Method and apparatus for producing land-surface shortwave radiation product, and method and apparatus for training long short-term memory network

Abstract

The present disclosure relates to the technical field of aerospace. Disclosed are a method and apparatus for producing a land-surface shortwave radiation product, and a method and apparatus for training a long short-term memory network. A specific implementation solution involves: extracting, from first land-surface shortwave radiation data collected by means of satellites, a first shortwave radiation data set and a second shortwave radiation data set that are aligned in terms of time; acquiring a photovoltaic shortwave radiation data set that is collected by means of ground photovoltaic radiometers and is aligned with the first shortwave radiation data set in terms of time; on the basis of satellite albedo products and land-surface land cover products, determining an albedo data set; on the basis of the first shortwave radiation data set, the second shortwave radiation data set, the albedo data set, the photovoltaic shortwave radiation data set, and coordinate vector points of the ground photovoltaic radiometers, obtaining an input shortwave radiation data sequence; inputting the input shortwave radiation data sequence into a pre-trained long short-term memory network, so as to obtain an initial land-surface shortwave radiation product; and on the basis of the initial land-surface shortwave radiation product, obtaining a target land-surface shortwave radiation product.

Description

Production of surface shortwave radiation products, methods and devices for training long short-term memory networks.

[0001] This application claims priority to Chinese Patent Application No. 202510855925.2, filed with the Chinese Patent Office on June 25, 2025, entitled "Method and Apparatus for Manufacturing Surface Shortwave Radiation Products and Training Long Short-Term Memory Networks", and the entire contents of the aforementioned patent application are incorporated herein by reference. Technical Field

[0002] This invention belongs to the field of aerospace technology, particularly satellite remote sensing technology, and specifically relates to a method and apparatus for producing surface shortwave radiation products, a method and apparatus for training long short-term memory networks, electronic equipment, and computer-readable storage media. Background Technology

[0003] Shortwave radiation from the Earth's surface is a crucial component of solar radiation, primarily encompassing ultraviolet, visible, and near-infrared radiation, with wavelengths ranging from 0.2 to 4 micrometers. The vast majority of solar energy is transmitted to Earth in the form of shortwave radiation, making it the primary source of energy received by the Earth's surface. Shortwave radiation is a key factor influencing changes in surface temperature, soil moisture evaporation, vegetation photosynthesis, and the Earth's surface energy balance, playing a fundamental role in natural processes and engineering applications.

[0004] In the field of photovoltaic (PV) power generation, the intensity of shortwave radiation energy directly determines the power generation efficiency of PV modules, and its spatiotemporal variability has a significant impact on power generation assessment and grid load forecasting. Especially with the integration of a high proportion of renewable energy sources, rapid changes in shortwave radiation can cause significant fluctuations in PV output, increasing the difficulty of grid regulation and even threatening grid operational safety. Therefore, obtaining accurate and stable shortwave radiation data is not only crucial for the design and operation of PV power plants but also of great significance for ensuring the stability and security of the power system.

[0005] Currently, surface shortwave radiation data is mainly acquired through ground-based equipment and satellite remote sensing. Ground-based equipment, such as radiometers, can provide high-precision real-time data, but their deployment costs are high and their coverage is limited. In contrast, satellite remote sensing has the advantage of wide coverage and can provide high-frequency radiation data, but its spatial resolution is lower. Therefore, existing surface radiation products present a contradiction between monitoring accuracy and wide-area coverage.

[0006] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this disclosure is to solve the technical problem that existing surface shortwave radiation products cannot achieve both spatiotemporal resolution, and to provide a method and apparatus for producing surface shortwave radiation products, a method and apparatus for training long short-term memory networks, an electronic device, and a computer-readable storage medium.

[0008] This disclosure provides a method for producing surface shortwave radiation products. The method includes: extracting a time-aligned first shortwave radiation dataset and a second shortwave radiation dataset from first surface shortwave radiation data acquired by a satellite. The first shortwave radiation dataset includes at least one first shortwave radiation data point, and the second shortwave radiation dataset includes at least one second shortwave radiation data point. The first time resolution of the first shortwave radiation data is greater than the second time resolution of the second shortwave radiation data, and the first spatial coverage of the first shortwave radiation data is smaller than the second spatial coverage of the second shortwave radiation data. The first surface shortwave radiation data has a first spatial resolution. The method also includes: acquiring a photovoltaic shortwave radiation dataset acquired by a ground-based photovoltaic radiometer and time-aligned with the first shortwave radiation dataset; and determining an albedo dataset with pixels having a second spatial resolution based on satellite albedo products and surface land cover products. The spatial resolution of the first shortwave radiation dataset is less than the land type spatial resolution of the land cover product, and the second spatial resolution is less than the land type spatial resolution but greater than the first spatial resolution. Based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground photovoltaic radiometer, an input shortwave radiation data sequence with a first temporal resolution and a second spatial resolution is obtained. The input shortwave radiation data sequence is then fed into a pre-trained long short-term memory network to obtain an initial land surface shortwave radiation product with a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage area. The long short-term memory network is used to predict future land surface shortwave radiation products that belong to the second spatial coverage area. Based on the initial land surface shortwave radiation product, a target land surface shortwave radiation product with a third spatial resolution is obtained, where the third spatial resolution is greater than the second spatial resolution.

[0009] The second aspect of this disclosure provides a method for training a Long Short-Term Memory (LSTM) network. The method includes: acquiring a data training set, which includes: a set of photovoltaic shortwave radiation values ​​aligned to minute time intervals, a set of surface reflectance values, a second set of shortwave radiation values, and a first set of shortwave radiation values. The first spatial coverage of the first shortwave radiation values ​​is smaller than the second spatial coverage of the second shortwave radiation values. The first shortwave radiation values, the second shortwave radiation values, the set of surface reflectance values, and the set of photovoltaic shortwave radiation values ​​all have a first temporal resolution and a second spatial resolution. Based on the data training set, multiple input sequences and output data pairs are constructed. Based on the input sequences and output data pairs, an initial LTM network is trained to obtain a trained LTM network. The LTM network is used to preset surface shortwave radiation products with a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage area at future time intervals.

[0010] This disclosure provides a surface shortwave radiation product production apparatus, comprising: an extraction unit configured to extract a time-aligned first shortwave radiation dataset and a second shortwave radiation dataset from first surface shortwave radiation data acquired by a satellite, wherein the first shortwave radiation dataset includes at least one first shortwave radiation data, the second shortwave radiation dataset includes at least one second shortwave radiation data, a first time resolution of the first shortwave radiation data is greater than a second time resolution of the second shortwave radiation data, and a first spatial coverage of the first shortwave radiation data is smaller than a second spatial coverage of the second shortwave radiation data, and the first surface shortwave radiation data has a first spatial resolution; an acquisition unit configured to acquire a photovoltaic shortwave radiation dataset time-aligned with the first shortwave radiation dataset acquired by a ground photovoltaic radiometer; and a determination unit configured to determine an albedo dataset of pixels with a second spatial resolution based on satellite albedo products and surface land cover products, wherein the satellite albedo products are satellite... The spatial resolution is smaller than the land type spatial resolution of the land cover product, and the second spatial resolution is smaller than the land type spatial resolution but larger than the first spatial resolution. The sequence obtaining unit is configured to obtain an input shortwave radiation data sequence with a first temporal resolution and a second spatial resolution based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground fiber optic radiometer. The input unit is configured to input the input shortwave radiation data sequence into a pre-trained long short-term memory network to obtain an initial land surface shortwave radiation product with a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage area. The long short-term memory network is used to predict land surface shortwave radiation products that belong to the second spatial coverage area at future times. The product obtaining unit is configured to obtain a target land surface shortwave radiation product with a third spatial resolution based on the initial land surface shortwave radiation product. The third spatial resolution is larger than the second spatial resolution.

[0011] This disclosure provides a long short-term memory (LSTM) network training device, comprising: a data acquisition unit configured to acquire a data training set, the data training set including: a photovoltaic shortwave radiation value set aligned to minute times, a land surface reflectance value set, a second shortwave radiation value set, and a first shortwave radiation value set, wherein the first spatial coverage of the first shortwave radiation value set is smaller than the second spatial coverage of the second shortwave radiation value set, and the first shortwave radiation value set, the second shortwave radiation value set, the land surface reflectance value set, and the photovoltaic shortwave radiation value set all have a first temporal resolution and a second spatial resolution; a construction unit configured to construct an input sequence and an output data pair based on the data training set; and a training unit configured to train an initial LTM network based on the input sequence and the output data pair to obtain a trained LTM network, wherein the LTM network is used to preset land surface shortwave radiation products with a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage range at future times.

[0012] The fifth aspect of this disclosure provides a computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method as described in the first or second aspect.

[0013] The sixth aspect of this disclosure provides a computer-readable storage medium in which a computer program, when executed by a processor, implements the steps of the method as described in the first or second aspect.

[0014] Compared with existing technologies, the technical effects achieved by this disclosure are as follows: By constructing a fusion framework of multi-satellite and ground radiometer data and employing a long short-term memory network, high temporal resolution data and data with second spatial coverage are effectively integrated. This overcomes the limitation of traditional single-satellite payloads being unable to simultaneously achieve spatiotemporal resolution, and for the first time, a standardized product output with global coverage and spatiotemporal resolution reaching the first temporal resolution / second spatial resolution is achieved. By introducing land cover data with land type spatial resolution into the albedo correction process, the reconstruction and refined correction of albedo at the sub-kilometer scale are realized, thereby significantly improving the spatial accuracy and resolution of surface shortwave radiation. Using the second spatial resolution albedo data as a benchmark, data completion is performed on satellite-covered areas that generate the first shortwave radiation dataset, generating a global shortwave radiation intermediate product with second spatial resolution. Subsequently, topographic radiation correction is performed using third spatial resolution data to generate a target surface shortwave radiation product with third spatial resolution, balancing radiation estimation accuracy with adaptability to complex terrain areas. Attached Figure Description

[0015] Figure 1 is a flowchart of an embodiment of a method for producing surface shortwave radiation products according to the present disclosure;

[0016] Figure 2 is a schematic diagram of the distribution of surface shortwave radiation values ​​of the surface shortwave radiation products of the first satellite disclosed herein.

[0017] Figure 3 is a schematic diagram of the distribution of surface shortwave radiation values ​​of the target surface shortwave radiation product of this disclosure.

[0018] Figure 4 is a flowchart of an embodiment of the long short-term memory network training method according to the present disclosure;

[0019] Figure 5 is a schematic diagram of a structure of an embodiment of a surface shortwave radiation product manufacturing apparatus according to the present disclosure;

[0020] Figure 6 is a schematic diagram of a structure of an embodiment of a long short-term memory network training device according to the present disclosure;

[0021] Figure 7 is a block diagram of an electronic device used to implement the method for producing surface shortwave radiation products or the method for training long short-term memory networks according to embodiments of the present disclosure. Detailed Implementation

[0022] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0023] The technical solution of the present invention is illustrated below through specific embodiments. It should be understood that the one or more steps mentioned in the present invention do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not for limiting the order of each method or limiting the scope of the present invention. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of the present invention.

[0024] The raw materials and instruments used in the examples are not subject to any specific restrictions on their source; they can be purchased from the market or prepared according to conventional methods known to those skilled in the art.

[0025] Currently, the main surface shortwave radiation products have relatively low spatiotemporal resolution, making them unsuitable for precise applications in scenarios such as photovoltaic power generation. This indicates that existing surface shortwave radiation products and estimations still have shortcomings. First, there is a trade-off between temporal and spatial resolution. Meteorological satellites can provide high-frequency radiation data, but their low spatial resolution makes it difficult to accurately reflect localized radiation changes, especially in photovoltaic power plant site selection and assessment where higher accuracy is required. Conversely, satellite products with higher spatial resolution have longer update cycles, making it difficult to meet the rapidly changing radiation monitoring needs. Furthermore, the surface albedo, a crucial parameter for surface shortwave radiation estimation, is too coarse. Traditional albedo estimation methods suffer from insufficient spatial heterogeneity in handling complex surface albedo variations, particularly in high-albedo areas (such as deserts and snow-covered areas) and complex terrain (such as plateaus and mountains), where surface albedo changes drastically, and existing technologies struggle to accurately capture its impact on radiation estimation, thus reducing the accuracy of surface shortwave radiation estimation. On the other hand, besides spatial resolution and albedo issues, existing technologies still do not adequately consider topographical factors. Topographic relief directly affects the incident angle and intensity of solar radiation, especially in mountainous and plateau regions, but existing radiation estimation methods generally fail to fully incorporate these topographic parameters.

[0026] Currently, combining ground monitoring and satellite remote sensing data is considered an important means to improve the accuracy of radiation estimation. However, it still faces many technical challenges in data fusion and correction. For example, the differences in temporal, spatial, and physical characteristics of multi-source data lead to complex fusion algorithms. Especially under complex weather and terrain conditions, how to effectively improve the spatiotemporal resolution and accuracy of radiation estimation remains a key research challenge. Overall, existing technologies still have considerable room for improvement in terms of insufficient spatial resolution, interference from meteorological and topographical factors, and a lack of multi-source data fusion strategies. Technological innovation is urgently needed to improve the accuracy and practicality of surface shortwave radiation estimation, providing more reliable data support for applications such as photovoltaic power generation, climate modeling, and agrometeorology.

[0027] To address the deficiencies in the prior art, this disclosure provides a method for producing surface shortwave radiation products. Figure 1 illustrates a flow chart 100 of an embodiment of the method for producing surface shortwave radiation products, which includes the following steps:

[0028] Step 101: Extract time-aligned first and second shortwave radiation datasets from the first surface shortwave radiation data acquired by the satellite.

[0029] In this embodiment, the first shortwave radiation dataset includes at least one first shortwave radiation data point, and the second shortwave radiation dataset includes at least one second shortwave radiation data point. The first time resolution of the first shortwave radiation data is greater than the second time resolution of the second shortwave radiation data, and the first spatial coverage area of ​​the first shortwave radiation data is smaller than the second spatial coverage area of ​​the second shortwave radiation data. The first surface shortwave radiation data has a first spatial resolution. The values ​​of the first and second time resolutions, as well as the first spatial resolution, can be set according to requirements. For example, the first time resolution is 10 minutes, the second time resolution is 15 minutes, and the first spatial resolution is 5 km. The first and second spatial coverage areas can be determined based on the coverage areas of different satellites. For example, the first spatial coverage area is East Asia, and the second spatial coverage area is the entire globe.

[0030] In this embodiment, the first surface shortwave radiation data can be minute-level surface shortwave radiation data. Surface shortwave radiation data refers to the shortwave portion of solar radiation received by the ground. Surface shortwave radiation data can be data recorded by satellite. The first surface shortwave radiation data contains at least two types of shortwave radiation data with different time resolutions, such as first shortwave radiation data and second shortwave radiation data. The spatial coverage of the first shortwave radiation data and the second shortwave radiation data are different.

[0031] In this embodiment, step 101 includes: acquiring first surface shortwave radiation data collected by satellite according to the required time period; extracting L2-level shortwave radiation datasets of the first satellite and the second satellite from the first surface shortwave radiation data; performing data preprocessing on the L2-level shortwave radiation datasets of the first satellite and the second satellite, wherein the preprocessing includes: converting nominal projection to latitude and longitude projection and converting nc format to tif format to obtain a first preprocessed dataset and a second preprocessed dataset; performing time matching on the first preprocessed dataset and the second preprocessed dataset to obtain a first shortwave radiation dataset and a second shortwave radiation dataset aligned to minute time.

[0032] In this embodiment, the first satellite has a first temporal resolution and a first spatial coverage, and the second satellite has a second temporal resolution and a second spatial coverage. By processing the L2-level shortwave radiation dataset of the first satellite, the surface shortwave radiation product of the first satellite is obtained. As shown in Figure 2, the distribution of surface shortwave radiation values ​​of the surface shortwave radiation product of the first satellite is as follows: the larger the surface shortwave radiation value of the surface shortwave radiation product, the lighter the color in the image; the smaller the surface shortwave radiation value, the darker the color in the image.

[0033] Step 102: Obtain the photovoltaic shortwave radiation dataset that is time-aligned with the first shortwave radiation dataset, collected by the ground photovoltaic radiometer.

[0034] In this embodiment, step 102 includes: selecting photovoltaic power stations with different landform types in different regions, obtaining minute-level surface shortwave radiation numerical data continuously monitored by ground photovoltaic radiometers configured on each photovoltaic power station, and aligning the time scale with the first shortwave radiation dataset and the interpolated and completed second shortwave radiation dataset to obtain a photovoltaic shortwave radiation dataset.

[0035] Step 103: Based on satellite albedo products and land cover products, determine the albedo dataset of pixels with a second spatial resolution.

[0036] In this embodiment, the spatial resolution of the satellite albedo product is lower than the land type spatial resolution of the land cover product. The second spatial resolution is lower than the land type spatial resolution but higher than the first spatial resolution. The land type spatial resolution is the spatial resolution of the land cover product, which varies depending on the specific land cover product, ranging from 10m to 1000m. In this embodiment, the land type spatial resolution can be 30m. The second spatial resolution can be set according to requirements; for example, the second spatial resolution is 500m, and the spatial resolution of the satellite albedo product is also 500m.

[0037] In this embodiment, step 103 includes: acquiring the albedo product of a satellite with a medium resolution imaging spectrometer, acquiring the land cover product, spatially superimposing the data of the two products, recalculating the albedo of each pixel of the medium resolution imaging spectrometer by the proportion of land cover type covered within the data pixel of each medium resolution imaging spectrometer, and obtaining the albedo dataset of pixels with a second spatial resolution.

[0038] Step 104: Based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground photovoltaic radiometer, an input shortwave radiation data sequence with a first temporal resolution and a second spatial resolution is obtained.

[0039] In this embodiment, the shortwave radiation data sequence is various types of data to be predicted, such as: a first shortwave radiation dataset after spatial resolution conversion, and a second shortwave radiation dataset after temporal and spatial resolution conversion. The shortwave radiation data sequence is input into a pre-trained long short-term memory network to obtain the initial surface shortwave radiation product output by the long short-term memory network.

[0040] In this embodiment, step 104 includes: resampling the first shortwave radiation dataset and the interpolated and completed second shortwave radiation dataset to a second spatial resolution (500 meters) to achieve spatial downscaling, reducing the spatial resolution of both from the first spatial resolution to the second spatial resolution, unifying them into the geographic coordinate system WGS84, and obtaining the resampled first shortwave radiation dataset; spatially superimposing the recalculated surface reflectance at the second spatial resolution (500 meters), and using a pixel area weighting method to register the interpolated and completed second shortwave radiation dataset and albedo dataset to the first shortwave radiation dataset, obtaining the registered second shortwave radiation dataset and the registered albedo dataset, ensuring that the data fusion is continuous and accurate.

[0041] In this embodiment, the pixel area weighting method refers to weighting the geographic data according to the actual ground area represented by each pixel to more accurately reflect the actual ground conditions. Due to the influence of geographic projection, pixels at different locations represent different actual ground areas; pixels near the equator have larger areas, while pixels near the North Pole have smaller areas. Directly averaging the data without area weighting would lead to inaccurate results.

[0042] In this embodiment, the coordinate vector points of the ground photovoltaic radiometer, the resampled first shortwave radiation dataset, the registered second shortwave radiation dataset, the registered albedo dataset, and the photovoltaic shortwave radiation dataset are spatially superimposed. The coordinate vector points of the ground photovoltaic radiometer are used to extract input shortwave radiation data sequences at a preset frequency (e.g., 10 minutes) using a point extraction method. The input shortwave radiation data sequences include: a first shortwave radiation data set, a second shortwave radiation data set, a surface reflectance data set, and a photovoltaic shortwave radiation data set. The first shortwave radiation data set is obtained by extracting the resampled first shortwave radiation dataset from the superimposed space at a preset frequency. The second shortwave radiation data set is obtained by extracting the registered second shortwave radiation dataset from the superimposed space at a preset frequency. The surface reflectance data set is obtained by extracting the registered albedo dataset from the superimposed space at a preset frequency. The photovoltaic shortwave radiation data set is obtained by extracting the photovoltaic shortwave radiation dataset from the superimposed space at a preset frequency.

[0043] Step 105: Input the input shortwave radiation data sequence into a pre-trained long short-term memory network to obtain an initial surface shortwave radiation product with a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage area.

[0044] In this embodiment, the Long Short-Term Memory Network is used to predict surface shortwave radiation products that are within the second spatial coverage area at future times.

[0045] In this embodiment, by combining the input shortwave radiation data sequence, the deep learning LSTM (Long Short Term Memory) network is used to predict the first shortwave radiation data belonging to the second spatial resolution, thereby completing the data of the uncovered areas in the first shortwave radiation data and forming an initial surface shortwave radiation product with the first temporal resolution and belonging to the second spatial coverage area.

[0046] Step 106: Based on the initial surface shortwave radiation product, obtain the target surface shortwave radiation product with third spatial resolution.

[0047] In this embodiment, the third spatial resolution is greater than the second spatial resolution. The third spatial resolution can be set based on development needs, for example, the third spatial resolution is 30m.

[0048] In this embodiment, step 106 includes: using a third spatial resolution Digital Elevation Model (DEM) to downscale and correct the terrain of the initial surface shortwave radiation product, thereby spatially downscaling and simultaneously correcting the terrain of the initial surface shortwave radiation product to obtain the target surface shortwave radiation product. Specifically, step 106 includes: using the DEM to calculate the slope, aspect, incident angle, sky visibility factor, and Earth-Sun distance correction factor for each pixel of the initial surface shortwave radiation product; and substituting the above calculation results into the equations for direct radiation and scattered radiation components in the radiation process according to the radiation transfer process to obtain the spatially downscaled and terrain-corrected target surface shortwave radiation product.

[0049] The method for producing surface shortwave radiation products disclosed herein includes: First, extracting a time-aligned first shortwave radiation dataset and a second shortwave radiation dataset from first surface shortwave radiation data acquired by a satellite. The first time resolution of the first shortwave radiation data is greater than the second time resolution of the second shortwave radiation data, and the first spatial coverage of the first shortwave radiation data is smaller than the second spatial coverage of the second shortwave radiation data. The first surface shortwave radiation data has a first spatial resolution. Second, acquiring a photovoltaic shortwave radiation dataset that is time-aligned with the first shortwave radiation dataset, acquired by a ground-based photovoltaic radiometer. Third, determining an albedo dataset with pixels having a second spatial resolution based on satellite albedo products and surface land cover products. The spatial resolution of the satellite albedo products is smaller than the land type spatial resolution of the surface land cover products. The second spatial resolution is less than the land type spatial resolution but greater than the first spatial resolution. Therefore, based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground photovoltaic radiometer, an input shortwave radiation data sequence with a first temporal resolution and a second spatial resolution is obtained. Then, the input shortwave radiation data sequence is fed into a pre-trained long short-term memory network to obtain an initial surface shortwave radiation product with a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage area. The long short-term memory network is used to predict future surface shortwave radiation products belonging to the second spatial coverage area. Finally, based on the initial surface shortwave radiation product, a target surface shortwave radiation product with a third spatial resolution is obtained, where the third spatial resolution is greater than the second spatial resolution. Therefore, by constructing a fusion framework of multi-satellite and ground radiometer data and employing a long short-term memory network, high temporal resolution data and second spatial coverage data are effectively integrated. Land cover data with land type spatial resolution is introduced into the albedo correction process, achieving sub-kilometer-scale albedo reconstruction and refined correction, thus significantly improving the spatial accuracy and resolution of surface shortwave radiation. Using the second spatial resolution albedo data as a benchmark, data completion is performed on satellite-covered areas that generated the first shortwave radiation dataset, generating a second spatial resolution global shortwave radiation intermediate product. Subsequently, topographic radiation correction is performed using third spatial resolution data to generate a target surface shortwave radiation product with third spatial resolution, balancing radiation estimation accuracy with adaptability to complex terrain areas. This disclosure establishes a scalable, high-precision, and high spatiotemporal resolution method for producing global surface shortwave radiation data, providing a reliable data foundation and technical support for applications such as solar energy resource assessment, climate modeling, and agro-meteorological analysis.

[0050] In some optional implementations of this disclosure, the extraction of time-aligned first and second shortwave radiation datasets from the first surface shortwave radiation data collected by satellite includes: extracting first surface shortwave radiation data from minute-level surface shortwave radiation data collected by satellite based on historical time periods; extracting first and second satellite datasets from the first surface shortwave radiation data; and performing time matching between the dataset with a first time resolution in the first satellite dataset and the dataset with a second time resolution in the second satellite dataset to obtain minute-time-aligned first and second shortwave radiation datasets.

[0051] In this optional implementation, the times of the first and second satellite datasets are aligned according to the principle of acquiring high-frequency data for data synthesis. The time series data of the second satellite dataset that does not correspond to the time series data periods of the first satellite dataset are then interpolated and completed using the mean interpolation method.

[0052] (1)

[0053] In equation (1), T is the time period of the first satellite dataset to be interpolated in the second satellite dataset, and t represents the median value within that time period. This represents the mean value of the pixels with row and column numbers r and c in time period T. t represents the pixel value of the first satellite dataset with row and column numbers r and c 15 minutes before and after time t, where t+15 represents the last 15 minutes and t-15 represents the first 15 minutes.

[0054] The optional implementation provides a method for obtaining first and second shortwave radiation data. First, it extracts first surface shortwave radiation data from surface shortwave radiation data. Then, it extracts first and second satellite data from the first surface shortwave radiation data. By time matching the first and second satellite data, it obtains the first and second shortwave radiation datasets, thus providing a reliable implementation for obtaining the first and second shortwave radiation datasets.

[0055] In some optional implementations of this disclosure, determining the albedo dataset of pixels with a second spatial resolution based on satellite albedo products and land cover products includes: acquiring satellite albedo products with a second spatial resolution and land cover products with land type spatial resolution; spatially overlaying the satellite albedo products and land cover products to calculate the proportion of each land cover type within each pixel; constructing a linear regression model based on the proportions; and using the linear regression model to spatially assign albedo to different land cover types to recalculate the albedo of each pixel, thereby obtaining the albedo dataset of pixels with a second spatial resolution.

[0056] In this optional implementation, the acquisition of satellite albedo products with second spatial resolution and land cover products with land type spatial resolution includes: acquiring satellite broadband directional albedo (Albedo_Broadband_BSA) products with a Moderate Resolution Imaging Spectroradiometer (MODIS), with the time scale aligned with the first shortwave radiometric data; simultaneously acquiring GlobeLand30 (short for Global Land Cover Dataset) land cover raster data; processing the two types of data into TIFF format; and unifying the coordinate system to the WGS84 geographic coordinate system.

[0057] In this optional implementation, the above-mentioned linear regression model based on proportion includes: based on proportion, for the satellite albedo product of each second spatial resolution pixel, constructing the following linear regression model:

[0058] (2)

[0059] In equation (2), It is the original MODIS albedo; It is the area proportion of the i-th land type in the pixels of the second spatial resolution (based on statistics for each land cover type). It is the albedo contribution factor corresponding to the i-th land type (solved by least squares regression).

[0060] In this optional implementation, the above-mentioned use of a linear regression model to assign albedo to different land cover types in space to recalculate the albedo of each cell and obtain an albedo dataset of cells with a second spatial resolution includes: recalculating the albedo of each cell using land cover data with a third spatial resolution to obtain an albedo dataset, as shown in Equation (3).

[0061] (3)

[0062] The optional implementation provides a method for obtaining an albedo dataset with second spatial resolution pixels. For the first time, it introduces land cover data with third spatial resolution into the albedo correction process. By calculating the proportion of different land types within each pixel, a linear regression model between land type distribution and albedo is constructed, realizing the reconstruction and fine correction of albedo at the sub-kilometer scale. This significantly improves the spatial accuracy and resolution of surface shortwave radiation and effectively solves the problem of insufficient spatial heterogeneity in traditional shortwave radiation estimation based on MODIS albedo products.

[0063] In some optional implementations of this disclosure, obtaining the input shortwave radiation data sequence with a first temporal resolution and a second spatial resolution based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, and the coordinate vector points of the ground photovoltaic radiometer includes: resampling the first and second shortwave radiation datasets so that the spatial resolution of both datasets is the second spatial resolution; registering the resampled second shortwave radiation dataset and the albedo dataset to the resampled first shortwave radiation dataset using a pixel area weighting method to obtain a registered second shortwave radiation dataset, a surface reflectance dataset, and a resampled first shortwave radiation dataset; spatially superimposing the registered second shortwave radiation dataset, the resampled first shortwave radiation dataset, the photovoltaic shortwave radiation dataset, and the surface reflectance dataset to obtain a superimposed dataset; and extracting the input shortwave radiation data sequence with the first temporal resolution from the superimposed dataset using the coordinate vector points of the ground photovoltaic radiometer through a point extraction value method.

[0064] In this optional implementation, the registration methods for the resampled second shortwave radiation dataset and the albedo dataset are the same. Taking the resampled second shortwave radiation dataset as an example, the pixel values ​​of the newly registered second shortwave radiation dataset under the pixels of the resampled first shortwave radiation dataset are... It can be calculated using the following formula (4):

[0065] (4)

[0066] In equation (4), i and j are the row and column numbers of the covered FY-4B shortwave radiometric data pixels, and n is the number of covered pixels. This represents the pixel value of the pixel in the i-th row and j-th column of the wind and cloud data. This represents the weight value of that pixel. It can be determined by the following formula (5):

[0067] (5)

[0068] In equation (5), For the second shortwave radiometric data after resampling, the pixel in the i-th row and j-th column occupies the area of ​​one pixel in the spatial resolution of the resampling first shortwave radiometric data that covers it. This represents the area of ​​a single pixel in the spatial resolution of the first shortwave radiometric data after resampling.

[0069] In this optional implementation, when extracting the input shortwave radiation data sequence using the point extraction value method, it can be done according to a preset frequency. The specific value of the preset frequency can be set as needed, such as a preset frequency of 10 minutes.

[0070] The optional implementation provides a method for obtaining the input shortwave radiation data sequence by resampling multiple datasets and registering the sampled datasets using a pixel area weighting method. After superimposing the datasets, the input shortwave radiation data sequence with a first time resolution is extracted from the superimposed datasets using the coordinate vector points of the ground photovoltaic radiometer through a point extraction value method. This provides a reliable means of obtaining the input shortwave radiation data sequence.

[0071] In some optional implementations of this disclosure, obtaining a target surface shortwave radiation product with a third spatial resolution based on the initial surface shortwave radiation product includes: resampling the second spatial resolution of the initial surface shortwave radiation product to match the data elevation model to obtain a processed surface shortwave radiation product; calculating the direct radiation component based on the data elevation model, the processed surface shortwave radiation product, and the expression for the direct radiation component in the surface shortwave radiation process; calculating the scattered radiation component based on the data elevation model, the processed surface shortwave radiation product, and the expression for the scattered radiation component in the surface shortwave radiation process; and summing the direct radiation component and the scattered radiation component to obtain the final surface shortwave radiation product with a third spatial resolution.

[0072] In this optional implementation, the above-mentioned resampling of the second spatial resolution of the initial surface shortwave radiation product to match the data elevation model to obtain the processed surface shortwave radiation product includes: using bilinear interpolation to the initial surface shortwave radiation product to the raster corresponding to the second spatial resolution, and overlaying it with the data elevation model having the second spatial resolution, and making the raster completely overlap through georegistration to obtain the processed surface shortwave radiation product.

[0073] In this optional implementation, the slope is calculated using a DEM. and slope Slope refers to the degree of inclination of the earth's surface at a point or area, usually expressed as an angle or percentage; aspect refers to the direction of the maximum slope of the earth's surface at a point or area, usually expressed as an angle, ranging from 0° to 360°. Slope and aspect are two very important parameters in topographic analysis, as they help to understand the undulation and directional characteristics of the terrain.

[0074] In this optional implementation, the solar zenith angle and solar azimuth angle are different at different geographical locations at the same time. The solar zenith angle on a large scale is calculated by using the following formulas (6) and (7). and solar azimuth .

[0075] (6)

[0076] In equation (6), Solar altitude angle δ: Geographic latitude of the observation point; δ: Solar declination angle (related to the date, which can be approximated by a formula). N is the yearly day), ω: hour angle (needs to be converted to the time difference between local time and noon - 15° per hour).

[0077] (7)

[0078] In equation (7), : Solar azimuth (usually 0° is due north, rotated clockwise to 360°); Solar altitude angle; : The geographical latitude and longitude of the observation point; Solar declination angle (related to date, can be approximated by a formula) (N represents the accumulated days in a year).

[0079] In this optional implementation method, the slope, aspect, and solar zenith angle are considered. and solar azimuth Sun angle, calculate the angle of incidence:

[0080] (8)

[0081] In equation (8), It is the angle of incidence. Solar zenith angle and solar altitude angle complementary For slope, The azimuth of the sun. It is a slope.

[0082] Sky visibility factor (SVF) for each pixel was calculated using DEM and line-of-sight analysis. Utilizing the DEM, solar azimuth, and solar elevation angles, the elevation from the pixel to be calculated to the sun was extracted, using the pixel as the origin. The elevation angles of the pixels and these points were calculated, and the maximum value was taken. Compared to the solar altitude angle, if If the solar altitude angle is less than that point, there is no nearby terrain obstruction at that point. =1), otherwise it will be blocked by the adjacent terrain ( =0).

[0083] Based on the Julian Day calculation, the Earth-Sun distance correction factor is calculated as follows:

[0084] (9)

[0085] In equation (9), N is the number of days in a year; based on the radiative transfer processes of direct radiation and scattered radiation, calculate the direct radiation component separately. ) and scattered radiation component ( ):

[0086] (10)

[0087] (11)

[0088] In equations (10) and (11), Angle of incidence The zenith angle of the sun. Terrain shading factor, SVF is the sky visibility factor. This is the solar shortwave radiation value before terrain correction. This is the Earth-Sun distance correction factor.

[0089] Finally, the shortwave radiation product of the target surface is obtained through equation (12). :

[0090] (12)

[0091] Figure 3 illustrates the distribution of surface shortwave radiometric values ​​in a target surface shortwave radiometric product. The third spatial resolution of the target surface shortwave radiometric product is greater than the first spatial resolution of the surface shortwave radiometric product of the first satellite shown in Figure 2. Furthermore, the second spatial coverage of the target surface shortwave radiometric product is greater than the first spatial coverage of the surface shortwave radiometric product of the first satellite shown in Figure 2. Both the target surface shortwave radiometric product and the surface shortwave radiometric product of the first satellite shown in Figure 2 have a first temporal resolution. In Figure 3, the larger the surface shortwave radiometric value, the lighter the color in the image; the smaller the surface shortwave radiometric value, the darker the color in the image.

[0092] The optional implementation method for obtaining target surface shortwave radiation products first resamples the initial surface shortwave radiation products, then calculates the direct radiation component and the scattered radiation component, and finally obtains the final surface shortwave radiation product through the direct radiation component and the scattered radiation component, providing a reliable implementation method for obtaining the final surface shortwave radiation product; by performing terrain correction based on DEM data, it effectively makes up for the shortcomings of traditional radiation products that ignore the influence of terrain, making the final product more physically consistent and geographically adaptable in complex terrain areas such as mountains and plateaus.

[0093] This disclosure provides a method for training a long short-term memory (LSTM) network. Figure 4 illustrates a flowchart 400 of an embodiment of the LSM network training method, which includes the following steps:

[0094] Step 401: Obtain the training data set.

[0095] In this embodiment, the data training set includes: a photovoltaic shortwave radiation value set aligned to minute time, a land surface reflectance value set, a second shortwave radiation value set, and a first shortwave radiation value set. The first spatial coverage of the first shortwave radiation value set is smaller than the second spatial coverage of the second shortwave radiation value set. The first shortwave radiation value set, the second shortwave radiation value set, the land surface reflectance value set, and the photovoltaic shortwave radiation value set all have a first temporal resolution and a second spatial resolution.

[0096] In this embodiment, the numerical set in the data training set can be a sequence obtained after collecting, resampling, and aligning the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, and the photovoltaic shortwave radiation dataset at a preset frequency. Specifically, the process of obtaining the data training set can refer to the process of obtaining the input shortwave radiation data sequence shown in Figure 1, which will not be repeated here.

[0097] Specifically, the training data set includes: , , as well as , i≥1. Where, Indicates the first A set of photovoltaic shortwave radiation data (time series) from a sample of photovoltaic meteorological stations. Indicates the first The second shortwave radiation numerical set (time series) of each sample; Indicates the first A set of surface albedo values ​​for a sample (time series). Indicates the first The first set of shortwave radiation values ​​(target values) for each sample.

[0098] Step 402: Based on the training data set, construct multiple input sequences and output data pairs.

[0099] In this embodiment, the above-mentioned training data set is combined to construct an input sequence, and an input-output pair is constructed using the sliding window method for initial training of the Long Short-Term Memory network. Specifically, the input sequence and output data pair include: an input feature sequence and a prediction target. The construction result of the input sequence and output data pair is as follows:

[0100] Input feature sequence: , , ;

[0101] Prediction target: ;

[0102] in To represent the length of the historical time window, the specific value of the historical time window can be set based on development requirements, for example... This indicates the previous hour, with a time interval of 10 minutes.

[0103] Step 403: Based on the input sequence and output data pairs, train the initial long short-term memory network to obtain the trained long short-term memory network.

[0104] In this embodiment, the Long Short-Term Memory Network is used to preset surface shortwave radiation products that have a first temporal resolution, a second spatial resolution, and belong to the second spatial coverage area at future times.

[0105] In this embodiment, the input features are normalized, missing test data is filled in, and the input sequence and output data are divided into a training set (e.g., accounting for 70% of all data), a validation set (e.g., accounting for 15% of all data), and a test set (e.g., accounting for 15% of all data) according to the region.

[0106] In this embodiment, when training the initial Long Short-Term Memory (LSTM) network, the model structure of the LSM network needs to be determined in advance, such as: input dimension input_dim: equal to the number of features (e.g., 3); hidden layer dimension hidden_dim=32; output dimension output_dim=1; LSTM layer number num_layers=2; different epochs (training rounds) and batch sizes can be set for each sample or region to accommodate differences in sample size. Mean squared error is used as the loss function, and the optimizer uses Adam.

[0107] In this embodiment, the weights and biases of the Long Short-Term Memory (LSTM) network are randomly initialized; the input data is fed into the LSTM step by step through forward propagation. At each time step, the LSTM calculates the hidden state and cell state for the current time step. Finally, the LSTM calculates the output based on the hidden state.

[0108] In this embodiment, during each iteration of LSTM training, a defined loss function is used to calculate the error between the model's predicted values ​​and the true values, i.e., the LSTM loss. The network weights and biases are updated by calculating the gradient of the loss function with respect to each parameter, and the gradient is propagated back to each layer of the network using the chain rule. Based on the calculated gradients, the network parameters are updated using an optimizer. The model performance is evaluated on a validation set, and hyperparameters (such as learning rate, number of hidden units, number of layers, etc.) are adjusted to optimize the model. The final performance of the model is evaluated on a test set to ensure that the Long Short-Term Memory network has good generalization ability.

[0109] The long short-term memory (LSTM) network training method disclosed herein first obtains a training data set; then, based on the training data set, it constructs multiple input sequences and output data pairs; finally, based on the input sequences and output data pairs, it trains the initial LSM network to obtain the trained LSM network. Thus, by combining the spatial coverage of multiple shortwave radiation values, the LSM network can be trained to have strong predictive performance, thereby improving the reliability and accuracy of LSM network training.

[0110] In some optional implementations of this disclosure, the acquisition of the data training set includes: extracting a first satellite dataset and a second satellite dataset from minute-level surface shortwave radiation data collected by satellites, wherein the surface shortwave radiation data has a first spatial resolution; performing time matching between the dataset with a first temporal resolution in the first satellite dataset and the dataset with a second temporal resolution in the second satellite dataset to obtain a first shortwave radiation dataset and a second shortwave radiation dataset aligned to minute time intervals; acquiring a photovoltaic shortwave radiation dataset collected by a ground-based photovoltaic radiometer aligned to minute time intervals of the first shortwave radiation dataset; determining an albedo dataset of pixels with a second spatial resolution based on satellite albedo products and surface land cover products, wherein the satellite spatial resolution of the satellite albedo products is less than the land type spatial resolution of the surface land cover products, and the second spatial resolution is less than the land type spatial resolution; and obtaining a photovoltaic shortwave radiation numerical set, a surface reflectance numerical set, a second shortwave radiation numerical set, and a first shortwave radiation numerical set aligned to minute time intervals based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground-based photovoltaic radiometer.

[0111] It should be noted that the photovoltaic shortwave radiation data set, the surface reflectance data set, the second shortwave radiation data set, and the first shortwave radiation data set constitute the input shortwave radiation data sequence. The acquisition of the input shortwave radiation data sequence is described in detail in the above embodiments and will not be repeated here.

[0112] Referring further to Figure 5, as an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a surface shortwave radiation product manufacturing apparatus, which corresponds to the method embodiment shown in Figure 1, and the apparatus can be specifically applied to various electronic devices.

[0113] In this embodiment, the specific processing of the extraction unit 501, photovoltaic acquisition unit 502, determination unit 503, sequence acquisition unit 504, input unit 505, and product acquisition unit 506 in the surface shortwave radiation product production device 500, and the resulting technical effects, can be referred to the relevant descriptions of steps 101, 102, 103, 104, 105, and 106 in the corresponding embodiment of Figure 1, which will not be repeated here.

[0114] In some embodiments of this disclosure, the extraction unit 501 is configured to: extract first surface shortwave radiation data from minute-level surface shortwave radiation data collected by satellite based on historical time periods; extract a first satellite dataset and a second satellite dataset from the first surface shortwave radiation data; and perform time matching between the dataset with a first time resolution in the first satellite dataset and the dataset with a second time resolution in the second satellite dataset to obtain a first shortwave radiation dataset and a second shortwave radiation dataset aligned to minute time intervals.

[0115] In some embodiments of this disclosure, the determining unit 503 is configured to: acquire satellite albedo products with a second spatial resolution and land cover products with land type spatial resolution; spatially overlay the satellite albedo products and land cover products to calculate the proportion of each land cover type within each pixel; construct a linear regression model based on the proportion; and use the linear regression model to spatially assign albedo to different land cover types to recalculate the albedo of each pixel, thereby obtaining an albedo dataset of pixels with a second spatial resolution.

[0116] In some embodiments of this disclosure, the sequence obtaining unit 504 can be configured to: resample the first shortwave radiation dataset and the second shortwave radiation dataset so that the spatial resolution of both the first and second shortwave radiation datasets is a second spatial resolution; register the resampled second shortwave radiation dataset and the albedo dataset to the resampled first shortwave radiation dataset using a pixel area weighting method, to obtain a registered second shortwave radiation dataset, a surface reflectance dataset, and a resampled first shortwave radiation dataset; spatially superimpose the registered second shortwave radiation dataset, the resampled first shortwave radiation dataset, the photovoltaic shortwave radiation dataset, and the surface reflectance dataset to obtain a superimposed dataset; and extract the input shortwave radiation data sequence with a first temporal resolution from the superimposed dataset using the coordinate vector points of the ground photovoltaic radiometer through a point extraction value method.

[0117] In some embodiments of this disclosure, the product obtaining unit 506 is configured to: resample the initial surface shortwave radiation product to a second spatial resolution that matches the data elevation model to obtain a processed surface shortwave radiation product; calculate the direct radiation component based on the data elevation model, the processed surface shortwave radiation product, and the expression for the direct radiation component in the surface shortwave radiation process; calculate the scattered radiation component based on the data elevation model, the processed surface shortwave radiation product, and the expression for the scattered radiation component in the surface shortwave radiation process; and sum the direct radiation component and the scattered radiation component to obtain the final surface shortwave radiation product P3 belonging to the third spatial resolution.

[0118] The surface shortwave radiation product production apparatus provided in this disclosure, by constructing a fusion framework of multi-satellite and ground radiometer data and employing a long short-term memory network, effectively integrates high temporal resolution data with data from the second spatial coverage area. It overcomes the limitation of traditional single-satellite payloads in simultaneously achieving spatiotemporal resolution, and for the first time achieves standardized product output with global coverage and spatiotemporal resolution reaching the first temporal resolution / second spatial resolution. By introducing land cover data with land type spatial resolution into the albedo correction process, it achieves sub-kilometer-scale albedo reconstruction and refined correction, thereby significantly improving the spatial accuracy and resolution of surface shortwave radiation. Using the second spatial resolution albedo data as a benchmark, it completes the data for satellite-covered areas in the first shortwave radiation dataset, generating a global shortwave radiation intermediate product with second spatial resolution. Subsequently, it combines third spatial resolution data for topographic radiation correction, generating a target surface shortwave radiation product with third spatial resolution, balancing radiation estimation accuracy with adaptability to complex terrain areas.

[0119] Referring further to FIG6, as an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a long short-term memory network training device, which is in an electronic device corresponding to the method embodiment shown in FIG2.

[0120] In this embodiment, the specific processing of the data acquisition unit 601, the construction unit 602, and the training unit 603 in the long short-term memory network training device 600 and the resulting technical effects can be referred to the relevant descriptions of steps 201, 202, and 203 in the corresponding embodiment of Figure 2, which will not be repeated here.

[0121] In some embodiments of this disclosure, the data acquisition unit 601 is configured to: extract a first satellite dataset and a second satellite dataset from minute-level surface shortwave radiation data collected by satellites, wherein the surface shortwave radiation data has a first spatial resolution; perform time matching between the dataset with a first temporal resolution in the first satellite dataset and the dataset with a second temporal resolution in the second satellite dataset to obtain a first shortwave radiation dataset and a second shortwave radiation dataset aligned to minute time; acquire a photovoltaic shortwave radiation dataset aligned to minute time of the first shortwave radiation dataset collected by a ground photovoltaic radiometer; determine the albedo dataset of pixels with a second spatial resolution based on satellite albedo products and surface land cover products, wherein the satellite spatial resolution of the satellite albedo products is less than the land type spatial resolution of the surface land cover products, and the second spatial resolution is less than the land type spatial resolution; and obtain a photovoltaic shortwave radiation numerical set, a surface reflectance numerical set, a second shortwave radiation numerical set, and a first shortwave radiation numerical set aligned to minute time based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground photovoltaic radiometer.

[0122] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0123] As shown in Figure 7, the electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 702 or a computer program loaded into a random access memory (RAM) 703 from a storage unit 708. The RAM 703 can also store various programs and data required for the operation of the electronic device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0124] Multiple components in electronic device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of displays, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0125] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as methods for producing surface shortwave radiation products or methods for training long short-term memory networks. For example, in some embodiments, the methods for producing surface shortwave radiation products or methods for training long short-term memory networks can be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the methods for producing surface shortwave radiation products or methods for training long short-term memory networks described above can be performed. Alternatively, in other embodiments, computing unit 701 may be configured by any other suitable means (e.g., by means of firmware) to perform a method for producing surface shortwave radiation products or a method for training long short-term memory networks.

[0126] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0127] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to the processor or controller of a general-purpose computer, special-purpose computer, or other programmable surface shortwave radiation product manufacturing apparatus or long short-term memory network training apparatus, such that when executed by the processor or controller, the program code causes the patterns / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, partially on a remote machine as a standalone software package, or entirely on a remote machine or server.

[0128] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0129] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0130] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0131] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0132] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for producing surface shortwave radiation products, characterized in that, The method includes: A time-aligned first shortwave radiation dataset and a second shortwave radiation dataset are extracted from first surface shortwave radiation data acquired by satellite. The first shortwave radiation dataset includes at least one first shortwave radiation data, and the second shortwave radiation dataset includes at least one second shortwave radiation data. The first time resolution of the first shortwave radiation data is greater than the second time resolution of the second shortwave radiation data, and the first spatial coverage of the first shortwave radiation data is smaller than the second spatial coverage of the second shortwave radiation data. The first surface shortwave radiation data has a first spatial resolution. Acquire a photovoltaic shortwave radiation dataset that is time-aligned with the first shortwave radiation dataset, collected by a ground-based photovoltaic radiometer; Based on satellite albedo products and land cover products, an albedo dataset with a second spatial resolution is determined, wherein the spatial resolution of the satellite albedo products is less than the land type spatial resolution of the land cover products, and the second spatial resolution is less than the land type spatial resolution but greater than the first spatial resolution. Based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground photovoltaic radiometer, an input shortwave radiation data sequence with the first temporal resolution and the second spatial resolution is obtained. The input shortwave radiation data sequence is input into a pre-trained long short-term memory network to obtain an initial surface shortwave radiation product with the first temporal resolution, the second spatial resolution and belonging to the second spatial coverage area. The long short-term memory network is used to predict surface shortwave radiation products that belong to the second spatial coverage area at future times. Based on the initial surface shortwave radiation product, a target surface shortwave radiation product with a third spatial resolution is obtained, wherein the third spatial resolution is greater than the second spatial resolution.

2. The method according to claim 1, characterized in that, The extraction of time-aligned first and second shortwave radiation datasets from the first surface shortwave radiation data acquired by satellite includes: Based on historical time periods, the first surface shortwave radiation data is extracted from minute-level surface shortwave radiation data collected by satellite; Extract the first satellite dataset and the second satellite dataset from the first surface shortwave radiation data; The dataset with the first time resolution in the first satellite dataset is time-matched with the dataset with the second time resolution in the second satellite dataset to obtain a first shortwave radiation dataset and a second shortwave radiation dataset aligned to minute time.

3. The method according to claim 1, characterized in that, The albedo dataset, which determines pixels with a second spatial resolution based on satellite albedo products and land cover products, includes: Obtain satellite albedo products with second spatial resolution and land cover products with land type spatial resolution; The satellite albedo product and the land cover product are spatially overlaid to calculate the proportion of each land cover type in each pixel; Based on the stated proportions, a linear regression model is constructed; The albedo is spatially assigned to different land cover types using the linear regression model to recalculate the albedo of each cell, resulting in an albedo dataset of cells with the second spatial resolution.

4. The method according to claim 1, characterized in that, The step of obtaining an input shortwave radiation data sequence with the first temporal resolution and the second spatial resolution based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, and the coordinate vector points of the ground photovoltaic radiometer includes: The first shortwave radiation dataset and the second shortwave radiation dataset are resampled so that the spatial resolution of both the first shortwave radiation dataset and the second shortwave radiation dataset is the second spatial resolution. The second shortwave radiation dataset and the albedo dataset are registered to the first shortwave radiation dataset after resampling using the cell area weighting method, resulting in the registered second shortwave radiation dataset, surface reflectance dataset, and the first shortwave radiation dataset after resampling. The registered second shortwave radiation dataset, the resampled first shortwave radiation dataset, the photovoltaic shortwave radiation dataset, and the surface reflectance dataset are spatially superimposed to obtain the superimposed dataset. The input shortwave radiation data sequence with a first time resolution is extracted from the superimposed dataset using the point extraction value method based on the coordinate vector points of the ground photovoltaic radiometer.

5. The method according to claim 1, characterized in that, The process of obtaining a target surface shortwave radiation product with a third spatial resolution based on the initial surface shortwave radiation product includes: The second spatial resolution of the initial surface shortwave radiation product is resampled to match the data elevation model to obtain the processed surface shortwave radiation product. Based on the data elevation model, the processed surface shortwave radiation product, and the expression for the direct radiation component in the surface shortwave radiation process, the direct radiation component is calculated. Based on the data elevation model, the processed surface shortwave radiation product, and the expression for scattered radiation during the surface shortwave radiation process, the scattered radiation component is calculated. The direct radiation component and the scattered radiation component are summed to obtain the final product of surface shortwave radiation with the third spatial resolution.

6. A method for training long short-term memory networks, characterized in that, The method includes: A data training set is obtained, which includes: a photovoltaic shortwave radiation value set aligned to minute time, a land surface reflectance value set, a second shortwave radiation value set, and a first shortwave radiation value set. The first spatial coverage range of the first shortwave radiation value set is smaller than the second spatial coverage range of the second shortwave radiation value set. The first shortwave radiation value set, the second shortwave radiation value set, the land surface reflectance value set, and the photovoltaic shortwave radiation value set all have a first temporal resolution and a second spatial resolution. Based on the aforementioned training data set, multiple input sequences and output data pairs are constructed; Based on the input sequence and output data pair, an initial long short-term memory network is trained to obtain a trained long short-term memory network. The long short-term memory network is used to preset surface shortwave radiation products with the first temporal resolution, the second spatial resolution, and belonging to the second spatial coverage area at future times.

7. The method according to claim 6, characterized in that, The acquired training data set includes: Extract a first satellite dataset and a second satellite dataset from minute-level surface shortwave radiation data collected by satellites, wherein the surface shortwave radiation data has a first spatial resolution; The dataset with the first time resolution in the first satellite dataset is time-matched with the dataset with the second time resolution in the second satellite dataset to obtain the first shortwave radiation dataset and the second shortwave radiation dataset aligned to minute time. Obtain the photovoltaic shortwave radiation dataset collected by the ground photovoltaic radiometer, which is aligned with the minute time of the first shortwave radiation dataset; Based on satellite albedo products and land cover products, an albedo dataset of pixels with a second spatial resolution is determined, wherein the satellite spatial resolution of the satellite albedo products is less than the land type spatial resolution of the land cover products, and the second spatial resolution is less than the land type spatial resolution. Based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground photovoltaic radiometer, a photovoltaic shortwave radiation data set, a surface reflectance data set, a second shortwave radiation data set, and a first shortwave radiation data set aligned to minute time intervals are obtained.

8. A device for producing surface shortwave radiation products, characterized in that, The device includes: The extraction unit is configured to extract a time-aligned first shortwave radiation dataset and a second shortwave radiation dataset from first surface shortwave radiation data acquired by a satellite. The first shortwave radiation dataset includes at least one first shortwave radiation data, and the second shortwave radiation dataset includes at least one second shortwave radiation data. The first time resolution of the first shortwave radiation data is greater than the second time resolution of the second shortwave radiation data, and the first spatial coverage of the first shortwave radiation data is smaller than the second spatial coverage of the second shortwave radiation data. The first surface shortwave radiation data has a first spatial resolution. The photovoltaic acquisition unit is configured to acquire a photovoltaic shortwave radiation dataset that is time-aligned with the first shortwave radiation dataset, collected by a ground-based photovoltaic radiometer. The determining unit is configured to determine an albedo dataset of pixels with a second spatial resolution based on satellite albedo products and land cover products, wherein the satellite spatial resolution of the satellite albedo products is less than the land type spatial resolution of the land cover products, and the second spatial resolution is less than the land type spatial resolution but greater than the first spatial resolution. The sequence obtaining unit is configured to obtain an input shortwave radiation data sequence with the first temporal resolution and the second spatial resolution based on the first shortwave radiation dataset, the second shortwave radiation dataset, the albedo dataset, the photovoltaic shortwave radiation dataset, and the coordinate vector points of the ground fiber optic radiometer. The input unit is configured to input the input shortwave radiation data sequence into a pre-trained long short-term memory network to obtain an initial surface shortwave radiation product having a first temporal resolution, a second spatial resolution, and belonging to the second spatial coverage area. The long short-term memory network is used to predict surface shortwave radiation products that belong to the second spatial coverage area at future times. The product obtaining unit is configured to obtain a target surface shortwave radiation product with a third spatial resolution, which is greater than the second spatial resolution, based on the initial surface shortwave radiation product.

9. A long short-term memory network training device, characterized in that, The device includes: The data acquisition unit is configured to acquire a data training set, which includes: a photovoltaic shortwave radiation value set aligned to minute time, a land surface reflectance value set, a second shortwave radiation value set, and a first shortwave radiation value set. The first spatial coverage range of the first shortwave radiation value set is smaller than the second spatial coverage range of the second shortwave radiation value set. The first shortwave radiation value set, the second shortwave radiation value set, the land surface reflectance value set, and the photovoltaic shortwave radiation value set all have a first temporal resolution and a second spatial resolution. The building unit is configured to construct input sequence and output data pairs based on the data training set; The training unit is configured to train an initial long short-term memory network based on the input sequence and output data pair to obtain a trained long short-term memory network. The long short-term memory network is used to preset surface shortwave radiation products with the first temporal resolution, the second spatial resolution, and belonging to the second spatial coverage area at future times.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images