Method and equipment for determining wind power data based on geographical constraint attention

Abstract

The invention provides a wind power data determination method and device based on geographical constraint attention. The wind power data determination method comprises the steps of obtaining topographic data, surface vegetation coverage data and wind power data; for the wind power data, combining the topographic data and the surface vegetation data, performing feature fusion through an adaptive attention mechanism to obtain a fusion feature containing multi-source information; performing shallow feature extraction, deep feature extraction and image up-sampling on the fusion features in sequence, wherein the deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through a multi-stage residual fusion module RFSTB, a wind field frequency domain decoupling module CFB and a geographic perception global attention module GAB; verifying the effectiveness of the model by using a peak signal-to-noise ratio (PSNR) and a structural similarity index (SSIM); and performing super-resolution reconstruction on the wind power data based on the trained model to generate wind power data with predetermined precision. The method provides support for fine evaluation and efficient development of wind energy resources in complex terrain areas.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical fields of renewable energy development and meteorological data processing, and in particular to a method and device for determining wind data based on geographically constrained attention. Background Art

[0002] In the context of the global transition to clean energy, refined wind energy assessment is crucial. Current wind energy resource assessment faces multiple technical bottlenecks. Mainstream global reanalysis datasets (such as ERA5 and MERRA-2) have spatial resolutions exceeding 3 km, making it difficult to resolve the nonlinear modulation of wind fields by complex terrain. In areas such as rugged mountainous terrain, local wind speed forecast errors can reach 25%-30% due to the inability to capture phenomena such as steep slope streamline separation and low-level jet streams over inversion layers. Errors in capturing nighttime vertical wind speed gradients exceed 40%, leading to biased wind farm siting. For one mountain wind farm, the annual equivalent full-power hours were 18% lower than the design value. Mesoscale numerical models (such as WRF) can improve resolution to 1-3 km, but these models are computationally expensive and suffer from flawed parameterization schemes. They also under-represent the "narrow channel effect" of steep slopes, reproducing only 15%-20% of the measured wind speed increase. Furthermore, the fusion of high-resolution geographic information and meteorological data presents significant challenges: 1 km DEMs cannot quantify flow field distortion in areas of sudden slope changes, resulting in systematic underestimation of wind speed forecasts by 12%-15%; low-resolution land use data underestimates roughness length, leading to an error of 14.6% in estimated near-surface wind speeds; and high-resolution land cover data, lacking an effective fusion mechanism, exhibits spatiotemporal alignment errors exceeding three grid intervals with meteorological data. Traditional deep learning models lack physical constraints, and reconstruction results often violate the law of conservation of mass. In the fusion of multi-source observation data, the spatiotemporal asynchrony between lidar and satellite remote sensing makes it difficult for shallow networks to adaptively weight heterogeneous features, resulting in poor fusion efficiency. Therefore, developing a method and device for determining wind data based on geographic constraint attention, which can effectively overcome the shortcomings of these related technologies, has become a pressing technical challenge for the industry. Summary of the Invention

[0003] In view of the above problems existing in the prior art, an embodiment of the present invention provides a method and device for determining wind data based on geographical constraint attention.

[0004] In the first aspect, an embodiment of the present invention provides a method for determining wind data based on geographic constraint attention, including: acquiring terrain data, surface vegetation cover data and wind data; for wind data, combining terrain data and surface vegetation data, performing feature fusion through an adaptive attention mechanism to obtain fusion features containing multi-source information; performing shallow feature extraction, deep feature extraction and image upsampling on the fusion features in sequence, and the deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through a multi-level residual fusion module RFSTB, a wind field frequency domain decoupling module CFB and a geographic perception global attention module GAB; using the peak signal-to-noise ratio PSNR and the structural similarity index SSIM to verify the effectiveness of the model; performing super-resolution reconstruction on the wind data based on the trained model to generate wind data with predetermined accuracy.

[0005] Based on the content of the above method embodiment, the embodiment of the present invention provides a wind data determination method based on geographic constrained attention, wherein the deep feature extraction adopts the CFFormer super-resolution generation framework, and the CFFormer super-resolution generation framework includes a shallow feature extraction module, a deep feature extraction module and an image reconstruction module; the shallow feature extraction module performs preliminary feature extraction on the input low-resolution wind data image and digital elevation model DEM, wherein the DEM data generates a terrain gradient mask to constrain the shallow feature extraction process; the deep feature extraction module is composed of a multi-level multi-level residual fusion module RFSTB, the wind field frequency domain decoupling module CFB and the geographic perception global attention module GAB, and the image reconstruction module is composed of a convolutional layer and a PixelShuffle block.

[0006] Based on the content of the above method embodiment, the embodiment of the present invention provides a method for determining wind data based on geographically constrained attention. The RFSTB internally includes multiple STL layers and SSTL layers. The inter-layer feature flow is enhanced by the residual fusion operation. The output features of the x-th layer STL and the x+3-th layer SSTL are spliced and adjusted by 1×1 convolution as the input of the subsequent layer. The STL layer is a Swin Transformer layer; the SSTL layer is a Shift Swin Transformer layer; the residual fusion operation includes:

[0007] The residual fusion operation includes:

[0008] I CFB =Conv(Cat(I CA ,I FB ))

[0009] Among them, I CA is the feature map after channel attention CA processing; I FBis the feature map after Fourier branch processing; Conv is the convolution function; Cat is the concatenation function; I CFB It is the output of the wind farm frequency domain decoupling module CFB.

[0010] Based on the content of the above-mentioned method embodiment, the embodiment of the present invention provides a wind data determination method based on geographical constraint attention, wherein the wind field frequency domain decoupling module CFB includes channel attention and Fourier block. After the channel attention performs Conv-Act-Conv processing on the feature map, the importance weights of different channel features are assigned through the channel attention mechanism; the Fourier block converts the feature map to the frequency domain through fast Fourier transform, and after deep convolution and point convolution processing, it is converted back to the spatial domain through inverse Fourier transform to restore the high-frequency details. The output of the wind field frequency domain decoupling module CFB is the sum of the channel attention and Fourier block processing results.

[0011] Based on the content of the above method embodiments, the embodiment of the present invention provides a method for determining wind data based on geographically constrained attention. The geographically aware global attention module dynamically allocates attention weights to geographically sensitive areas: after the input feature map is processed by Conv-Act-Conv, three feature maps are obtained by reshaping three times. Attention weights are generated through transposition, multiplication, maximum pooling and expansion operations to determine terrain contour lines and land use types.

[0012] Based on the content of the above method embodiment, the method for determining wind data based on geographically constrained attention provided in the embodiment of the present invention, wherein the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM) are used to verify the effectiveness of the model, includes:

[0013]

[0014] Among them, PSNR is the peak signal-to-noise ratio; SSIM is the structural similarity index; x is the true value; y is the reconstructed value; log 10 is the logarithmic function with base 10; MSE is the mean square error; μ x is the pixel mean of image x; μ y is the pixel mean of image y; c1 and c2 are the first and second stability constants, respectively, to prevent the calculation from becoming unstable when the denominator is close to 0; σ xy is the covariance of image x and image y; σ x is the pixel variance of image x; σ y is the pixel variance of image y.

[0015] Based on the content of the above method embodiment, the embodiment of the present invention provides a method for determining wind data based on geographic constraint attention, wherein the terrain data includes: ASTERGDEM digital elevation data with a spatial resolution of 30 meters, land use type data with a resolution of 1 meter, and OpenStreetMap vector data.

[0016] In the second aspect, an embodiment of the present invention provides a wind data determination device based on geographic constraint attention, including: a first main module, used to acquire terrain data, surface vegetation cover data and wind data; a second main module, used to implement feature fusion for wind data, combined with terrain data and surface vegetation data, through an adaptive attention mechanism to obtain fusion features containing multi-source information; a third main module, used to implement shallow feature extraction, deep feature extraction and image upsampling of the fusion features in sequence, and the deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through a multi-level residual fusion module RFSTB, a wind field frequency domain decoupling module CFB and a geographic perception global attention module GAB; a fourth main module, used to implement the use of peak signal-to-noise ratio PSNR and structural similarity index SSIM to verify the effectiveness of the model; a fifth main module, used to implement super-resolution reconstruction of wind data based on the trained model to generate wind data with predetermined accuracy.

[0017] In a third aspect, an embodiment of the present invention provides an electronic device, including:

[0018] At least one processor, at least one memory and a communication interface; wherein,

[0019] The processor, memory and communication interface communicate with each other;

[0020] The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the wind data determination method based on geographical constraint attention provided by any one of the various implementations of the first aspect.

[0021] In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to execute the wind data determination method based on geographic constraint attention provided by any one of the various implementation methods of the first aspect.

[0022] The embodiment of the present invention provides a method and device for determining wind data based on geographically constrained attention. By constructing a wind field frequency domain decoupling block CFB, the method and device extract multi-scale periodic features to reconstruct a high-resolution meteorological field, significantly improving the accuracy of generating meteorological details such as typhoon eyewall structure, terrain, and rainfall distribution; dynamically allocating attention weights to geographically sensitive areas to forcibly embed prior knowledge such as terrain contour lines and land use types to ensure consistency between the meteorological field and the geographic space and avoid generating data that violates physical laws; achieving pixel-level alignment of multi-source data through jump connections and channel attention mechanisms to solve the problem of spatiotemporal alignment of heterogeneous data; combining staged residual learning with adaptive gradient clipping strategies to alleviate the gradient vanishing and explosion problems of deep networks and enhance the stability of model training; breaking through the bottlenecks in detail accuracy, physical rationality, and multi-source compatibility, and providing support for the refined assessment and efficient development of wind energy resources in complex terrain areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 A schematic flow chart of a method for determining wind data based on geographically constrained attention provided by an embodiment of the present invention;

[0025] Figure 2 A schematic structural diagram of a device for determining wind data based on geographically constrained attention provided by an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention;

[0027] Figure 4 A schematic diagram of the overall structure of CFFormer provided in an embodiment of the present invention;

[0028] Figure 5 Schematic diagram of RSFTB structure under different numbers of STLs provided by an embodiment of the present invention;

[0029] Figure 6 Schematic diagram of a wind farm frequency domain decoupling module CFB provided in an embodiment of the present invention;

[0030] Figure 7 A schematic diagram of the structure of the geographic awareness global attention module GAB provided in an embodiment of the present invention;

[0031] Figure 8 A schematic diagram of wind speed distribution at an accuracy of 200 meters in a predetermined area provided by an embodiment of the present invention;

[0032] Figure 9 A schematic diagram showing the comparison of the bicubic interpolation and SwinFIR at a 4x scale provided by an embodiment of the present invention;

[0033] Figure 10 A schematic diagram of the comparative effects of ablation experiments provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be combined with each other arbitrarily to form a feasible technical solution. This combination is not subject to the constraints of the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be considered that this combination of technical solutions does not exist and is not within the scope of protection required by the present invention. If there are step numbers in the following embodiments, they are only set for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0035] The embodiment of the present invention provides a method for determining wind data based on geographical constraint attention, see Figure 1 The method includes: acquiring terrain data, surface vegetation cover data and wind data; for the wind data, combining the terrain data and surface vegetation data, performing feature fusion through an adaptive attention mechanism to obtain fused features containing multi-source information; performing shallow feature extraction, deep feature extraction and image upsampling on the fused features in sequence, and the deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through a multi-level residual fusion module RFSTB, a wind field frequency domain decoupling module CFB and a geographic perception global attention module GAB; using the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM) to verify the effectiveness of the model; and performing super-resolution reconstruction on the wind data based on the trained model to generate wind data with predetermined accuracy.

[0036] The CFFormer super-resolution generation framework is proposed. This framework closely fits the physical characteristics of wind data and is designed to solve the three core problems of wind field reconstruction in complex terrain areas: terrain dynamic effect modeling, multi-source observation collaborative fusion and spectral feature decoupling. Figure 4As shown in the figure, CFFormer consists of a multi-level residual fusion SwinTransformer block (RFSTB), a wind field frequency domain decoupling block (CFB), and a geographic perception global attention block (GAB). CFFormer aims to more effectively adopt and fuse global information, process feature maps that are difficult to handle in the spatial domain, enhance feature reuse, and effectively solve problems such as gradient disappearance and explosion to improve RSSR performance. The overall structure of CFFormer is shown in the figure. Figure 1 As shown in the figure, CFormer consists of shallow feature extraction, deep feature extraction, and image upsampling. Specifically, the input low-resolution wind data image ILR and digital elevation model DEM pass through the shallow feature extraction part consisting of a convolutional layer, where the DEM data generates a terrain gradient mask to constrain the shallow feature extraction process. It then passes through the deep feature extraction part consisting of a series of residual fusion SwinTransformer blocks RFSTB, wind field frequency domain decoupling blocks CFB, and geographic awareness global attention blocks GAB at the head and tail of the blocks. Finally, it passes through the image reconstruction part consisting of a series of convolutional layers and PixelShuffle blocks:

[0037] I SR =F PixelShuffle (F DFE (Conv(I LR ))) (1)

[0038] FDFE represents the deep feature extraction function, FPixelShuffle represents the image upsampling function, Conv represents convolution, and ISR represents the final reconstructed image. In the deep feature extraction part, each RFSTB contains a series of STL, SSTL, and CFB to achieve deep extraction of feature maps.

[0039] Based on the content of the above method embodiment, as an optional embodiment, the wind data determination method based on geographic constrained attention provided in the embodiment of the present invention, the deep feature extraction adopts the CFFormer super-resolution generation framework, and the CFFormer super-resolution generation framework includes a shallow feature extraction module, a deep feature extraction module and an image reconstruction module; the shallow feature extraction module performs preliminary feature extraction on the input low-resolution wind data image and digital elevation model DEM, wherein the DEM data generates a terrain gradient mask to constrain the shallow feature extraction process; the deep feature extraction module is composed of a multi-level multi-level residual fusion module RFSTB, the wind field frequency domain decoupling module CFB and the geographic perception global attention module GAB, and the image reconstruction module is composed of a convolutional layer and a PixelShuffle block.

[0040] Each Swin Transformer Layer (STL) or Shift Swin Transformer Layer (SSTL) is mostly connected in series. This structure will cause the relationship between each layer to be too sparse, which will hinder the flow of features and easily cause the loss of local-global features of wind resource information images. Therefore, in order to solve this problem, it is proposed to construct a residual fusion Swin Transformer block (RFSTB) to enhance the connection between each layer, such as Figure 5 Specifically, the output of each STL is concatenated with the output of its second-closest SSTL, and then a 1x1 convolution is used to adjust the feature channels. The same feature fusion operation is then performed on each SSTL and its second-closest STL. This operation strengthens the connection between the features captured by the model at lower layers and the high-level abstract features captured at higher layers, optimizing the information flow, enhancing the expressiveness of the features, and improving the performance of the overall model.

[0041] Based on the content of the above method embodiment, as an optional embodiment, the embodiment of the present invention provides a method for determining wind data based on geographically constrained attention, wherein the RFSTB internally includes multiple STL layers and SSTL layers, and the inter-layer feature flow is enhanced by the residual fusion operation. The output features of the x-th layer STL and the x+3-th layer SSTL are spliced and adjusted by 1×1 convolution as the input of the subsequent layer; wherein the STL layer is a Swin Transformer layer; the SSTL layer is a Shift Swin Transformer layer; the residual fusion operation includes:

[0042] I CFB =Conv(Cat(I CA ,I FB )) (1.1)

[0043] Among them, I CA is the feature map after channel attention CA processing; I FB is the feature map after Fourier branch processing; Conv is the convolution function; Cat is the concatenation function; I CFB It is the output of the wind farm frequency domain decoupling module CFB.

[0044] like Figure 5As shown, taking 6 (S)STL as an example, 1).STL is fused with 4).SSTL, 2).STL is fused with 5).SSTL, and 3).STL is fused with 6).SSTL. Taking the first RFSTB as an example, after being processed by the geographic-aware global attention block GAB, IGAB is processed by the first layer 1).STL to obtain its output ISR1, which continues to serve as the input of the next STL until ISR4 is obtained:

[0045]

[0046] Where FSTL and FSSTL represent the STL and SSTL functions, respectively, and ISRx represents the feature map after the x-th layer of STL processing. Next, the resulting ISR1 and ISR4 are fused and used as the input to the fifth layer (5).STL. ISR2 and ISR5 are fused and used as the input to the sixth layer (6).SSTL. Finally, ISR3 and ISR6 are fused and used as the input to CFB. The output of CFB is added to IGAB to obtain the final output of RFSTB:

[0047]

[0048] Where ISRx,y represents the feature map after fusing the results of the xth and yth layers (S)STL, Conv(·) represents the convolution function, Cat(·) represents the concatenation function, and FCFB(·) represents the channel Fourier block function. By introducing fusion layers between different layers, the architecture effectively integrates information from different processing stages. This not only optimizes the feature flow process and enhances the network's ability to recognize complex patterns, but also significantly alleviates the problem of vanishing and exploding gradients, effectively enhancing RSSR performance.

[0049] In wind farm images, some important image information (such as wind speed detail branches, terrain boundaries, etc.) becomes blurred due to low resolution. By converting the feature map from the spatial domain to the frequency domain for processing through Fast Fourier Transform (FFT), these high-frequency details can be effectively restored, making the boundaries clearer and significantly improving the resolution and practicality of the image. Therefore, in order to enhance the detail recovery and overall quality of the image, the wind farm frequency domain decoupling block (CFB) is carefully designed. Figure 3 As shown in Figure 1, CFB consists of a channel attention part CA and a Fourier block part FB. The input feature map will be processed by the channel attention part and the Fourier block part respectively, and the results of the two branches will be fused as the output of CFB. Take the CFB in each RFSTB as an example:

[0050]

[0051] Where FCA(·) represents the channel attention function and FFB(·) represents the Fourier block function.

[0052] Channel Attention (CA): In order to enhance the performance of convolutional neural networks, the traditional channel attention mechanism is applied to CFFormer. The difference is that, Figure 6 As shown in part (a) of the figure (channel attention CA), a Conv-Act-Conv operation is performed at the beginning of the channel attention to extract features. The feature map processed by Conv-Act-Conv is element-wise multiplied with the feature map processed by channel attention, and then feature fusion is performed:

[0053]

[0054] Where Act(·) represents the activation function, and FCA'(·) represents the channel attention mechanism function. CA optimizes the model's feature representation by assigning different importance to features of different channels, allowing the network to focus more on features that are more useful for the current task.

[0055] Fourier block (FB): As mentioned above, in order to better restore image details and structures, the Fourier block FB is carefully designed in combination with the fast Fourier transform. Figure 6 As shown in part (b) (Fourier block FB), after the input feature map is processed by Conv-Act, it is transformed into the frequency domain space by Fourier transform FFT. In the frequency domain space, Conv-Act is used again to extract features in the deep layer, respectively through 1x1 point convolution and 3x3 depth convolution:

[0056]

[0057] Here, DConv and PConv represent depthwise convolution and pointwise convolution, respectively, and FFFT(·) represents the Fourier transform function. Pointwise convolution is used to mix information and transform features between channels. Although the convolution kernel is small, it integrates information from different channels to learn more complex feature representations. Depthwise convolution, on the other hand, applies the convolution kernel independently to each input channel. Unlike traditional convolutional layers, it does not mix information between channels, but instead processes the spatial features of each channel independently.

[0058] After feature fusion of the feature maps processed by depthwise convolution and pointwise convolution, they are then subjected to maximum pooling and average pooling, respectively. Next, the results of the maximum pooling and average pooling processes are concatenated, and further depthwise convolution is performed. Act-Conv is performed to adjust the channel dimension. After sigmoid processing, element-wise intelligent multiplication is performed with the feature map IFB4 before the pooling operation. The modified feature map is then converted from the frequency domain back to the spatial domain through the inverse Fourier transform (Inv FFT), and then added to the feature map IFB1 before the Fourier transform. The channel is adjusted through a 1x1 convolution:

[0059]

[0060] Where FMAX(·) represents the maximum pooling function, FAVG(·) represents the average pooling function, Sig(·) represents the sigmoid function, and FinvFFT(·) represents the inverse Fourier transform function. Then, the output of CA is fused with the output of FB to obtain the final CFB output ICFB:

[0061] I CFB =Conv(Cat(I CA ,I FB )) (8)

[0062] Based on the content of the above method embodiment, as an optional embodiment, the wind data determination method based on geographical constraint attention provided in the embodiment of the present invention, the wind field frequency domain decoupling module CFB includes channel attention and Fourier block, the channel attention performs Conv-Act-Conv processing on the feature map, and then assigns importance weights of different channel features through the channel attention mechanism; the Fourier block converts the feature map to the frequency domain through fast Fourier transform, and after deep convolution and point convolution processing, it is converted back to the spatial domain through inverse Fourier transform to restore high-frequency details. The output of the wind field frequency domain decoupling module CFB is the sum of the channel attention and Fourier block processing results.

[0063] Although most previous SR methods have achieved remarkable results, they still have great deficiencies in extracting more distant information and are still very limited in processing long-distance information. In addition, it is determined that the rich channel dimension contains more features, and most previous literature has not fully utilized these features. Therefore, in order to solve these problems, a geographically aware global attention block (GAB) is proposed, which is placed at the head and tail of deep feature extraction, such as Figure 4 As shown in the figure, the head GAB is used to extract richer features for LR, and the tail GAB is used together with the head GAB to better capture information of longer distances and channel dimensions.

[0064] Based on the content of the above method embodiment, as an optional embodiment, the embodiment of the present invention provides a wind data determination method based on geographically constrained attention, and the geographically aware global attention module dynamically allocates attention weights to geographically sensitive areas: after the input feature map is processed by Conv-Act-Conv, three feature maps are obtained by three reshape operations, and attention weights are generated through transposition, multiplication, maximum pooling and expansion operations to determine terrain contour lines and land use types.

[0065] The proposed geo-aware global attention block is as follows: Figure 7 As shown, the input feature map undergoes Conv-Act-Conv extraction and is reshaped three times to obtain three feature maps: IGA1, IGA2, and IGA3. IGA1 is transposed and multiplied with IGA2 to obtain I'GA12. After max pooling and expansion, it is subtracted from I'GA12 and normalized and highlighted using a sigmoid operation. The normalized result is multiplied by IGA3 and then added to the features before the initial transposition. Finally, feature fusion is performed to obtain the final GAB output, IGAB. Take the GAB of the deep feature extraction head as an example:

[0066]

[0067] FReshape and Ftranspose represent the reshape function and transpose function respectively, and MAX represents the maximum pooling and expansion operation. This operation enables the model to better focus on key features instead of being treated equally by all activation values.

[0068] Based on the content of the above method embodiment, as an optional embodiment, the method for determining wind data based on geographically constrained attention provided in the embodiment of the present invention, wherein the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM) are used to verify the effectiveness of the model, includes:

[0069]

[0070] Among them, PSNR is the peak signal-to-noise ratio; SSIM is the structural similarity index; x is the true value; y is the reconstructed value; log 10 is the logarithmic function with base 10; MSE is the mean square error; μ x is the pixel mean of image x; μ y is the pixel mean of image y; c1 and c2 are the first and second stability constants, respectively, to prevent the calculation from becoming unstable when the denominator is close to 0; σ xy is the covariance of image x and image y; σ xis the pixel variance of image x; σ y is the pixel variance of image y.

[0071] A total of 26 ASTER GDEM digital elevation data sets with a spatial resolution of 30 meters were collected for the designated areas. This data will serve as the basis for subsequent analysis, helping to more accurately assess the impact of topography on the wind farm's power generation potential. Furthermore, the plan is to use the digital elevation model data to derive key information such as slope and aspect, which are crucial for generating wind maps and conducting in-depth analysis of wind farm site selection.

[0072] We also utilized 1-meter-resolution land use type data for the designated areas, generated using a deep learning framework and open data. This data, totaling 11 images and 6.7GB of data, detailedly categorizes various land use types, including forest, shrubland, grassland, cultivated land, buildings, transportation routes, bare land, snow and ice, water bodies, wetlands, and tundra, providing rich and accurate information for wind turbine site selection. The land use type classification is shown in Table 1.

[0073] Table 1

[0074]

[0075]

[0076] In addition, vector data layers covering 16 predefined regions were used, provided by Geofabrik. Geofabrik provides OpenStreetMap (OSM) data for global regions, covering a wide range of landform types, including natural features (such as mountains, rivers, lakes, and forests), urban features (such as buildings, roads, bridges, and parks), agricultural features (such as farmland and orchards), and transportation features (such as railways, highways, and aviation facilities). This detailed geospatial information provides rich and accurate foundational data support for wind power site selection analysis.

[0077] The wind speed and wind density at 10m, 50m, 100m, 150m, and 200m in the predetermined area are also collected, and the example visualization is as follows Figure 8 As shown in the figure, the wind speed in the predetermined area is mainly used, supplemented by the longitude and latitude information corresponding to the wind speed and wind density data in the predetermined area clipped from the above-mentioned spatial resolution digital elevation data, land use type numbers, and vector data layers. The proposed CFFormer structure is used for super-resolution.

[0078] The experimental environment used an RTX 4090D for training, using the Adam (Adaptive Momentum) optimizer, with an initial learning rate η set to 0.0002 and an L1 loss function. The method was comprehensively evaluated on wind and meteorological data using five metrics: PSNR, SSIM, SAM, QI, and SCC.

[0079] Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM): Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) are the most commonly used metrics for measuring image reconstruction quality. PSNR evaluates image quality by comparing the pixel differences between the original image and the processed image. It is expressed in decibels (dB). A higher value indicates less image distortion and better image quality. SSIM measures the similarity between two images based on brightness, contrast, and structure. Unlike PSNR, SSIM takes into account the characteristics of the human visual system and places greater emphasis on the structural information of the image content. PSNR and SSIM are shown in Equations (9) and (10).

[0080] Based on the content of the above method embodiment, as an optional embodiment, the embodiment of the present invention provides a method for determining wind data based on geographic constraint attention, wherein the terrain data includes: ASTERGDEM digital elevation data with a spatial resolution of 30 meters, land use type data with a resolution of 1 meter, and OpenStreetMap vector data.

[0081] The SAM metric is used to measure the angular similarity between two pixels, regardless of their absolute intensity. The lower the value, the smaller the angular difference between the spectral vectors of the two pixels. It is widely used to effectively measure the similarity of spectra between different pixels. The calculation formula of SAM includes:

[0082]

[0083] Where a and b represent two spectral vectors, and n is the number of bands.

[0084] Universal Quality Index (UQI): The UQI is used to assess the similarity between two images. It takes into account the mean, variance, and covariance of the images. Its range is [-1, 1]. Values closer to 1 indicate greater similarity between the two images. A moving window is used to traverse the image, and the UQI value of each window in the image is calculated and averaged.

[0085] Spatial Correlation Coefficient (SCC): SCC is an indicator that measures the spatial correlation between two images and is used to evaluate spatial structural similarity. A high spatial correlation coefficient indicates a high similarity in the spatial distribution and structural features between the images. The calculation formula for SCC includes:

[0086]

[0087] Where X and Y represent the pixel values of the two images respectively, and N is the total number of pixels. and Represents the average pixel value of X and Y respectively. Figure 9 To determine the effectiveness of the proposed method, we compared it with bicubic interpolation and SwinFIR. Bicubic interpolation resulted in significant blurring and loss of detail, while SwinFIR still struggled to extract some details. SwinFIR achieved the best performance. To verify the effectiveness of each module in the proposed method, we conducted ablation experiments using a 200-meter wind speed map of a designated area, as shown in Figure 10. The model achieved the best performance when only CFB and GAB were present.

[0088] The embodiment of the present invention provides a wind data determination method based on geographically constrained attention. By constructing a wind field frequency domain decoupling block CFB, the method extracts multi-scale periodic features and reconstructs a high-resolution meteorological field, significantly improving the accuracy of generating meteorological details such as typhoon eyewall structure, terrain and rainfall distribution; dynamically allocating attention weights to geographically sensitive areas to forcibly embed prior knowledge such as terrain contour lines and land use types to ensure consistency between the meteorological field and the geographic space and avoid generating data that violates physical laws; through skip connections and channel attention mechanisms, pixel-level alignment of multi-source data is achieved to solve the problem of spatiotemporal alignment of heterogeneous data; combining staged residual learning with adaptive gradient clipping strategies to alleviate the gradient vanishing and explosion problems of deep networks and enhance the stability of model training; breaking through the bottlenecks in detail accuracy, physical rationality and multi-source compatibility, and providing support for the refined assessment and efficient development of wind energy resources in complex terrain areas.

[0089] The implementation basis of each embodiment of the present invention is to implement programmed processing through a device with processor functions. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the present invention can be encapsulated into various modules. Based on this reality, on the basis of the above embodiments, an embodiment of the present invention provides a device for determining wind data based on geographical constraint attention, which is used to execute the method for determining wind data based on geographical constraint attention in the above method embodiment. Figure 2The device includes: a first main module for acquiring terrain data, surface vegetation cover data and wind data; a second main module for implementing feature fusion for wind data by combining terrain data and surface vegetation data through an adaptive attention mechanism to obtain fusion features containing multi-source information; a third main module for sequentially performing shallow feature extraction, deep feature extraction and image upsampling on the fusion features, wherein the deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through a multi-level residual fusion module RFSTB, a wind field frequency domain decoupling module CFB and a geographic perception global attention module GAB; a fourth main module for verifying the effectiveness of the model using the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM); and a fifth main module for implementing super-resolution reconstruction of wind data based on the trained model to generate wind data with predetermined accuracy.

[0090] The embodiment of the present invention provides a device for determining wind data based on geographical constraint attention, which adopts Figure 2 Several modules in it, by constructing the wind field frequency domain decoupling block CFB, extracting multi-scale periodic features to reconstruct high-resolution meteorological fields, significantly improving the accuracy of generating meteorological details such as typhoon eyewall structure, topography and rainfall distribution; dynamically allocating attention weights to geographically sensitive areas, forcibly embedding prior knowledge such as terrain contours and land use types to ensure the consistency between meteorological fields and geographic space, and avoid generating data that violates physical laws; through skip connections and channel attention mechanisms, pixel-level alignment of multi-source data is achieved to solve the problem of spatiotemporal alignment of heterogeneous data; combining staged residual learning with adaptive gradient clipping strategy to alleviate the gradient vanishing and explosion problems of deep networks and enhance model training stability; breaking through the bottlenecks in detail accuracy, physical rationality and multi-source compatibility, and providing support for the refined assessment and efficient development of wind energy resources in complex terrain areas.

[0091] It should be noted that the device in the device embodiment provided by the present invention can be used to implement the method in the above-mentioned method embodiment as well as the method in other method embodiments provided by the present invention. The only difference is that the corresponding functional modules are set. The principle is basically the same as the principle of the above-mentioned device embodiment provided by the present invention. As long as those skilled in the art refer to the specific technical solutions in other method embodiments on the basis of the above-mentioned device embodiment, obtain the corresponding technical means and the technical solutions composed of these technical means by combining technical features, and ensure the practicality of the technical solutions, they can improve the device in the above-mentioned device embodiment to obtain the corresponding device class embodiment, thereby obtaining the corresponding device class embodiment for implementing the methods in other method class embodiments. For example:

[0092] Based on the content of the above-mentioned device embodiment, as an optional embodiment, the wind data determination device based on geographic constrained attention provided in the embodiment of the present invention also includes: a first sub-module, used to realize the deep feature extraction using the CFFormer super-resolution generation framework, the CFFormer super-resolution generation framework includes a shallow feature extraction module, a deep feature extraction module and an image reconstruction module; the shallow feature extraction module performs preliminary feature extraction on the input low-resolution wind data image and digital elevation model DEM, wherein the DEM data generates a terrain gradient mask to constrain the shallow feature extraction process; the deep feature extraction module is composed of a multi-level multi-level residual fusion module RFSTB, the wind field frequency domain decoupling module CFB and the geographic perception global attention module GAB, and the image reconstruction module is composed of a convolutional layer and a PixelShuffle block.

[0093] Based on the content of the above device embodiment, as an optional embodiment, the wind data determination device based on geographical constraint attention provided in the embodiment of the present invention further includes: a second submodule, which is used to realize that the RFSTB contains multiple STL layers and SSTL layers, enhances the inter-layer feature flow through the residual fusion operation, splices the output features of the x-th layer STL and the x+3-th layer SSTL, and uses them as the input of the subsequent layer after 1×1 convolution adjustment; wherein the STL layer is a SwinTransformer layer; the SSTL layer is a Shift Swin Transformer layer; the residual fusion operation includes:

[0094] The residual fusion operation includes:

[0095] I CFB =Conv(Cat(I CA ,I FB ))

[0096] Among them, I CA is the feature map after channel attention CA processing; I FB is the feature map after Fourier branch processing; Conv is the convolution function; Cat is the concatenation function; I CFB It is the output of the wind farm frequency domain decoupling module CFB.

[0097] Based on the content of the above-mentioned device embodiment, as an optional embodiment, the wind data determination device based on geographical constraint attention provided in the embodiment of the present invention further includes: a third sub-module, used to realize that the wind field frequency domain decoupling module CFB includes channel attention and Fourier block, and the channel attention assigns importance weights of different channel features through the channel attention mechanism after performing Conv-Act-Conv processing on the feature map; the Fourier block converts the feature map to the frequency domain through fast Fourier transform, and after depth convolution and point convolution processing, it is converted back to the spatial domain through inverse Fourier transform to restore high-frequency details. The output of the wind field frequency domain decoupling module CFB is the sum of the channel attention and Fourier block processing results.

[0098] Based on the content of the above-mentioned device embodiment, as an optional embodiment, the wind data determination device based on geographically constrained attention provided in the embodiment of the present invention further includes: a fourth sub-module, used to realize the dynamic allocation of attention weights of geographically sensitive areas by the geographically aware global attention module: after the input feature map is processed by Conv-Act-Conv, three feature maps are obtained by reshaping three times, and attention weights are generated through transposition, multiplication, maximum pooling and expansion operations to determine the terrain contour lines and land use types.

[0099] Based on the content of the above device embodiment, as an optional embodiment, the device for determining wind data based on geographically constrained attention provided in the embodiment of the present invention further includes: a fifth submodule for implementing the verification of model validity using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), including:

[0100]

[0101]

[0102] Among them, PSNR is the peak signal-to-noise ratio; SSIM is the structural similarity index; x is the true value; y is the reconstructed value; log 10 is the logarithmic function with base 10; MSE is the mean square error; μ x is the pixel mean of image x; μ y is the pixel mean of image y; c1 and c2 are the first and second stability constants, respectively, to prevent the calculation from becoming unstable when the denominator is close to 0; σ xy is the covariance of image x and image y; σ x is the pixel variance of image x; σ y is the pixel variance of image y.

[0103] Based on the content of the above-mentioned device embodiment, as an optional embodiment, the wind data determination device based on geographic constraint attention provided in the embodiment of the present invention further includes: a sixth sub-module, used to realize that the terrain data includes: ASTER GDEM digital elevation data with a spatial resolution of 30 meters, land use type data with a resolution of 1 meter, and OpenStreetMap vector data.

[0104] The method of the embodiment of the present invention is implemented by electronic devices, so it is necessary to introduce the relevant electronic devices. Based on this purpose, the embodiment of the present invention provides an electronic device, such as Figure 3 As shown, the electronic device includes: at least one processor, a communications interface, at least one memory, and a communications bus, wherein the at least one processor, the communications interface, and the at least one memory communicate with each other via the communications bus. The at least one processor can call logic instructions in the at least one memory to execute all or part of the steps of the methods provided in the aforementioned method embodiments.

[0105] In addition, the logic instructions in the at least one memory mentioned above can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each method embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0106] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0107] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0108] The flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. Based on this understanding, each box in the flowchart or block diagram can represent a module, program segment or part of the code, and the module, program segment or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, or sometimes in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0109] It should be noted that the terms "include", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include..." do not exclude the presence of other identical elements in the process, method, article or device that includes the elements. Any "predetermined threshold", "preset threshold" or similar expressions that do not indicate a specific value can be determined by a person of ordinary skill in the art through simple experiments or corresponding debugging.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for determining wind data based on geographically constrained attention, characterized in that: include: Acquire terrain data, surface vegetation cover data and wind data; for wind data, combine terrain data and surface vegetation data, and perform feature fusion through an adaptive attention mechanism to obtain fused features containing multi-source information; perform shallow feature extraction, deep feature extraction and image upsampling on the fused features in sequence; the deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through a multi-level residual fusion module RFSTB, a wind field frequency domain decoupling module CFB and a geographic perception global attention module GAB; use the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM) to verify the effectiveness of the model; perform super-resolution reconstruction of the wind data based on the trained model to generate wind data with predetermined accuracy.

2. The method for determining wind data based on geographically constrained attention according to claim 1, characterized in that: The deep feature extraction adopts the CFFormer super-resolution generation framework, which includes a shallow feature extraction module, a deep feature extraction module and an image reconstruction module; the shallow feature extraction module performs preliminary feature extraction on the input low-resolution wind data image and digital elevation model DEM, wherein the DEM data generates a terrain gradient mask to constrain the shallow feature extraction process; the deep feature extraction module is composed of a multi-level multi-level residual fusion module RFSTB, the wind field frequency domain decoupling module CFB and the geographic perception global attention module GAB, and the image reconstruction module is composed of a convolutional layer and a PixelShuffle block.

3. The method for determining wind data based on geographically constrained attention according to claim 2, characterized in that: The RFSTB internally contains multiple STL layers and SSTL layers. The residual fusion operation is used to enhance the inter-layer feature flow. The output features of the x-th layer STL and the x+3-th layer SSTL are spliced and adjusted by 1×1 convolution as the input of the subsequent layer. Among them, the STL layer is the SwinTransformer layer; the SSTL layer is the Shift Swin Transformer layer. The residual fusion operation includes: The residual fusion operation includes: I CFB =Conv(Cat(I CA ,I FB )) Among them, I CA is the feature map after channel attention CA processing; I FB is the feature map after Fourier branch processing; Conv is the convolution function; Cat is the concatenation function; I CFB It is the output of the wind farm frequency domain decoupling module CFB.

4. The method for determining wind data based on geographically constrained attention according to claim 3, characterized in that: The wind farm frequency domain decoupling module CFB includes channel attention and Fourier block. After the channel attention performs Conv-Act-Conv processing on the feature map, it assigns importance weights to different channel features through the channel attention mechanism; the Fourier block converts the feature map to the frequency domain through fast Fourier transform, and after depth convolution and point convolution processing, it converts it back to the spatial domain through inverse Fourier transform to restore high-frequency details. The output of the wind farm frequency domain decoupling module CFB is the sum of the channel attention and Fourier block processing results.

5. The method for determining wind data based on geographically constrained attention according to claim 4, characterized in that: The geo-aware global attention module dynamically assigns attention weights to geo-sensitive areas: after the input feature map undergoes Conv-Act-Conv processing, it is reshaped three times to obtain three feature maps. Attention weights are generated through transposition, multiplication, maximum pooling and expansion operations to determine terrain contours and land use types.

6. The method for determining wind data based on geographically constrained attention according to claim 5, characterized in that: The use of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) to verify the effectiveness of the model includes: Among them, PSNR is the peak signal-to-noise ratio; SSIM is the structural similarity index; x is the true value; y is the reconstructed value; log 10 is the logarithmic function with base 10; MSE is the mean square error; μ x is the pixel mean of image x; μ y is the pixel mean of image y; c1 and c2 are the first and second stability constants, respectively, to prevent the calculation from becoming unstable when the denominator is close to 0; σ xy is the covariance of image x and image y; σ x is the pixel variance of image x; σ y is the pixel variance of image y.

7. The method for determining wind data based on geographically constrained attention according to claim 6, characterized in that: The terrain data includes: ASTERGDEM digital elevation data with a spatial resolution of 30 meters, land use type data with a resolution of 1 meter, and OpenStreetMap vector data.

8. A device for determining wind data based on geographically constrained attention, characterized in that: include: The first main module is used to obtain terrain data, surface vegetation cover data and wind data; The second main module is used to implement feature fusion of wind data by combining terrain data and surface vegetation data through an adaptive attention mechanism to obtain fused features containing multi-source information; the third main module is used to implement shallow feature extraction, deep feature extraction and image upsampling of the fused features in sequence. The deep feature extraction realizes multi-scale feature extraction and geographic constraint fusion through the multi-level residual fusion module RFSTB, the wind field frequency domain decoupling module CFB and the geographic perception global attention module GAB; the fourth main module is used to verify the effectiveness of the model using the peak signal-to-noise ratio PSNR and the structural similarity index SSIM; the fifth main module is used to implement super-resolution reconstruction of wind data based on the trained model to generate wind data with predetermined accuracy.

9. An electronic device, characterized in that: include: At least one processor, at least one memory and a communication interface; wherein, The processor, memory and communication interface communicate with each other; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, which cause a computer to execute the method of any one of claims 1 to 7.

Images