier characteristics, and sample fixed-dimension surface roughness matrix of multiple sample terrains into an initial network model to determine the predicted wind resource result map of each sample terrain; Based on the linear interpolation algorithm, perform conversion processing, outlier removal processing, and maximum and minimum value processing on the sample three-dimensional flow field data of each sample terrain, and extract the wind speed component of each sample terrain data point at a distance from the ground target height in the processed sample three-dimensional flow field data; Calculate the actual wind speed modulus of each sample terrain at the target height based on the wind speed component of each sample terrain data point; Iteratively train the initial network model based on the mean square error and mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map to determine the wind resource prediction model.

6. The wind resource prediction method according to claim 5, characterized in that, The iteratively training the initial network model based on the mean square error and mean absolute error between the actual wind speed modulus of each sample terrain and the wind speed modulus in the corresponding predicted wind resource result map to determine the wind resource prediction model includes: Detect whether the mean square error and the mean absolute error are less than or equal to a preset threshold; If either is no, perform hyperparameter adjustment and network parameter adjustment on the initial network model, and continue to iteratively train the initial network model after parameter adjustment; If both are yes, use the initial network model as the wind resource prediction model.

7. A wind resource prediction device for meso-scale terrain, characterized in that, The wind resource prediction device includes: A data acquisition module, configured to extract surface roughness data of a target terrain, and perform hydrodynamic numerical simulation processing on the target terrain after grid division to determine three-dimensional flow field data of the target terrain; A preprocessing module, configured to perform data preprocessing on the grid coordinates of the target terrain, the three-dimensional flow field data, and the surface roughness data to determine a fixed-dimension terrain elevation matrix and a fixed-dimension surface roughness matrix; A wind field blocking feature construction module, configured to calculate a blocking coefficient in the oncoming flow direction for the fixed-dimension terrain elevation matrix to determine the wind field blocking feature of the target terrain, specifically: For each terrain data point, starting from this terrain data point, linearly judge point by point along the reverse wind direction to determine whether the terrain elevation data of the upwind terrain data points on the reverse wind straight line is greater than the terrain elevation data of this terrain data point; If so, based on the height difference between the terrain elevation data of the upwind terrain data point and the terrain elevation data of the terrain data point, and the straight-line distance between the upwind terrain data point and the terrain data point, determine the blocking coefficient of the upwind terrain data point to this terrain data point; Accumulate the blocking coefficients of each terrain data point to determine the comprehensive blocking coefficient of each terrain data point; Perform maximum-minimum normalization processing on multiple comprehensive blocking coefficients to determine the wind field blocking feature of the target terrain; A wind resource prediction module, configured to input the fixed-dimension terrain elevation matrix, the wind field blocking feature, and the fixed-dimension surface roughness matrix into a wind resource prediction model for convolutional downsampling, non-linear transformation, self-attention mechanism, convolutional upsampling, and feature fusion to predict a wind resource result map of the target terrain; wherein, the wind resource prediction model is obtained by training an initial network model with a target height fixed-dimension flow field matrix as the training target, and the initial network model is a UNet architecture convolutional neural network model with multiple Transformer layers added.

8. An electronic device, characterized in that, Including: A processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are run by the processor, the steps of the wind resource prediction method for mesoscale terrain according to any one of claims 1 to 6 are executed.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is run by the processor, the steps of the wind resource prediction method for mesoscale terrain according to any one of claims 1 to 6 are executed.

Images