Wind resource assessment method and device and storage medium

Abstract

The invention provides a wind resource assessment method and device and a storage medium. According to the method, feature extraction is carried out on a first wind resource map and a second wind resource map to obtain a first type of features, and the first wind resource map and the second wind resource map are wind resource maps of different spatial resolutions for the same wind power plant; performing image decomposition processing on the first wind resource map and the second wind resource map to obtain a second type of features of the first wind resource map and a second type of features of the second wind resource map; performing feature fusion processing on the first type of features, the second type of features of the first wind resource atlas and the second type of features of the second wind resource atlas to obtain a fusion atlas; and evaluating the wind resources of the wind power plant based on the fusion atlas. The maps with different spatial resolutions are fused into the map with higher precision, so that the accuracy of a wind resource evaluation result is improved, and the effect of deploying the wind power plant is improved.

Description

Technical Field

[0001] The present application relates to the field of wind power generation, and particularly relates to a wind resource assessment method, device, and storage medium. Background Art

[0002] Wind resource assessment refers to detecting wind resources through certain technical means to determine the wind resource density, distribution, and availability in a specific area. Wind resource assessment provides accurate data basis for wind power development. For example, wind resource assessment can provide important basis for the site selection of wind farms. Based on the wind resources evaluated in an area, it helps to determine whether the area is suitable for building a wind farm and the expected power generation of the wind farm.

[0003] The wind resource map is the data basis for wind resource assessment. The wind resource map draws a geographic information map representing the spatial distribution characteristics of wind resources through spatial interpolation and statistical analysis of wind speed and wind direction data. The wind resource map provides the spatial distribution information of wind resources, including potential wind power development areas, wind resource density in different areas, etc., and can intuitively reflect the utilization potential of wind resources. The wind resource map is one of the basic data for wind resource assessment. The wind resource map provides spatial distribution and time variation data on wind resource characteristics such as wind speed and wind direction, and these data are crucial for wind resource assessment.

[0004] Currently, the accuracy of many wind resource maps is poor, resulting in poor accuracy of the wind resource assessment results. And the poor accuracy of the wind resource assessment results will also affect the effect of deploying wind farms. Summary of the Invention

[0005] The present application provides a wind resource assessment method, device, and storage medium. Since maps with different spatial resolutions are fused into a map with higher accuracy, the accuracy of the wind resource assessment results is improved, and further the effect of deploying wind farms is improved. The technical solutions are as follows.

[0006] In a first aspect, a wind resource assessment method is provided. The method includes:

[0007] Performing feature extraction on a first wind resource map and a second wind resource map to obtain a first type of features. The first wind resource map and the second wind resource map are wind resource maps with different spatial resolutions for the same wind farm;

[0008] Performing image decomposition processing on the first wind resource map and the second wind resource map respectively to obtain a second type of features of the first wind resource map and a second type of features of the second wind resource map;

[0009] Perform feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map;

[0010] Evaluate the wind resources of the wind farm based on the fusion map.

[0011] In some embodiments, the first wind resource map and the second wind resource map include the same wind resource parameters, and / or, the first wind resource map and the second wind resource map are wind resource maps at the same wind speed height.

[0012] In some embodiments, the feature extraction of the first wind resource map and the second wind resource map to obtain the first type of features includes:

[0013] Input the first wind resource map and the second wind resource map into multiple convolutional layers in a deep learning model to obtain the feature map of the first wind resource map and the feature map of the second wind resource map;

[0014] Based on the feature map, obtain the first type of features through the activation function of the deep learning model, and the first type of features are used to characterize the fusion weights of the first wind resource map and the second wind resource map.

[0015] In some embodiments, the second type of features includes residual images. The image decomposition processing of the first wind resource map and the second wind resource map respectively to obtain the second type of features of the first wind resource map and the second type of features of the second wind resource map includes:

[0016] Decompose the first wind resource map and the second wind resource map respectively by using a pyramid to obtain N first pyramid maps with different scales corresponding to the first wind resource map and N second pyramid maps with different scales corresponding to the second wind resource map, where N is the number of layers of the pyramid and N≥2.

[0017] For the N first pyramid maps with different scales, perform upsampling processing on the N-1 layer first pyramid map and perform differential processing with the Nth layer first pyramid map to obtain the residual image corresponding to the first wind resource map;

[0018] For the N second pyramid maps with different scales, perform upsampling processing on the N-1 layer second pyramid map and perform differential processing with the Nth layer second pyramid map to obtain the residual image corresponding to the second wind resource map.

[0019] In some embodiments, the feature fusion process for the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map includes:

[0020] Determine the map fusion mode parameters based on the second type of features of the first wind resource map and the second type of features of the second wind resource map;

[0021] Perform a fusion process on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map using the fusion strategy corresponding to the map fusion mode parameters to obtain a fusion map.

[0022] In some embodiments, the performing a fusion process on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map using the fusion strategy corresponding to the map fusion mode parameters to obtain a fusion map includes:

[0023] In response to the map fusion mode parameters belonging to the first threshold range, perform a weighted sum on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map based on the weight of the first wind resource map and the weight of the second wind resource map to obtain a fusion map;

[0024] In response to the map fusion mode parameters belonging to the second threshold range, perform an inverse transformation on the second type of features of the first wind resource map to obtain a fusion map;

[0025] In response to the map fusion mode parameters belonging to the third threshold range, perform an inverse transformation on the second type of features of the second wind resource map to obtain a fusion map.

[0026] In some embodiments, the weight of the first wind resource map is determined based on the spatial resolution of the first wind resource map, the weight of the second wind resource map is determined based on the spatial resolution of the second wind resource map, and the weight of the wind resource map is positively correlated with the spatial resolution of the wind resource map.

[0027] In some embodiments, after the feature fusion process for the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map, the method further includes:

[0028] Determine the accuracy of the fusion map based on the deviation between the wind resource parameters in the fusion map and the wind resource parameters detected at the wind measurement points.

[0029] In some embodiments, after performing feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map, the method further includes:

[0030] Based on the deviation between the wind resource parameters in the fusion map and the wind resource parameters in the first wind resource map, and the deviation between the wind resource parameters in the fusion map and the wind resource parameters in the second wind resource map, determine the accuracy of the fusion map.

[0031] In some embodiments, the performing feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map includes:

[0032] Fuse the first type of features, the second type of features of the first wind resource map, the second type of features of the second wind resource map, and the wind resource parameters detected at the wind measurement points to obtain fusion features.

[0033] In a second aspect, a wind resource evaluation device is provided, and the device includes:

[0034] A feature extraction unit, configured to perform feature extraction on a first wind resource map and a second wind resource map to obtain a first type of features, where the first wind resource map and the second wind resource map are wind resource maps with different spatial resolutions for the same wind farm;

[0035] An image decomposition unit, configured to perform image decomposition processing on the first wind resource map and the second wind resource map respectively to obtain the second type of features of the first wind resource map and the second type of features of the second wind resource map;

[0036] A feature fusion unit, configured to perform feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map;

[0037] An evaluation unit, configured to evaluate the wind resource of the wind farm based on the fusion map.

[0038] In some embodiments, the first wind resource map and the second wind resource map include the same wind resource parameters, and / or the first wind resource map and the second wind resource map are wind resource maps at the same wind speed height.

[0039] In some embodiments, the feature extraction unit is configured to input the first wind resource map and the second wind resource map into multiple convolutional layers in a deep learning model to obtain a feature map of the first wind resource map and a feature map of the second wind resource map; and based on the feature maps, obtain the first type of features through an activation function of the deep learning model, where the first type of features is used to characterize the fusion weights of the first wind resource map and the second wind resource map.

[0040] In some embodiments, the second type of features includes residual images. The image decomposition unit is configured to perform decomposition processing on the first wind resource map and the second wind resource map respectively by using a pyramid to obtain N first pyramid maps with different scales corresponding to the first wind resource map and N second pyramid maps with different scales corresponding to the second wind resource map, where N is the number of layers of the pyramid and N≥2. For the N first pyramid maps with different scales, perform upsampling processing on the N - 1 layer first pyramid maps and perform differential processing with the Nth layer first pyramid maps to obtain the residual image corresponding to the first wind resource map; for the N second pyramid maps with different scales, perform upsampling processing on the N - 1 layer second pyramid maps and perform differential processing with the Nth layer second pyramid maps to obtain the residual image corresponding to the second wind resource map.

[0041] In some embodiments, the feature fusion unit is configured to determine a map fusion mode parameter based on the second type of features of the first wind resource map and the second type of features of the second wind resource map; and perform fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map by using a fusion strategy corresponding to the map fusion mode parameter to obtain a fused map.

[0042] In some embodiments, the feature fusion unit is configured to, in response to the map fusion mode parameter belonging to a first threshold range, perform weighted summation on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map based on the weight of the first wind resource map and the weight of the second wind resource map to obtain a fused map; in response to the map fusion mode parameter belonging to a second threshold range, perform inverse transformation on the second type of features of the first wind resource map to obtain a fused map; and in response to the map fusion mode parameter belonging to a third threshold range, perform inverse transformation on the second type of features of the second wind resource map to obtain a fused map.

[0043] In some embodiments, the weight of the first wind resource map is determined based on the spatial resolution of the first wind resource map, the weight of the second wind resource map is determined based on the spatial resolution of the second wind resource map, and the weight of the wind resource map is positively correlated with the spatial resolution of the wind resource map.

[0044] In some embodiments, the apparatus further includes:

[0045] a determination unit, configured to determine the accuracy of the fusion map based on the deviation between the wind resource parameters in the fusion map and the wind resource parameters detected at the wind measurement points; or determine the accuracy of the fusion map based on the deviation between the wind resource parameters in the fusion map and the wind resource parameters in the first wind resource map and the deviation between the wind resource parameters in the fusion map and the wind resource parameters in the second wind resource map.

[0046] In some embodiments, the feature fusion unit is configured to fuse the first type of features, the second type of features of the first wind resource map, the second type of features of the second wind resource map, and the wind resource parameters detected at the wind measurement points to obtain fusion features.

[0047] In a third aspect, a computing device is provided, where the computing device includes: a processor, the processor is coupled to a memory, and at least one computer program instruction is stored in the memory, and the at least one computer program instruction is loaded and executed by the processor to enable the computing device to implement the method provided in the first aspect or any optional manner of the first aspect.

[0048] In a fourth aspect, a computer-readable storage medium is provided, where at least one instruction is stored in the storage medium, and when the instruction runs on a computer, the computer is enabled to execute the method provided in the first aspect or any optional manner of the first aspect.

[0049] Thus, the embodiments of the present application have the following beneficial effects:

[0050] In the embodiments of the present application, feature extraction is performed on wind resource maps with different spatial resolutions of the same wind farm and image decomposition processing is performed, the features of the obtained different wind resource maps are subjected to feature fusion processing to obtain a fusion map, and the wind resources of the wind farm are evaluated based on the fusion map. Since the wind resource parameters and wind speed heights in two wind resource maps are simultaneously referred to in the fusion map, the fusion map helps to solve the problem of low accuracy in some regions in a single map. The fusion map has higher accuracy than any wind resource map in the original two wind resource maps before fusion, thereby improving the accuracy of wind resource evaluation, and further helping to improve the deployment effect of the wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic architecture diagram of a map fusion system provided by an embodiment of the present application;

[0052] Figure 2 It is a flowchart of a wind resource assessment method provided by an embodiment of the present application;

[0053] Figure 3 It is a flowchart of an algorithm for fusing maps provided by an embodiment of the present application;

[0054] Figure 4 It is a schematic structural diagram of a deep learning model provided by an embodiment of the present application;

[0055] Figure 5 It is a schematic structural diagram of a wind resource assessment device provided by an embodiment of the present application;

[0056] Figure 6 It is a schematic structural diagram of a computing device provided by an embodiment of the present application. Detailed implementation manners

[0057] To make the objectives, technical solutions, and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0058] The following explains some term concepts related to the embodiments of the present application.

[0059] (1) Wind resource map

[0060] The wind resource map includes wind resource assessment data for a certain period of time in a certain area (such as one or more wind farms). Specifically, the wind resource map is a geographical information map that visualizes the spatial distribution characteristics of wind resources by analyzing and processing wind speed and wind direction data. The wind resource map uses different colors or contour lines, etc., to represent the potential and availability of wind resources in a specific area, as well as the variation laws of wind speed and wind direction. The wind resource map is usually drawn on a geographical information system (GIS) platform and can display wind resource conditions based on geographical coordinates and spatial data.

[0061] (2) Multi-dimensional map

[0062] The multi-dimensional map in the embodiments of the present application refers to a wind resource map with multiple feature dimensions. For example, the wind resource map includes three feature dimensions, namely wind resource parameters, wind speed height, and spatial resolution.

[0063] (3) Wind resource parameters in the wind resource map

[0064] Wind resource parameters refer to the parameters used to describe and characterize wind resource characteristics in a wind resource atlas. Wind resource parameters help provide data on aspects such as wind speed, direction, turbulence, energy, etc., and contribute to a comprehensive and accurate assessment of wind resources.

[0065] Exemplarily, the wind resource parameters include at least one of longitude, latitude, wind speed, wind direction, shear, parameters of the Weibull distribution, and wind power density.

[0066] Longitude and latitude are coordinate values used to locate places in a geographic coordinate system. In a wind resource atlas, longitude and latitude can be used to identify the geographical location of an observation point or measurement position.

[0067] Wind speed refers to the speed of air movement per unit time, generally measured in meters per second (m / s). Wind speed is one of the important indicators for measuring the intensity of wind resources.

[0068] Wind direction refers to the direction from which the wind blows, usually expressed in angles (such as north, south, east, west, etc.) or azimuth angles (with due north as 0 degrees, the angle measured clockwise). Wind direction is of great significance for the layout of wind farms and the siting of wind turbines.

[0069] Shear refers to the change in wind speed or / and wind direction in the vertical direction. Shear is usually used to describe the instability and turbulence degree of the wind. Shear has an important impact on the selection and layout of wind power equipment.

[0070] The Weibull distribution is a commonly used probability distribution for describing the frequency distribution of wind speed. The parameters of the Weibull distribution include the shape parameter and the scale parameter, which can be used to estimate the probability density function and cumulative distribution function of wind speed, thereby evaluating the reliability and stability of wind resources.

[0071] Wind power density refers to the average flow rate of wind resources in the air per unit area or unit volume. Wind power density represents the richness of wind resources, usually measured in watts per square meter (W / m 2 ) or watts per cubic meter (W / m 3 ). Wind power density is an important indicator for evaluating the potential of wind resources and can be used to calculate the expected power generation of a wind farm.

[0072] Exemplarily, the wind resource atlas includes wind speed distribution. For example, by using contour lines or color gradients, etc., the wind speed distribution at different locations in a specific area is shown. Higher contour lines or colors indicate higher wind speeds, and vice versa. This helps to determine areas with rich wind resources and the siting of wind farms.

[0073] Exemplarily, the wind resource map includes wind direction distribution. For example, the distribution of wind direction in a specific area is represented by symbols such as arrows or rose diagrams. The direction of the arrow represents the wind direction, and the length or density of the arrow indicates the frequency or proportion of the corresponding wind direction. This can help understand the main wind directions and variation rules in a specific area.

[0074] Exemplarily, the wind resource map includes wind resource density. For example, wind resource density is an indicator to measure the availability of wind resources. The wind resource map can represent the distribution of wind resource density in a specific area through color gradients or contour lines, etc. Higher density areas indicate richer wind resources, which are suitable for the construction and development of wind farms.

[0075] (4) Wind speed height in the wind resource map

[0076] Wind speed height, also known as the wind speed measurement height, refers to the height position used to measure wind resource parameters in wind resource assessment. For example, the wind speed height is the height of wind speed, wind direction height, wind power density height, or wind shear height. For example, the wind speed height is 70m, 80m, 90m, 100m, 110m, 120m, 130m, 140m, or 150m, etc.

[0077] (5) Spatial resolution of the wind resource map

[0078] Spatial resolution refers to the actual geographical space size represented by a grid in the wind resource map. Spatial resolution can also be understood as the fineness of the wind resource parameters shown on the map in the geographical space. The grid in the wind resource map is the smallest component unit of the map. The grid in the wind resource map is also called a pixel, grid, or voxel. Spatial resolution is usually measured in unit length (such as meters, kilometers) or grid size (such as how many square meters per grid). Exemplarily, the spatial resolution of the wind resource map is 1km or 200m. Specifically, the smaller the spatial resolution value, the higher the accuracy of the wind resource map, and the richer the wind resource parameters shown. Conversely, it is coarser. For example, if the spatial resolution of the wind resource map is 1km x 1km, each grid represents a geographical area of 1 square kilometer, and the wind resource parameters shown on the map are the average or statistical results of this 1 square kilometer area. If the spatial resolution of the wind resource map is 100m x 100m, each grid represents a geographical area of 0.01 square kilometers, and the wind resource parameters shown on the map are more refined and can more accurately reflect the wind resource parameters of this area.

[0079] Exemplarily, if a high-precision lidar device is used for wind resource measurement, a high spatial resolution can be obtained, such as a wind resource map with a resolution of 10m x 10m. The high spatial resolution helps to very detailedly display the wind speed and wind direction changes in each small area, providing more accurate data support for the micro-siting of wind farms and the layout of wind turbines. If traditional meteorological observation stations or remotely sensed data with low resolution are used for wind resource assessment, only a low spatial resolution may be obtained, such as a wind resource map with a resolution of 1km x 1km. Although the low-spatial-resolution map can reflect the regional wind resource distribution characteristics, it may not be fine enough for specific wind farm planning and wind turbine siting.

[0080] The application scenarios of the embodiments of the present application are exemplified below.

[0081] The embodiments of the present application are applicable to scenarios for wind resource assessment based on wind resource maps.

[0082] Currently, there are usually deviations in the accuracy of different wind resource maps. Once the accuracy of the wind resource assessment results is poor, it will also affect the deployment effect of the wind farm. The problems affecting the deployment effect of the wind farm are, for example, first, affecting the accuracy of wind farm siting. For example, the wind resource map is used to evaluate the wind resource situation to determine the siting of the wind farm. If the accuracy of the map is insufficient, it may lead to a decrease in the accuracy of siting and the potential wind resources cannot be fully utilized. Second, affecting the accuracy of the number of wind turbines deployed. Specifically, the accuracy problem of the wind resource map will affect the assessment of the number of wind turbines deployed. If the map accuracy is poor and the estimation of the wind resources is inaccurate, it may lead to too many or too few wind turbines being deployed, which will directly affect the power generation capacity of the wind farm. Third, affecting the accuracy of wind turbine height selection. Specifically, the accuracy of the wind resource map will also affect the assessment of the wind turbine height. An inaccurate map may not accurately reflect the vertical distribution of wind speed and wind direction, thus leading to an incorrect selection of wind turbine height. The incorrectly selected wind turbine height may cause problems such as unstable power generation output.

[0083] The deviation in accuracy between different wind resource maps may be due to the different data sources of the maps. Exemplarily, the data source of wind resource map A is meteorological service provider A. Meteorological service provider A deployed p1 anemometers in region A and q1 anemometers in region B. Based on the wind resource parameters detected by the p1 anemometers in region A and the wind resource parameters detected by the q1 anemometers in region B, meteorological service provider A obtained wind resource map A; meteorological service provider B deployed p2 anemometers in region A and q2 anemometers in region B. Based on the wind resource parameters detected by the p2 anemometers in region A and the wind resource parameters detected by the q2 anemometers in region B, meteorological service provider B obtained wind resource map B. Among them, p1 is greater than p2, and q1 is less than p2. Since the more anemometers are deployed, the higher the accuracy of the detected wind resource parameters usually is, the accuracy of the wind resource parameters of wind resource map A in region B is lower than that of wind resource map B, and the accuracy of the wind resource parameters of wind resource map B in region A is lower than that of wind resource map A.

[0084] In addition, the anemometers from which different wind resource maps are derived may have different measurement accuracies, observation times, and observation conditions, which will also lead to differences in the accuracies of wind resource maps. In addition, there are many types of wind resource maps, and the spatial resolutions of different wind resource maps are inconsistent, especially for wind resource maps of complex terrains, the regional deviations are relatively large.

[0085] Even if the wind resource map is corrected by the anemometry data observed by the anemometer tower, since the anemometer tower can only represent the wind speed within a certain range around the anemometer tower and cannot well represent the wind speed of the entire region beyond a certain distance from the anemometer tower, the representativeness of the anemometer tower is insufficient. Therefore, using the anemometer tower data cannot optimize the accuracy of other regions outside the representative region corresponding to the anemometer tower in the map, that is, the correction process cannot correct all regions.

[0086] In view of this, in some embodiments of the present application, by fusing the wind resource maps provided by different data sources, a wind resource map with higher accuracy is obtained, thereby solving the problem of low accuracy in some regions of a single map, and further improving the accuracy of wind resource assessment, which helps to improve the deployment effect of the wind farm.

[0087] As a specific example, the wind resource parameter accuracy of wind resource map A in region B is lower than that of wind resource map B, while the wind resource parameter accuracy of wind resource map B in region A is lower than that of wind resource map A. After fusing wind resource map A and wind resource map B, in the obtained fused map, the wind resource parameters of region A in the fused map are higher in accuracy than those of wind resource map A in region A because they refer to wind resource map B. Similarly, the wind resource parameters of region B in the fused map are higher in accuracy than those of wind resource map B in region B because they refer to wind resource map A.

[0088] Thus, it can be seen that the fused map compensates for the defect of insufficient accuracy of a single map in some regions, and the accuracy of the fused map is improved compared with that of the two original maps (wind resource map A and wind resource map B).

[0089] In some embodiments, considering that the spatial resolutions of wind resource maps in the same region (such as the same wind farm) may be different. For example, the spatial resolution of wind resource map A of a wind farm is 1 km, the spatial resolution of wind resource map B of this wind farm is 200 m, and the spatial resolution of wind resource map C of this wind farm is 300 m. Generally, the lower the spatial resolution of a wind resource map, the worse its accuracy. Based on this, wind resource maps with different spatial resolutions in the same region (such as the same wind farm) are fused to obtain a fused map of this region (such as the wind farm). Since the wind resource parameters in the two original maps are referred to in the fused map, compensating for the defect of insufficient accuracy of the wind resource map with a lower spatial resolution, the accuracy of the fused map is higher than that of the two original maps, thus meeting the requirements for wind resource assessment. Unifying multiple wind resource maps with different spatial resolutions into a wind resource map with the same spatial resolution also reduces the amount of calculation and processing complexity.

[0090] Next, an example of the software system architecture applied in the embodiments of the present application will be given.

[0091] Refer to Figure 1 , Figure 1 which shows a schematic architecture diagram of a map fusion system 10 provided by an embodiment of the present application. Figure 1 The shown map fusion system 10 includes a map data analysis module 110, a map fusion module 120, and an accuracy evaluation module 130. The map fusion system 10 is, for example, set on a computing device or on a computing device cluster including multiple computing devices.

[0092] The wind resource map data analysis module 110 is used to analyze the data in at least two wind resource maps so as to input the data in the at least two wind resource maps into the map fusion module 120. The at least two wind resource maps are, for example, all multi-dimensional maps. For example, the wind resource map data analysis module 110 inputs the spatial resolution, wind resource parameters, and wind speed height of each wind resource map into the map fusion module 120.

[0093] The map fusion module 120 is used to fuse two wind resource maps to obtain a fused map. The map fusion module 120 includes a feature extraction module 121 and a feature fusion module 122. The feature extraction module 121 is used to extract the features of the wind resource map based on the spatial resolution, wind resource parameters, and wind speed height of the wind resource map. The feature fusion module 122 is used to fuse the features of each wind resource map.

[0094] The accuracy evaluation module 130 is used to evaluate the accuracy of the fused map. The accuracy evaluation module 130 includes a wind measurement point evaluation module 131 and a grid space representativeness evaluation module 132. The wind measurement point evaluation module 131 is used to evaluate the accuracy of the fused map based on the wind resource parameters detected at the wind measurement points. The grid space representativeness evaluation module 132 is used to evaluate the accuracy of the fused map based on the deviation between the fused map and the original wind resource map.

[0095] Optionally, the wind resource map data analysis module 110 further includes a format conversion sub-module 114. The format conversion sub-module 114 is used to convert the format of the wind resource map into an image file format. Considering that the format of the wind resource map is usually the file generation format of data acquisition software, such as the Network Common Data Format (also known as the NetCDF format or nc format), by converting the format of the wind resource map into an image file format, such as converting it into the Tagged Image File Format (tif), the complexity of map fusion by the map fusion module 120 is reduced. In particular, in the case where the algorithm adopted in the map fusion module 120 includes a CNN, since images are the main input data of the CNN. The tif format, as a common image file format, can be directly supported by the image processing libraries and frameworks of the CNN. Therefore, the complexity of feature extraction from the wind resource map by the CNN is reduced.

[0096] Optionally, the format conversion sub-module 114 is further used to perform image segmentation on the wind resource map in the image file format to obtain multiple image blocks, and input the data in the multiple image blocks into the map fusion module 120.

[0097] Alternatively, the step of converting the format of the wind resource map into an image file format is omitted, and the data in at least two wind resource maps is directly input into the map fusion module 120.

[0098] Attached Figure 2 is a flowchart of a wind resource assessment method provided by an embodiment of the present application. Attached Figure 2 The method 200 shown is, for example, performed by Figure 1 the map fusion system 10 shown. Attached Figure 2 The method shown is, for example, performed by a computing device or by a computing device cluster including multiple computing devices. Attached Figure 2 The method shown involves a fusion process for multiple wind resource maps. To distinguish different wind resource maps, "the first wind resource map" and "the second wind resource map" are used to distinguish and describe different wind resource maps, or "wind resource map A" and "wind resource map B" are used to distinguish and describe different wind resource maps. Attached Figure 2 The method shown includes the following steps.

[0099] Step S210, obtain the first wind resource map and the second wind resource map.

[0100] The first wind resource map and the second wind resource map are wind resource maps with different spatial resolutions for the same wind farm.

[0101] In some embodiments, the first wind resource map and the second wind resource map include the same wind resource parameters, and / or the first wind resource map and the second wind resource map are wind resource maps at the same wind speed height. In other embodiments, the first wind resource map and the second wind resource map are wind resource maps that include the same wind resource parameters and have different wind speed heights.

[0102] Step S220, perform feature extraction on the first wind resource map and the second wind resource map to obtain the first type of features.

[0103] In some embodiments, the first type of features is used to reflect the commonality between the first wind resource map and the second wind resource map. For example, the first type of features includes the similarity between the wind resource parameters in the first wind resource map and the wind resource parameters in the second wind resource map and / or the similarity between the wind speed heights in the first wind resource map and the wind speed heights in the second wind resource map.

[0104] In some embodiments, the first type of feature is used to characterize the fusion weight of the first wind resource map and the second wind resource map. For example, the first type of feature can be a weight map. The weight map includes the fusion weights of the first wind resource map and the second wind resource map, and the fusion weight is, for example, the similarity between the first wind resource map and the second wind resource map. For example, the value of any pixel point in the weight map represents the similarity between the feature (wind resource parameter or / and wind speed height) of the first wind resource map at this pixel point and the feature (wind resource parameter or / and wind speed height) of the second wind resource map at this pixel point. For example, the data input into the deep learning model is a pair of image patches in the same region of the first wind resource map and the second wind resource map, and the value range of each pixel point in the weight map output by the deep learning model is between (0, 1). The value of each pixel point in the weight map represents the similarity information of a pair of image patches of the two input wind resource maps in terms of (wind resource parameter or / and wind speed height). For example, when the value of a pixel point in the weight map is closer to 1, it indicates that the features (wind resource parameter or / and wind speed height) of the first wind resource map and the second wind resource map at this pixel point are not similar, and the amount of information of the features (wind resource parameter or / and wind speed height) included in the first wind resource map at this pixel point is greater; when the value of a pixel point in the weight map is closer to 0.5, it indicates that the first wind resource map and the second wind resource map are relatively similar; when the value of a pixel point in the weight map is closer to 0, it indicates that the features (wind resource parameter or / and wind speed height) of the first wind resource map and the second wind resource map are not similar, and the amount of information of the features (wind resource parameter or / and wind speed height) included in the second wind resource map at this pixel point is smaller.

[0105] In some embodiments, the first wind resource map and the second wind resource map are input into multiple convolutional layers in the deep learning model to obtain the feature map of the target wind resource map; the first type of feature is obtained based on the feature map through the activation function of the deep learning model.

[0106] In some embodiments, the deep learning model is a Convolutional Neural Networks (CNN). CNN is a deep learning model commonly used to process image data. The convolutional layers in CNN can effectively extract local features in the image, while the activation function can increase the non-linear expression ability of the model. In some other embodiments, the deep learning model is a Fully Convolutional Network (FCN), a Self-Attention Mechanism neural network (such as the TranS2ormer model), a Recurrent Convolutional Neural Network (RCNN), or a Residual Neural Network (ResNet), etc.

[0107] The Convolutional Layer is an important component in deep learning models. The convolutional layer is used to extract local features from the input first wind resource map and second wind resource map. The convolutional layer performs a point-by-point convolution operation on the input first wind resource map and second wind resource map respectively through a sliding convolutional kernel (a type of filter), obtaining the feature map of the first wind resource map and the feature map of the second wind resource map. The convolutional layer can capture the local spatial relationships of the input first wind resource map and second wind resource map, and reduce the number of parameters in the deep learning model through weight sharing, with translational invariance and position awareness. The convolutional layer usually includes multiple convolutional kernels, and each convolutional kernel is responsible for detecting different features, such as edges, textures, etc. By slidingly applying the convolutional kernel, the convolutional layer can detect the local features of the input first wind resource map and second wind resource map, and use the detected features as the input data for the next layer.

[0108] The Activation Function is a non-linear function in deep learning models, used to introduce non-linearity. The main role of the activation function is to increase the non-linear expression ability of the deep learning model, enabling the deep learning model to better learn and simulate complex input-output relationships. Without an activation function, no matter how many layers the neural network has, its output will be a linear combination of the inputs, which will greatly limit the expression ability of the deep learning model. The activation function can transform the weighted sum of the inputs into a non-linear response of the output, enabling the deep learning model to learn complex non-linear relationships. Activation functions include ReLU (Rectified Linear Unit), Sigmoid, tanh, etc.

[0109] The Feature Map is a data structure in deep learning models, especially playing an important role in Convolutional Neural Networks (CNNs). The feature map is the result obtained by performing a convolution operation on the input wind resource map through a convolutional kernel, and can be regarded as an abstract representation of the original wind resource map. In the convolutional layer of a deep learning model, each convolutional kernel performs a convolution operation on the input wind resource map, and the output obtained is the feature map. The size and number of feature maps depend on parameters such as the size and number of convolutional kernels, as well as the stride and padding method of the convolution. These feature maps can be regarded as the filter responses of the input wind resource map, and each feature map captures a certain specific type of feature of the input wind resource map.

[0110] In some embodiments, the deep learning model further includes a pooling layer, which is connected to the convolutional layer and is used to perform a pooling operation on the feature map output by the convolutional layer. The main function of the pooling layer is to reduce the dimension of the input feature map, thereby reducing the computational amount and extracting the important features of the input feature map. The pooling layer usually follows the convolutional layer closely and is used to reduce the size of the output of the convolutional layer. By applying the pooling layer, the model can maintain good feature extraction ability while reducing computational complexity. For example, the pooling layer is used to perform max pooling or average pooling on the feature map. Max pooling means dividing the input feature map into non-overlapping regions and then selecting the maximum value in each region as the output. Average pooling performs feature extraction by calculating the average value within a local region. Specifically, average pooling divides the input feature map into non-overlapping regions and then calculates the average value in each region as the output.

[0111] In some embodiments, the deep learning model further includes a fully connected layer (also known as a dense layer or a linear layer). Each node in the fully connected layer is connected to all nodes in the previous layer (usually a convolutional layer or a pooling layer). This means that each node in the fully connected layer receives the outputs of all nodes in the previous layer, performs a weighted combination and a non-linear transformation on these inputs, and generates new features. The output of the fully connected layer can be used as the prediction result of the last layer. Technically, the fully connected layer is usually implemented through matrix multiplication operations and activation function operations. For example, the outputs of the nodes in the previous layer are flattened into a vector and then multiplied by a weight matrix to obtain a new feature representation. Usually, a bias parameter is also introduced for each node to adjust the sensitivity of the node to the input data. Finally, the obtained result is subjected to a non-linear transformation through an activation function to introduce non-linear relationships.

[0112] Exemplarily, please refer to Figure 3 , Figure 3 shows a flowchart of an algorithm for fusing atlases provided by an embodiment of the present application. As Figure 3 shown, the first wind resource atlas is an atlas with a first spatial resolution provided by data source A, and the second wind resource atlas is an atlas with a second spatial resolution provided by data source B. The first wind resource atlas and the second wind resource atlas are input into a CNN, and a weight map G{W} is obtained through the CNN. The weight map G{W} represents the similarity between the first wind resource atlas and the second wind resource atlas. The input data is designed according to a two-channel design, that is, the dimensions of the feature variables of the first wind resource atlas and the second wind resource atlas are the same, and both include three dimensions: spatial resolution, wind resource parameters, and wind speed height. The image block size is set to 16X16. The deep learning model includes 4 convolutional layers, 1 max pooling layer, and 1 fully connected layer.

[0113] In some embodiments, the convolution kernel size in the convolutional layer of the deep learning model is 3X3. The convolution kernel size determines the receptive field size of the convolutional layer for feature extraction, or in other words, the neighborhood range to be considered during each sliding step. A convolution kernel size of 3X3 means that the size of the convolution kernel is 3 rows and 3 columns, or in other words, the size of the convolution kernel is 3x3 pixels.

[0114] In some embodiments, the stride of the convolution kernel in the deep learning model is 2. The stride refers to the step size at which the convolution kernel slides on the input data. A stride of 2 means that each convolution kernel moves two pixels to the right or down each time. The stride can control the downsampling degree of the model on the input data. Reducing the stride can reduce the spatial size of the data and improve the generalization ability of the model.

[0115] In some embodiments, the objective function is set as the softmax loss function. The objective function is used to measure the gap between the model prediction result and the true result. The softmax loss function measures the prediction accuracy by calculating the cross-entropy between the prediction result and the true result.

[0116] In some embodiments, the SGD (Stochastic Gradient Descent) algorithm is used to train the deep learning model. Specifically, during the training process, the gradient value of the model parameters is calculated, and the model parameters are updated according to the direction of the gradient value at a certain learning rate. The SGD algorithm updates the parameters based on randomly sampled mini-batch data, making the training process more efficient.

[0117] In some embodiments, the deep learning framework is selected as Pytorch. This open-source machine learning library, Pytorch, is used as a tool for building and training deep learning models. Pytorch provides rich functions and convenient APIs, making the development and training of deep learning models simpler.

[0118] Step S240, perform image decomposition processing on the first wind resource map and the second wind resource map respectively to obtain the second type of features of the first wind resource map and the second type of features of the second wind resource map.

[0119] In some embodiments, the second type of features includes residual images. The first wind resource map and the second wind resource map are respectively decomposed using a pyramid to obtain N first pyramid maps of different scales corresponding to the first wind resource map and N second pyramid maps of different scales corresponding to the second wind resource map, where N is the number of pyramid layers and N≥2. For the N first pyramid maps of different scales, the N - 1 first pyramid maps are upsampled and differenced from the Nth first pyramid map to obtain the residual image corresponding to the first wind resource map; for the N second pyramid maps of different scales, the N - 1 second pyramid maps are upsampled and differenced from the Nth second pyramid map to obtain the residual image corresponding to the second wind resource map.

[0120] A pyramid is a structured multi-scale image representation. Decomposing the first wind resource map using a pyramid, for example, is downsampling the first wind resource map at N different scales until a termination condition is reached to stop sampling. Exemplarily, the bottom layer of the first pyramid map is the first wind resource map, and as the level in the first pyramid map increases, the spatial resolution gradually decreases.

[0121] The construction of a Gaussian pyramid starts from the original image and obtains a series of images with different resolutions through continuous filtering and downsampling, forming a pyramid-like structure.

[0122] In some embodiments, a Gaussian pyramid is used to decompose the first wind resource map to obtain N first pyramid maps of different scales. For example, a Gaussian filter and downsampling are used to gradually generate wind resource maps of reduced scales. The specific process, for example, is to first apply a low-pass filter (such as a Gaussian filter) to the first wind resource map for smoothing to remove high-frequency noise and detailed information; then, the smoothed first wind resource map is downsampled, that is, the size of the smoothed first wind resource map is reduced. The downsampling process can be achieved, for example, by adjusting the sampling interval or using an interpolation algorithm. The scale of each pyramid level is reduced by half compared to the previous level, and the detailed information of the image is also correspondingly reduced. By repeating the steps of smoothing and downsampling until the bottom layer of the pyramid or a specified number of layers is reached. Different-scale smoothed images are obtained for each layer. In the decomposition process of the wind resource map, a Gaussian pyramid can be used to extract different spatial frequency components of the image. By performing feature extraction and processing on images of different levels, the distribution characteristics of wind resources at different scales can be obtained, which helps to analyze the distribution of wind resources.

[0123] In some embodiments, the Laplacian pyramid is used to decompose the first wind resource map to obtain N first pyramid maps of different scales. The Laplacian pyramid also obtains a series of images with different resolutions through continuous filtering and downsampling. However, the Laplacian pyramid extracts the edge and detail information in the image by calculating the difference between adjacent-level images. For example, the Laplacian pyramid is constructed by taking the difference between each level of the Gaussian pyramid and its upper-level image. Each level of the Laplacian pyramid contains the detail information in the original wind resource map, and the resolution of each level of the Laplacian pyramid gradually decreases. For example, a series of initial pyramid maps of different scales are first obtained through Gaussian pyramid decomposition. For the top layer of the Laplacian pyramid, it is the same as the corresponding layer image of the Gaussian pyramid, that is, it serves as the bottom layer of the pyramid. Starting from the second-to-last layer, the upsampling method is used to enlarge the lower-resolution image to the same size as the upper-layer image. The upsampled image in the upper layer is subtracted from the Gaussian pyramid image of the current layer to obtain the Laplacian pyramid image of the current layer. Repeat the above steps until the top layer of the pyramid is reached, thus obtaining N first pyramid maps of different scales. In the decomposition process of the wind resource map, the Laplacian pyramid can be used to extract high-frequency components such as edges and vortices in the image, and these components usually correspond to the movement and change characteristics of the wind resource. By analyzing and processing these high-frequency components, the dynamic characteristics and vortex structure of the wind resource can be obtained, which helps to better predict and understand the distribution and evolution of the wind resource.

[0124] Differential processing refers to performing a subtraction operation on two images, subtracting the pixel values of the two image pixels at the corresponding positions. For the generation of the residual image in the pyramid map decomposition, differential processing is used to obtain the detail information between the high-level pyramid map and the low-level map. For example, for the pyramid map of the first wind resource map, the first pyramid map of the N-1th layer is upsampled to make its size the same as that of the first pyramid map of the Nth layer. Then, a differential operation is performed on the two pyramid maps, subtracting the upsampled image from the first pyramid map of the Nth layer pixel by pixel. This differential operation will extract the detail changes between the high-level and low-level images.

[0125] The residual image is the result obtained through differential processing. The residual image can be regarded as the difference or detail map between two pyramid maps. The residual image describes the changing part of the high-frequency information or detail information because in the pyramid decomposition, each level stores the detail information at different scales. By extracting and analyzing the residual image, it is convenient to process the detail features in the image.

[0126] The process of performing image decomposition on the second wind resource map is the same as that of performing image decomposition on the first wind resource map, and reference can be made to the process of performing image decomposition on the first wind resource map.

[0127] In some embodiments, the Laplacian pyramid is used to decompose the first wind resource map into multiple band images of different scales. Each band image contains the detailed features of the first wind resource map at the corresponding scale and corresponding spatial frequency. Different band images correspond to the components of different spatial frequencies in the first wind resource map. Similarly, the Laplacian pyramid is used to decompose the second wind resource map into multiple band images of different scales. Subsequently, by using the band images of different scales in the first wind resource map and the band images of different scales in the second wind resource map, it is helpful to fully fuse the features and details of different wind resource characteristics. Exemplarily, the first wind resource map is decomposed by using the Laplacian pyramid method to obtain the coefficient L{I}, and the second wind resource map is decomposed by using the Laplacian pyramid method to obtain the coefficient L{V}, as shown in the following formula (1).

[0128] L(i) = G(i) - PyrUp(G(i + 1)); (1)

[0129] Where G(i) represents the image of this layer, G(i + 1) represents the image of the previous layer, and PyrUp represents the upsampling operation.

[0130] This embodiment does not limit the timing of feature extraction using a deep learning model and image decomposition using a pyramid. In some embodiments, feature extraction using a deep learning model and image decomposition using a pyramid can be executed sequentially. For example, first execute feature extraction using a deep learning model, and then execute image decomposition using a pyramid; or first execute image decomposition using a pyramid, and then execute feature extraction using a deep learning model. In other embodiments, feature extraction using a deep learning model and image decomposition using a pyramid can also be executed in parallel, that is, feature extraction using a deep learning model and image decomposition using a pyramid can be executed simultaneously.

[0131] Step S260, perform feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fused map.

[0132] In some embodiments, the map fusion mode parameters are determined based on the second type of features of the first wind resource map and the second type of features of the second wind resource map; based on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map, fusion processing is performed using the fusion strategy corresponding to the map fusion mode parameters to obtain a fused map.

[0133] In some embodiments of determining the atlas fusion mode parameters, the first wind resource atlas is decomposed into N first pyramid atlases L{I} at different scales, where l represents the l-th decomposition level, and the second wind resource atlas is decomposed into N second pyramid atlases L{V} at different scales, where l represents the l-th decomposition level. A local energy map of the first wind resource atlas is obtained based on the first pyramid atlas L{I}. A local energy map of the second wind resource atlas is obtained based on the second pyramid atlas L{V}. The atlas fusion mode parameters are determined based on the local energy map of the first wind resource atlas and the local energy map of the second wind resource atlas.

[0134] Exemplarily, the local energy map of the first wind resource atlas and the local energy map of the second wind resource atlas are determined based on the following formula (2).

[0135]

[0136]

[0137] Where (x, y) represents a coordinate point in the first pyramid atlas or the second pyramid atlas. represents the local energy map of the first pyramid atlas at the l-th decomposition level at the position (x, y). represents the local energy map of the second pyramid atlas at the l-th decomposition level at the position (x, y). For each decomposition level l, the local energy map is calculated by iteratively calculating around the pixels at the position (x, y).

[0138] For example, for the first wind resource atlas, use L{I}l(x + m, y + n) to represent the pixel value at the position offset by (m, n) on the l-th decomposition level of the first pyramid atlas. Calculate the square of L{I}l(x + m, y + n), and sum all the squared values to get Similarly, for the second wind resource atlas, use L{V}l(x + m, y + n) to represent the pixel value at the position offset by (m, n) on the l-th decomposition level of the second pyramid atlas. Calculate the square of L{V}l(x + m, y + n), and sum all the squared values to get

[0139] The local energy map is obtained by calculating the sum of the squares of adjacent elements in the Laplacian pyramid, which reflects the wind resource intensity and variation degree at that position. By calculating the local energy map for each decomposition level, it helps to analyze and fuse the wind resource distribution characteristics at different scales.

[0140] In some embodiments, based on the local energy map of the first wind resource map, the local energy map of the second wind resource map, the first pyramid map, and the second pyramid map, the map fusion mode parameters are determined. For example, the map fusion mode parameters are determined by the following formula (3).

[0141]

[0142] Where M represents the map fusion mode parameter. M l (x, y) represents the value of the parameter M of the fusion mode at the coordinate point (x, y) when the decomposition level is l. L{I} l (x + m, y + n) represents the pixel value of the first pyramid map at the position offset by (m, n) at the decomposition level l. L{V} l (x + m, y + n) represents the pixel value of the second pyramid map at the position offset by (m, n) at the decomposition level l. represents the value of the local energy map of the first wind resource map at the coordinate point (x, y) at the decomposition level l. represents the value of the local energy map of the second wind resource map at the coordinate point (x, y) at the decomposition level l. The purpose of the above formula is to calculate the map fusion mode parameter M according to the local energy and its correlation of the first wind resource map and the second wind resource map at each decomposition level and each position. For example, the fusion mode parameter M is calculated by adding the Laplacian pyramid coefficients of the two maps and dividing by the sum of the local energies.

[0143] In some embodiments, in response to the map fusion mode parameter belonging to the first threshold range, weighted summation is performed on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map based on the weight of the first wind resource map and the weight of the second wind resource map to obtain a fused map; in response to the map fusion mode parameter belonging to the second threshold range, the second type of features of the first wind resource map are inversely transformed to obtain a fused map; in response to the map fusion mode parameter belonging to the third threshold range, the second type of features of the second wind resource map are inversely transformed to obtain a fused map.

[0144] For example, the fused map is obtained by using the following formula (4).

[0145]

[0146] Where t represents the threshold. G{W} represents the weight map of the wind resource map. The values of all pixel points in the weight map represent the similarity of the first wind resource map and the second wind resource map at the corresponding pixel points.

[0147] If the pixel value M of the fusion mode parameter M at the coordinate point (x, y) at the decomposition level ll If (x, y) is greater than or equal to the threshold t, then based on the weight map W, a weighted summation fusion mode is adopted for fusion. Specifically, based on the value G(W)l(x, y) of the weight map at the coordinate point (x, y) at the decomposition level l, the pixel value of the first pyramid spectrum at the coordinate point (x, y) at the decomposition level l, and the value of the weight coefficient map of the second wind resource spectrum at the coordinate point (x, y), a weighted summation is performed to obtain the fusion eigenvalue L{F}l(x, y) of the first wind resource spectrum and the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l.

[0148] If the pixel value M of the fusion mode parameter M at the coordinate point (x, y) at the decomposition level l l If (x, y) is less than the threshold t, then the value of the local energy map of the first wind resource spectrum at the coordinate point (x, y) at the decomposition level l and the value of the local energy map of the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l are compared, and the maximum selection fusion mode is adopted. Specifically, if the value of the local energy map of the first wind resource spectrum at the coordinate point (x, y) at the decomposition level l is greater than the value of the local energy map of the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l, then the value of the local energy map of the first wind resource spectrum at the coordinate point (x, y) at the decomposition level l is selected to obtain the fusion eigenvalue L{F}l(x, y) of the first wind resource spectrum and the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l. If the value of the local energy map of the first wind resource spectrum at the coordinate point (x, y) at the decomposition level l is less than the value of the local energy map of the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l, then the value of the local energy map of the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l is selected to obtain the fusion eigenvalue L{F}l(x, y) of the first wind resource spectrum and the second wind resource spectrum at the coordinate point (x, y) at the decomposition level l.

[0149] In some embodiments, after obtaining the fusion feature L{F} of the first wind resource spectrum and the second wind resource spectrum at each coordinate point at each decomposition level, an inverse transform is performed on the fusion feature L{F} using a pyramid, thereby

[0150] Reconstruct the fused map. For example, the way to reconstruct the map is to use the inverse pyramid transform to re-stack the fused features with different resolutions together, so as to obtain a fused map with the same size as the original wind resource map. For example, the Laplacian pyramid method is to subtract the residual information obtained by subtracting the upsampled and convolved prediction maps corresponding to each layer of the pyramid downsampling from each layer of the image, and then combine the images of each layer to gradually construct the original wind resource map, that is, the final fused map. For example, in the process of inverse transformation, the fused features of each level are stacked on the reconstructed image of the previous level. In this way, the fused features of the lower resolution levels are gradually stacked on the images of the higher resolution levels until the inverse transformation is completed at the lowest resolution level. The finally obtained fused map is composed of the results of the inverse transformation at all resolution levels stacked together and has the same size as the original wind resource map.

[0151] In some embodiments, the first type of features, the second type of features of the first wind resource map, the second type of features of the second wind resource map, and the wind resource parameters detected at the wind measurement points are fused to obtain fused features.

[0152] In some embodiments, the weight of the first wind resource map is determined based on the spatial resolution of the first wind resource map, the weight of the second wind resource map is determined based on the spatial resolution of the second wind resource map, and the weight of the wind resource map is positively correlated with the spatial resolution of the wind resource map.

[0153] Step S270, determine that the accuracy of the fused map meets the requirements.

[0154] In some embodiments, the accuracy of the fused map is determined based on the deviation between the wind resource parameters in the fused map and the wind resource parameters detected at the wind measurement points.

[0155] Exemplarily, the wind resource map set is divided into a training set and a test set. The training set is used to train the deep learning model, and the test set is used to evaluate the performance of the deep learning model. When the accuracy of the fused map corresponding to the wind resource map in the training set reaches the accuracy threshold and the accuracy of the fused map corresponding to the wind resource map in the test set reaches the accuracy threshold, it is determined that the accuracy of the fused map meets the requirements.

[0156] For example, the wind resource atlas is divided into a training set and a test set at a ratio of 4:1. On the training set, the deviation between the wind resource parameters in the fused atlas and the wind resource parameters detected at the wind measurement points is evaluated, and thus the accuracy is calculated. The accuracy represents the ratio of the wind resource parameters in the fused atlas that are consistent with those detected at the wind measurement points. In addition, on the test set, the deviation between the wind resource parameters in the fused atlas and the wind resource parameters detected at the wind measurement points is evaluated, and thus the accuracy is calculated. If the accuracy of both the training set and the test set reaches over 60%, that is, the wind resource parameters in the fused atlas can well match the wind resource parameters detected at the wind measurement points on both data sets, then it can be determined that the accuracy of the fused atlas meets the requirements.

[0157] In some embodiments, based on the deviation between the wind resource parameters in the fused atlas and the wind resource parameters in the first wind resource atlas, and the deviation between the wind resource parameters in the fused atlas and the wind resource parameters in the second wind resource atlas, the accuracy of the fused atlas is determined. Specifically, in a way of evaluating the representativeness of the grid space, the accuracy of the fused atlas is determined from the data included in the whole fused atlas. The grid space is the grid in the fused atlas. Focus is placed on the wind resource parameters of the grids in the fused atlas that are far from the wind measurement points. By judging the deviation between the wind resource parameters in the fused atlas and the original atlases (the first wind resource atlas and the second wind resource atlas), and comprehensively judging the quality of the correction effect in combination with the surrounding terrain and altitude difference, if the deviation between the wind resource parameters in the fused atlas and the wind resource parameters in the first wind resource atlas and the second wind resource atlas does not exceed a certain predetermined threshold in more than 80% of the areas, then it is determined that the accuracy of the fused atlas meets the requirements. In other words, when the deviation between the fused atlas and the original atlases is small in most areas (i.e., does not exceed a certain threshold), it is determined that the accuracy of the fused atlas meets the requirements.

[0158] Optionally, before performing the feature extraction described in step S220, the format of the first wind resource map is also converted into an image file format, and the format of the second wind resource map is converted into an image file format. In the process of executing step S220, feature extraction is performed on the first wind resource map in image file format and the second wind resource map in image file format to obtain the first type of features. For example, the format of the first wind resource map is converted from nc format to tif format, and the format of the second wind resource map is converted from nc format to tif format, and feature extraction is performed on the first wind resource map in tif format and the second wind resource map in tif format to obtain the first type of features. Considering that the map in tif format is easier to perform spatial analysis and processing. The tif format is suitable for spatial operations such as image segmentation, registration, and interpolation, and is convenient for fusion and comparison with other maps. In addition, the tif format supports a variety of color modes and compression methods, which can save richer information and higher image quality in the map, which helps to provide more features and information for the CNN algorithm, so as to better process the map.

[0159] Step S280: Evaluate the wind resources of the wind farm based on the fused graph.

[0160] For example, the location of the wind farm to be deployed is determined on the fusion map. The wind resource data at the location of the wind farm is extracted from the wind resource map, including wind speed, wind direction, shear and other information at different heights. The extracted wind resource data is analyzed, including statistical distribution characteristics of wind speed and wind direction, and calculation of parameters such as wind power density to evaluate the richness of wind energy resources in the wind farm. Based on the wind resource data and the design parameters of the wind farm, the theoretical power generation, capacity factor and other performance indicators of the wind farm are predicted to determine the feasibility of deploying the wind farm. For another example, according to the wind resource parameters, the layout of the wind farm is optimized, including the selection and layout of wind turbines, so as to improve the power generation efficiency of the wind farm. For another example, the wind resource parameters and prediction models in the wind resource map are used to predict the future power generation of the wind farm, providing a reference for the operation and scheduling of the wind farm.

[0161] The method provided in this embodiment extracts features from wind resource maps of different spatial resolutions of the same wind farm and performs image decomposition processing, fuses the features of the obtained different wind resource maps, obtains a fused map, and evaluates the wind resources of the wind farm based on the fused map. Since the fused map refers to the wind resource parameters and wind speed height in the two wind resource maps at the same time, the fused map helps to solve the problem of low accuracy in some areas of a single map. The fused map is more accurate than any wind resource map in the two original wind resource maps before fusion, thereby improving the accuracy of wind resource evaluation, and further helping to improve the deployment effect of the wind farm.

[0162] Attached Figure 2The method shown above can optionally be used to evaluate wind resources after fusing three or more wind resource maps. In scenarios where there are three or more wind resource maps, in some embodiments, the maps are grouped in pairs for fusion. For example, for wind resource maps A, B, and C in the same region (such as in the same wind farm), the method shown in Appendix Figure 2 is used to fuse wind resource maps A and B to obtain fused map 1, the method shown in Appendix Figure 2 is used to fuse wind resource maps B and C to obtain fused map 2, and the method shown in Appendix Figure 2 is used to fuse fused map 1 and fused map 2 to obtain fused map 3. The wind resources of the wind farm are evaluated based on fused map 3.

[0163] The method shown in the above appendix will be illustrated below with an example. Figure 2 The method shown above will be illustrated below with an example.

[0164] During the execution of S210, wind resource maps provided by different data sources are obtained. For example, wind resource map A with a spatial resolution of 1 km provided by meteorological service provider A, wind resource map B with a spatial resolution of 200 m provided by meteorological service provider A, and wind resource map C with a spatial resolution of 200 m provided by meteorological service provider B are obtained.

[0165] Exemplarily, the data included in wind resource map A is shown in Table 1 below.

[0166] Table 1

[0167]

[0168] Exemplarily, the data included in wind resource map B is shown in Table 2 below.

[0169] Table 2

[0170]

[0171] Exemplarily, the data included in wind resource map C is shown in Table 3 below.

[0172] Table 3

[0173]

[0174] The spatial resolution, wind resource parameters, and wind speed height are selected as the data that will be subsequently input into the CNN and Laplacian pyramid.

[0175] During the execution of S220 and S240, the spatial resolution, wind resource parameters, wind speed height of the wind resource map A obtained by executing S210, the spatial resolution, wind resource parameters, and wind speed height of the wind resource map B, and the spatial resolution, wind resource parameters, and wind speed height of the wind resource map C are respectively input into the CNN and the Laplacian pyramid for model training.

[0176] Exemplarily, please refer to Figure 4 , Figure 4 which is a schematic structural diagram of a deep learning model provided by an embodiment of the present application. Figure 4 The shown deep learning model is a CNN, which includes four convolutional layers, a max pooling layer, a fully connected layer, and an activation function. The size of the convolutional kernel is 3x3, and the convolutional kernel stride is 2. The output of the CNN is a weight map, and the value of each pixel point in the weight map is in the interval (0, 1). The value of each pixel point in the weight map represents the similarity information of a pair of image patches in the same region of the two input wind resource maps. For example, the wind resource map A includes an image patch p1, the wind resource map B includes an image patch p2, the region where the image patch p1 is located in the wind resource map A is the same as the region where the image patch p2 is located in the wind resource map B, the weight map includes an image patch p3, the region where the image patch p3 is located in the weight map is the same as the region where the image patch p1 is located in the wind resource map A, and the value of the image patch p3 represents the similarity between the wind resource parameters in the image patch p1 and the wind resource parameters in the image patch p2.

[0177] This training process is completed on a local server. The CPU is an interl i7 with 52 cores, the memory is 256G, the graphics card is an RTXA5000, and the video memory is 24G. According to the parameter settings, parameter optimization is selected. The simulation takes a total of 48 hours, and the optimal result is selected after parameter optimization for simulation to obtain the final result.

[0178] During the execution of S270, modeling calculations are performed based on the fusion map. The training set and the test set are divided according to 4:1. The final accuracy of the training set is 89%, and the result of the test set is 75%. It is determined that the accuracy of the fusion map meets the requirements. In addition, from the judgment of the grid space representativeness, the difference between the fusion map and the result of the previous version map is calculated, and the deviation of 86% of the region is controlled within 0.5 m / s, which meets the evaluation result.

[0179] The map fusion algorithm provided in this embodiment fully considers the feature information in different maps, assigns higher weights to high-resolution maps, and is more effective for complex terrain regions, thereby improving the map accuracy.

[0180] In addition, the dominant information is fused from the existing maps to make up for the problems of a single wind resource map.

[0181] In addition, the used map fusion algorithm combines the Laplacian pyramid method, extracts more feature information of the wind resource map, is more effective for the complex terrain area structure, and improves the accuracy of the wind resource map.

[0182] Figure 5 FIG. 4 is a schematic structural diagram of a wind resource assessment device 600 provided by an embodiment of the present application. The device 600 includes:

[0183] A feature extraction unit 610, configured to extract features from a first wind resource map and a second wind resource map to obtain a first type of features. The first wind resource map and the second wind resource map are wind resource maps with different spatial resolutions for the same wind farm;

[0184] An image decomposition unit 620, configured to perform image decomposition processing on the first wind resource map and the second wind resource map respectively to obtain a second type of features of the first wind resource map and a second type of features of the second wind resource map;

[0185] A feature fusion unit 630, configured to perform feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fused map;

[0186] An evaluation unit 640, configured to evaluate the wind resources of the wind farm based on the fused map.

[0187] In some embodiments, the first wind resource map and the second wind resource map include the same wind resource parameters, and / or the first wind resource map and the second wind resource map are wind resource maps at the same wind speed height.

[0188] In some embodiments, the feature extraction unit 610 is configured to input the first wind resource map and the second wind resource map into multiple convolutional layers in a deep learning model to obtain a feature map of the first wind resource map and a feature map of the second wind resource map; and obtain the first type of features based on the feature maps through an activation function of the deep learning model. The first type of features is used to characterize the fusion weights of the first wind resource map and the second wind resource map.

[0189] In some embodiments, the second type of features includes residual images. The image decomposition unit 620 is configured to decompose the first wind resource map and the second wind resource map respectively by using a pyramid to obtain N first pyramid maps with different scales corresponding to the first wind resource map and N second pyramid maps with different scales corresponding to the second wind resource map, where N is the number of layers of the pyramid and N≥2. For the N first pyramid maps with different scales, the N - 1 first pyramid maps are upsampled and differenced from the Nth first pyramid map to obtain the residual image corresponding to the first wind resource map; for the N second pyramid maps with different scales, the N - 1 second pyramid maps are upsampled and differenced from the Nth second pyramid map to obtain the residual image corresponding to the second wind resource map.

[0190] In some embodiments, the feature fusion unit 630 is configured to determine a map fusion mode parameter based on the second type of features of the first wind resource map and the second type of features of the second wind resource map; and perform fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map by using a fusion strategy corresponding to the map fusion mode parameter to obtain a fused map.

[0191] In some embodiments, the feature fusion unit 630 is configured to, in response to the map fusion mode parameter belonging to a first threshold range, perform weighted summation on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map based on the weight of the first wind resource map and the weight of the second wind resource map to obtain a fused map; in response to the map fusion mode parameter belonging to a second threshold range, perform inverse transformation on the second type of features of the first wind resource map to obtain a fused map; and in response to the map fusion mode parameter belonging to a third threshold range, perform inverse transformation on the second type of features of the second wind resource map to obtain a fused map.

[0192] In some embodiments, the weight of the first wind resource map is determined based on the spatial resolution of the first wind resource map, the weight of the second wind resource map is determined based on the spatial resolution of the second wind resource map, and the weight of the wind resource map is positively correlated with the spatial resolution of the wind resource map.

[0193] In some embodiments, the apparatus further includes:

[0194] A determination unit, configured to determine the accuracy of the fusion map based on the deviation between the wind resource parameters in the fusion map and the wind resource parameters detected at the wind measurement points; or determine the accuracy of the fusion map based on the deviation between the wind resource parameters in the fusion map and the wind resource parameters in the first wind resource map and the deviation between the wind resource parameters in the fusion map and the wind resource parameters in the second wind resource map.

[0195] In some embodiments, the feature fusion unit 630 is configured to fuse the first type of features, the second type of features of the first wind resource map, the second type of features of the second wind resource map, and the wind resource parameters detected at the wind measurement points to obtain fused features.

[0196] Figure 6 FIG. 7 is a schematic structural diagram of a computing device 800 provided by an embodiment of the present application. The computing device 800 includes a processor 801, and the processor 801 is coupled to a memory 802. At least one computer program instruction is stored in the memory 802, and the at least one computer program instruction is loaded and executed by the processor 801 so that the computing device 800 implements the above method.

[0197] Each embodiment in this specification is described in a progressive manner. Similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.

[0198] A refers to B, which means that A is the same as B or A is a simple deformation of B.

[0199] The terms "first" and "second" in the description and claims of the embodiments of the present application are used to distinguish different objects, rather than to describe a specific order of the objects, nor can they be understood as indicating or implying relative importance. For example, the first wind resource map and the second wind resource map are used to distinguish different wind resource maps, rather than to describe a specific order of the wind resource maps, nor can it be understood that the first wind resource map is more important than the second wind resource map.

[0200] The information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of the present application are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions. For example, the wind resource maps involved in the present application are all obtained under full authorization.

[0201] In the embodiments of the present application, unless otherwise specified, "at least one" means one or more, and "a plurality" means two or more. For example, a plurality of wind resource maps means two or more wind resource maps.

[0202] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state disk (SSD)).

[0203] The above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; 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 they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A wind resource assessment method, characterized in that, the method includes: performing feature extraction on a first wind resource map and a second wind resource map to obtain a first type of features, where the first wind resource map and the second wind resource map are wind resource maps with different spatial resolutions for the same wind farm; respectively performing image decomposition processing on the first wind resource map and the second wind resource map to obtain a second type of features of the first wind resource map and a second type of features of the second wind resource map; performing feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map; assessing the wind resource of the wind farm based on the fusion map.

2. The method according to claim 1, characterized in that, the first wind resource map and the second wind resource map include the same wind resource parameters, and / or, the first wind resource map and the second wind resource map are wind resource maps at the same wind speed height.

3. The method according to claim 1, characterized in that, the performing feature extraction on the first wind resource map and the second wind resource map to obtain a first type of features includes: inputting the first wind resource map and the second wind resource map into multiple convolutional layers in a deep learning model to obtain a feature map of the first wind resource map and a feature map of the second wind resource map; obtaining the first type of features based on the feature map through an activation function of the deep learning model, where the first type of features is used to represent the fusion weights of the first wind resource map and the second wind resource map.

4. The method according to claim 1, characterized in that, the second type of features includes residual images, and the respectively performing image decomposition processing on the first wind resource map and the second wind resource map to obtain a second type of features of the first wind resource map and a second type of features of the second wind resource map includes: respectively performing decomposition processing on the first wind resource map and the second wind resource map by using a pyramid to obtain N first pyramid maps with different scales corresponding to the first wind resource map and N second pyramid maps with different scales corresponding to the second wind resource map, where N is the number of layers of the pyramid and N≥2; for the N first pyramid maps with different scales, performing upsampling processing on the N - 1 layer first pyramid maps and performing differential processing with the N layer first pyramid maps to obtain the residual image corresponding to the first wind resource map; for the N second pyramid maps with different scales, performing upsampling processing on the N - 1 layer second pyramid maps and performing differential processing with the N layer second pyramid maps to obtain the residual image corresponding to the second wind resource map.

5. The method according to claim 1, characterized in that, the performing feature fusion processing on the first type of features, the second type of features of the first wind resource map, and the second type of features of the second wind resource map to obtain a fusion map includes: Determine the map fusion mode parameters based on the second - type features of the first wind resource map and the second - type features of the second wind resource map; Perform fusion processing using the fusion strategy corresponding to the map fusion mode parameters based on the first - type features, the second - type features of the first wind resource map, and the second - type features of the second wind resource map to obtain a fused map.

6. The method according to claim 5, wherein, The performing fusion processing using the fusion strategy corresponding to the map fusion mode parameters based on the first - type features, the second - type features of the first wind resource map, and the second - type features of the second wind resource map to obtain a fused map includes: In response to the map fusion mode parameters belonging to the first threshold range, perform weighted summation on the first - type features, the second - type features of the first wind resource map, and the second - type features of the second wind resource map based on the weight of the first wind resource map and the weight of the second wind resource map to obtain a fused map; In response to the map fusion mode parameters belonging to the second threshold range, perform inverse transformation on the second - type features of the first wind resource map to obtain a fused map; In response to the map fusion mode parameters belonging to the third threshold range, perform inverse transformation on the second - type features of the second wind resource map to obtain a fused map.

7. The method according to claim 6, wherein, The weight of the first wind resource map is determined based on the spatial resolution of the first wind resource map, the weight of the second wind resource map is determined based on the spatial resolution of the second wind resource map, and the weight of the wind resource map is positively correlated with the spatial resolution of the wind resource map.

8. The method according to any one of claims 1 - 6, wherein, After performing feature fusion processing on the first - type features, the second - type features of the first wind resource map, and the second - type features of the second wind resource map to obtain a fused map, the method further includes: Determine the accuracy of the fused map based on the deviation between the wind resource parameters in the fused map and the wind resource parameters detected at the wind measurement points.

9. The method according to any one of claims 1 - 6, wherein, After performing feature fusion processing on the first - type features, the second - type features of the first wind resource map, and the second - type features of the second wind resource map to obtain a fused map, the method further includes: Determine the accuracy of the fused map based on the deviation between the wind resource parameters in the fused map and the wind resource parameters in the first wind resource map and the deviation between the wind resource parameters in the fused map and the wind resource parameters in the second wind resource map.

10. The method according to any one of claims 1 - 6, wherein, The performing feature fusion processing on the first - type features, the second - type features of the first wind resource map, and the second - type features of the second wind resource map to obtain a fused map includes: Fuse the first type of features, the second type of features of the first wind resource map, the second type of features of the second wind resource map, and the wind resource parameters detected at the wind measurement points to obtain fused features.

11. A computing device, characterized in that the computing device includes: a processor, the processor is coupled to a memory, and at least one computer program instruction is stored in the memory, and the at least one computer program instruction is loaded and executed by the processor so that the computing device implements the method according to any one of claims 1-10.

12. A computer-readable storage medium, characterized in that at least one instruction is stored in the storage medium, and when the instruction runs on a computer, the computer executes the method according to any one of claims 1-10.

Images