Earth surface solar radiation degree prediction method and device, and electronic equipment

Abstract

The invention relates to the technical field of solar radiation degree prediction, and discloses a prediction method and device for earth surface solar radiation degree, and electronic equipment. The prediction method comprises the following steps: acquiring numerical weather forecast, ground meteorological observation data and a satellite channel cloud picture, and constructing a three-channel pseudo-color cloud picture based on cloud top temperature; constructing a layered cloud image set with time dimension information according to the three-channel pseudo-color cloud image; a pre-designed U-Net network structure is adopted to extract multi-scale features of the layered cloud image set, and a global satellite feature map is constructed; and intercepting a local feature map corresponding to the target photovoltaic power station from the global satellite feature map, and inputting the local feature map into a pre-trained photovoltaic prediction model to predict the earth surface solar radiation degree of the target photovoltaic power station. According to the invention, the prediction precision of the earth surface solar radiation degree can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solar radiation prediction, for example, to a method and device for predicting ground surface solar radiation, and an electronic device. BACKGROUND

[0002] Ground surface solar radiation is a core parameter for evaluating solar energy resource potential and optimizing photovoltaic system design. Therefore, accurately predicting ground surface solar radiation is of great significance to the work of photovoltaic power stations.

[0003] In the related art, numerical weather prediction or single-layer stationary orbit cloud images of a photovoltaic power station are first collected, and then the numerical weather prediction or single-layer stationary orbit cloud images are input into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the photovoltaic power station in the short term.

[0004] In the process of implementing the embodiments of the present disclosure, it is found that the related art at least has the following problems:

[0005] The scheme of the related art has a large error between the predicted ground surface solar radiation and the actual solar radiation when facing relatively complex weather conditions (for example, cloudy weather). Therefore, how to improve the prediction accuracy of the ground surface solar radiation has become a technical problem to be solved. SUMMARY

[0006] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an overall description of the application, nor is it intended to identify key / important elements or delineate the scope of the embodiments. It is only a prelude to the detailed description that follows.

[0007] The embodiments of the present disclosure provide a method and device for predicting ground surface solar radiation, and an electronic device, which can improve the prediction accuracy of the ground surface solar radiation.

[0008] In some embodiments, the method for predicting ground surface solar radiation comprises: acquiring numerical weather prediction, ground meteorological observation data and satellite channel cloud images, and constructing a three-channel pseudo-color cloud image based on cloud top temperature; constructing a layered cloud image set with time dimension information according to the three-channel pseudo-color cloud image; extracting multi-scale features of the layered cloud image set using a pre-designed U-Net network structure, and constructing a global satellite feature map; intercepting a local feature map corresponding to a target photovoltaic power station from the global satellite feature map, and inputting the local feature map into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station.

[0009] Optionally, the numerical weather prediction, the ground meteorological observation data and the satellite channel cloud image are acquired, and a three-channel pseudo-color cloud image based on cloud top temperature is constructed, including: interpolating the numerical weather prediction, the ground meteorological station data and the satellite channel cloud image to a set time frequency; mapping the numerical weather prediction and the ground meteorological station data to a spatial grid scale corresponding to the satellite channel cloud image; splicing the three-channel pseudo-color data of the satellite channel cloud image with the numerical weather prediction and the ground meteorological station data to obtain the three-channel pseudo-color cloud image based on the cloud top temperature. Figure One

[0010] Optionally, according to the three-channel pseudo-color cloud image, a hierarchical cloud image set with time dimension information is constructed, including: continuously acquiring multiple frames of satellite TBB channel images from the three-channel pseudo-color cloud image according to a set time interval, constructing a time sliding window, and determining a brightness temperature change sequence; performing histogram division and threshold cutting on each frame of satellite TBB channel image according to the brightness temperature change sequence, and extracting a brightness temperature distribution area representing a cloud body at different height levels; stacking the images formed by the brightness temperature distribution areas in time order to construct an image sequence tensor, and obtaining the hierarchical cloud image set.

[0011] Optionally, after the image sequence tensor is constructed, the prediction method further includes: pre-processing the image sequence tensor in a differential filtering and local standard deviation enhancement manner.

[0012] Optionally, the pre-designed U-Net network structure includes an up-sampling block, a down-sampling block and an FPN module; a multi-scale feature of the hierarchical cloud image set is extracted by using the pre-designed U-Net network structure, and a global satellite feature map is constructed, including: extracting a multi-scale feature from the hierarchical cloud image set by using the up-sampling block and the down-sampling block, and obtaining multiple feature maps of different scales; pre-processing the multiple feature maps of different scales by using the FPN module; and connecting the multiple feature maps of different scales after pre-processing by using 2D convolution operation, and obtaining the global satellite feature map.

[0013] Optionally, the FPN module is used to pre-process the multiple feature maps of different scales, including: determining a feature map of the largest scale in the multiple feature maps of different scales; performing 2D convolution operation and jump connection on the feature map of the largest scale; performing 2D convolution operation and spatial pyramid pooling on the feature maps other than the feature map of the largest scale in the multiple feature maps, and enlarging the feature maps other than the feature map of the largest scale in the multiple feature maps to the same scale as the largest scale by using a bilinear interpolation algorithm.

[0014] ​Optionally, the local feature map is input into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station, including: performing gray standardization and scale resampling processing on the local feature map; inputting the processed local feature map into a convolutional layer network of the photovoltaic prediction model, and extracting texture structure features and brightness distribution patterns of the local feature map layer by layer; inputting the texture structure features and the brightness distribution patterns into a fully connected network of the photovoltaic prediction model, and determining the ground surface solar radiation of the target photovoltaic power station.

[0015] Optionally, after predicting the ground surface solar radiation of the target photovoltaic power station according to the local feature map, the prediction method further includes: using a Dual TV-L1 optical flow algorithm to calculate the flow field between continuous cloud images in the hierarchical cloud image set, and performing short-time cloud extrapolation on the hierarchical cloud image set; using a multi-layer residual network to extract features of different cloud layers in the hierarchical cloud image set after cloud extrapolation, calculating the contribution of each cloud layer to the ground surface total radiation through a dynamic weight mechanism, and correcting the predicted ground surface solar radiation.

[0016] In some embodiments, the prediction device for ground surface solar radiation includes: a first construction module configured to acquire numerical weather prediction, ground meteorological observation data and satellite channel cloud images, and construct a three-channel pseudo-color cloud image based on cloud top temperature; a second construction module configured to construct a hierarchical cloud image set with time dimension information according to the three-channel pseudo-color cloud image; a third construction module configured to extract multi-scale features of the hierarchical cloud image set using a pre-designed U-Net network structure, and construct a global satellite feature map; and a prediction module configured to cut a local feature map corresponding to the target photovoltaic power station from the global satellite feature map, and input the local feature map into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station.

[0017] In some embodiments, the prediction device for ground surface solar radiation includes a processor and a memory storing program instructions, and the processor is configured to be able to execute the prediction method for ground surface solar radiation as described above.

[0018] In some embodiments, the electronic device includes: a device body; and the prediction device for ground surface solar radiation as described above, which is arranged in the device body.

[0019] The prediction method and device for ground surface solar radiation, and the electronic device provided by the embodiments of the present disclosure can achieve the following technical effects:

[0020] The disclosure embodiment finds that, in the related art, when facing relatively complex weather conditions, the reason why the predicted ground surface solar radiation degree has a large error with the actual solar radiation degree is that the data source relied on is single, and when the input data source is a track cloud picture, the detailed information in the track cloud picture cannot be well extracted, and the time and space information of different height levels in the cloud picture is ignored. In the disclosure embodiment, when predicting the ground surface solar radiation degree of a target photovoltaic power station, by obtaining numerical weather prediction, ground meteorological observation data and satellite channel cloud Figure Three A data source, a three-channel pseudo-color cloud picture based on cloud top temperature is constructed, which can form a complete meteorological information chain and avoid the limitation of a single data source. After the three-channel pseudo-color cloud picture is constructed, a layered cloud picture set with time dimension information is constructed according to the three-channel pseudo-color cloud picture. The vertical structure information of the cloud layer can be retained, and the attenuation effect of different height cloud layers on solar radiation can be distinguished. By using a pre-designed U-Net network structure to extract the multi-scale features of the layered cloud picture set, a global satellite feature map is constructed. With the help of the context information in the whole picture (i.e. in the global satellite feature map), the influence of cloud clusters and various weather factors on the ground surface solar radiation degree in the time dimension and the space dimension can be determined. After the global satellite feature map is constructed, the local feature map corresponding to the target photovoltaic power station is intercepted from the global satellite feature map, and then the local feature map is input into the photovoltaic prediction model to predict the ground surface solar radiation degree of the target photovoltaic power station, which can ensure that sufficient context information is referred to and irrelevant regional noise interference is avoided when predicting the ground surface solar radiation degree of the target photovoltaic power station. Therefore, even when facing relatively complex weather conditions, the disclosure embodiment can accurately predict the ground surface solar radiation degree of the target photovoltaic power station, and the prediction accuracy of the ground surface solar radiation degree is improved.

[0021] The foregoing general description and the following description are only exemplary and explanatory, and are not used to limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0022] One or more embodiments are exemplarily illustrated by corresponding drawings, which do not constitute limitation on the embodiments, elements with the same reference numerals in the drawings are shown as similar elements, the drawings do not constitute proportional limitation, and wherein:

[0023] Figure 1 is a schematic diagram of an electronic device provided by the disclosure embodiment;

[0024] Figure 2 is a schematic diagram of a prediction method for ground surface solar radiation degree provided by the disclosure embodiment;

[0025] Figure 3 is a schematic diagram of a U-Net network structure provided by the disclosure embodiment;

[0026] Figure 4 is another schematic diagram of a method for predicting ground surface solar radiation provided by an embodiment of the present disclosure;

[0027] Figure 5 is a schematic diagram of a device for predicting ground surface solar radiation provided by an embodiment of the present disclosure;

[0028] Figure 6 is another schematic diagram of a device for predicting ground surface solar radiation provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] In order to enable persons skilled in the art to more fully understand the features and technical contents of the embodiments of the present disclosure, the implementation of the embodiments of the present disclosure is described in detail below with reference to the accompanying drawings, which are used only for reference and are not intended to limit the embodiments of the present disclosure. In the following technical description, in order to facilitate explanation, a plurality of details are provided to provide a full understanding of the disclosed embodiments. However, one or more embodiments can still be implemented without these details. In other cases, well-known structures and devices can be simplified to facilitate the drawings.

[0030] The terms "first", "second", and the like in the specification and claims of the embodiments of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion.

[0031] Unless otherwise specified, the term "a plurality of" features means two or more.

[0032] In the embodiments of the present disclosure, the character " / " features represent a "or" relationship between the objects before and after. For example, A / B features represent: A or B.

[0033] The term "and / or" is a description of the association relationship between the objects, and the features represent that there can be three relationships. For example, A and / or B, the features represent: A or B, or, A and B, the three relationships.

[0034] The term "corresponding" can refer to an association relationship or a binding relationship, A corresponding to B means that there is an association relationship or a binding relationship between A and B.

[0035] It should be noted that the embodiments in the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0036] As Figure 1As shown, the electronic device 100 provided by the embodiment of the present disclosure includes a device body 110 and a prediction device 500 (600) for ground surface solar radiation.

[0037] Specifically, the prediction device 500 (600) for ground surface solar radiation is arranged on the device body 110.

[0038] Optionally, the prediction device 600 for ground surface solar radiation includes a processor. When predicting the ground surface solar radiation of a target photovoltaic power station, the processor can acquire numerical weather prediction, ground meteorological observation data and satellite channel cloud images, and construct a three-channel pseudo-color cloud image based on cloud top temperature. After the three-channel pseudo-color cloud image is constructed, the processor can construct a hierarchical cloud image set with time dimension information according to the three-channel pseudo-color cloud image, and can extract multi-scale features of the hierarchical cloud image set by using a pre-designed U-Net network structure to construct a global satellite feature map. After the global satellite feature map is constructed, the processor can intercept a local feature map corresponding to the target photovoltaic power station from the global satellite feature map, and input the local feature map into a photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station.

[0039] In combination with the above-mentioned electronic device, the embodiment of the present disclosure provides a prediction method for ground surface solar radiation. The execution subject of the prediction method can be the processor as described above, such as Figure 2 As shown, the prediction method includes:

[0040] S201, the processor acquires numerical weather prediction, ground meteorological observation data and satellite channel cloud images, and constructs a three-channel pseudo-color cloud image based on cloud top temperature.

[0041] Specifically, the numerical weather prediction (NWP, Numerical Weather Prediction) is cloud coverage, water vapor, aerosol and other data provided by models such as ECMWF, GFS, etc. The ground meteorological observation data is a series of meteorological parameters monitored and recorded by the ground meteorological observation station in real time, such as ground temperature, air pressure, wind speed, wind direction, humidity, sunshine duration, etc. The processor can access the ground meteorological observation data of the ground meteorological observation station. The satellite channel cloud image is the multi-spectral channel data of the geostationary satellite (such as Himawari-8, GOES and Japanese Himawari-8 satellite, etc.). The processor can access the official authorized platform corresponding to the satellite to acquire the satellite channel cloud image.

[0042] Specifically, based on the hierarchical sampling and physical index mapping mechanism, the numerical weather prediction, ground meteorological observation data and satellite channel cloud images can be fused by time alignment, space mapping and variable normalization to construct a three-channel pseudo-color cloud image based on cloud top temperature. The cloud top temperature is an important feature of the cloud and is related to the height of the cloud. In the construction of the three-channel pseudo-color cloud image based on the cloud top temperature, different colors can be used to represent the cloud layers in different cloud top temperature ranges.

[0043] Optionally, the three channels of the three-channel pseudo-color cloud image are infrared, visible light and water vapor channels.

[0044] S202, the processor constructs a hierarchical cloud image set with time dimension information according to the three-channel pseudo-color cloud image.

[0045] Specifically, the three-channel pseudo-color cloud image is constructed as a hierarchical cloud image set, that is, the three-channel pseudo-color cloud image is processed hierarchically, and the purpose is to divide the cloud image according to different features or attributes. For example, the cloud can be divided according to the cloud top height, the cloud layer thickness, the type of the cloud and other factors. Each layer of the cloud image contains different information of the cloud layer, so that the influence of the cloud layer on the solar radiation can be analyzed in more detail, which helps to extract more targeted features.

[0046] Optionally, the three-channel pseudo-color cloud image is processed hierarchically by using channel-specific extraction and thermodynamic hierarchical processing.

[0047] Specifically, by introducing a time evolution clue when processing the three-channel pseudo-color cloud image hierarchically, a hierarchical cloud image set with time dimension information is constructed.

[0048] S203, a pre-designed U-Net network structure is used to extract multi-scale features of the hierarchical cloud image set to construct a global satellite feature map.

[0049] Specifically, U-Net is a deep learning network structure widely used in image segmentation and feature extraction, which has an encoder-decoder architecture. The encoder part gradually reduces the size of the image through a series of convolutional layers and pooling layers, while extracting high-level semantic features of the image. The decoder part gradually restores the size of the image through upsampling and convolution operations, and fuses and refines the features extracted by the encoder to generate more accurate feature maps.

[0050] The pre-designed U-Net network structure in the embodiment of the present disclosure retains the upsampling block, downsampling block and feature fusion method compared with the existing U-Net, and adds FPN (Feature Pyramid Network) to process the multi-scale features of the cloud image.

[0051] Specifically, the layered cloud atlas is input into the pre-designed U-Net network, and the U-Net network extracts features from each layer of the layered cloud atlas. In the encoder part, different levels of convolutional layers and pooling layers extract features of different scales, thereby obtaining multi-scale features of the layered cloud atlas.

[0052] Specifically, by fusing the multi-scale features extracted by the U-Net network, a global satellite feature map can be constructed. The global satellite feature map contains all important feature information extracted from the layered cloud atlas, which reflects the distribution, type, and dynamic change of the cloud layer in the global range, and these features are closely related to the ground solar radiation. By constructing the global satellite feature map, cloud information of different regions and different time points can be uniformly represented, thereby providing more comprehensive data support for subsequent ground solar radiation prediction.

[0053] S204, the processor extracts a local feature map corresponding to the target photovoltaic power station from the global satellite feature map, and inputs the local feature map into a pre-trained photovoltaic prediction model to predict the ground solar radiation of the target photovoltaic power station.

[0054] Specifically, since the global satellite feature map contains cloud feature information in the global range, and the embodiment of the present disclosure needs to focus on the situation of the region where the target photovoltaic power station is located, the processor needs to first extract a local region feature map corresponding to the target photovoltaic power station from the global satellite feature map according to the geographical location information of the target photovoltaic power station, so as to reduce the amount of data to be processed in subsequent prediction and accurately focus on the region where the target photovoltaic power station is located during prediction.

[0055] Specifically, the local region feature map is extracted by intercepting a rectangular region corresponding to the target photovoltaic power station in the global satellite feature map.

[0056] Specifically, the extracted local feature map is input into the pre-trained photovoltaic prediction model, and the ground solar radiation of the target photovoltaic power station can be obtained.

[0057] Optionally, the photovoltaic prediction model is constructed by using a machine learning algorithm or a deep learning algorithm, for example, a regression model, a neural network, etc. When training the prediction model, a large amount of historical data needs to be used in advance, including historical ground solar radiation observation data and corresponding local feature maps in the global satellite feature map, so that the model accurately learns the mapping relationship between the local feature map and the ground solar radiation.

[0058] The embodiments of the present disclosure find that, in the related art, when facing relatively complex weather conditions, the error between the predicted ground surface solar radiation and the actual solar radiation is large because the data source for prediction is single, and when the input data source is a track cloud image, the details in the image cannot be extracted well, and the time and space information of different height levels in the cloud image is ignored. In the embodiments of the present disclosure, when predicting the ground surface solar radiation of a target photovoltaic power station, numerical weather prediction, ground meteorological observation data and satellite channel cloud Figure Three data are obtained, a three-channel pseudo-color cloud image based on cloud top temperature is constructed, a complete meteorological information chain can be formed, and the limitation of a single data source is avoided. After the three-channel pseudo-color cloud image is constructed, a hierarchical cloud atlas with time dimension information is constructed according to the three-channel pseudo-color cloud image. The vertical structure information of the cloud layer can be retained, and the attenuation effect of different height cloud layers on solar radiation can be distinguished. By using a pre-designed U-Net network structure to extract the multi-scale features of the hierarchical cloud atlas, a global satellite feature map is constructed. The influence of cloud clusters and various weather factors on the ground surface solar radiation in the time dimension and the space dimension can be determined with the aid of the context information in the whole image (i.e. in the global satellite feature map). After the global satellite feature map is constructed, the local feature map corresponding to the target photovoltaic power station is intercepted from the global satellite feature map, and then the local feature map is input into a photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station, so that sufficient context information can be referred to and irrelevant regional noise interference can be avoided when predicting the ground surface solar radiation of the target photovoltaic power station. Therefore, the embodiments of the present disclosure can accurately predict the ground surface solar radiation of the target photovoltaic power station even when facing relatively complex weather conditions, and the prediction accuracy of the ground surface solar radiation is improved.

[0059] In addition, although the satellite channel cloud image has a medium resolution (the resolution is greater than 5 km) compared with the single-layer geostationary orbit cloud image used in the related art, the satellite channel cloud image has a wide coverage and can reflect the continuous structural characteristics of large-scale weather systems (such as cloud bands and vortices). The photovoltaic prediction model learns the spatial arrangement and evolution trend of these large-scale cloud clusters through training, determines the regulation effect of the cloud clusters on the ground surface radiation in different geographical regions, and reverses the fine-grained radiation distribution at a future time based on the context information even if the local pixel is small after the local feature map corresponding to the target photovoltaic power station is intercepted from the global satellite feature map. Therefore, even if the original resolution of the input satellite channel cloud image is greater than 5 km, the prediction accuracy of the ground surface solar radiation can still be improved in a finer scale, and the performance of high-precision radiation prediction with a spatial resolution of less than 5 km x 5 km can be achieved.

[0060] In some embodiments, the method for obtaining numerical weather prediction, ground meteorological observation data and satellite channel cloud image, constructing a three-channel pseudo-color cloud image based on cloud top temperature, comprises: interpolating the numerical weather prediction, ground meteorological station data and satellite channel cloud image to a set time frequency; mapping the numerical weather prediction and ground meteorological station data to a spatial grid scale corresponding to the satellite channel cloud image; and splicing the three-channel pseudo-color data of the satellite channel cloud image with the numerical weather prediction and ground meteorological station data to obtain a three-channel pseudo-color cloud image based on cloud top temperature. Figure One

[0061] Specifically, since different data sources (i.e. numerical weather prediction, ground meteorological station, satellite channel cloud image) can have different time frequencies. Therefore, directly splicing the three data sources can easily lead to disorder of the time dimension of the constructed three-channel pseudo-color cloud image. Therefore, in order to ensure the consistency of the data in the time dimension, the numerical weather prediction, ground meteorological station data and satellite channel cloud image need to be interpolated first, and their time frequencies are unified to a set time frequency (e.g. 10 minutes, 15 minutes or 20 minutes, etc.).

[0062] Optionally, the interpolation method includes but is not limited to linear interpolation, spline interpolation or other time interpolation algorithms.

[0063] Specifically, the numerical weather prediction and ground meteorological station data can come from different spatial observation points or model outputs, and therefore, a spatial interpolation method (such as bilinear interpolation, nearest neighbor interpolation, etc.) is needed to map these data to a spatial grid scale corresponding to the satellite channel cloud image, so as to facilitate subsequent fusion of the three data sources. Figure One

[0064] Specifically, the three-channel pseudo-color data of the satellite channel cloud image can provide visual information of the cloud top temperature, and the numerical weather prediction and ground meteorological station data can provide other meteorological information (e.g. temperature, humidity and wind speed, etc.) of the atmosphere and the ground. By splicing the numerical weather prediction and ground meteorological station data with the three-channel pseudo-color data of the satellite channel cloud image, multi-dimensional meteorological information integrating the visual information of the cloud top temperature, numerical weather prediction and ground meteorological station data can be generated, which provides more comprehensive data for subsequent solar radiation prediction.

[0065] Optionally, before splicing the three-channel pseudo-color data of the satellite channel cloud image with the numerical weather prediction and ground meteorological station data, the prediction method further comprises: using a Z-score normalization method to perform physical index normalization processing on the numerical weather prediction, ground meteorological station data and satellite channel cloud image. In order to eliminate the dimensional influence and retain the fluctuation trend when the value ranges of the three data sources are different.

[0066] ​​In this embodiment, when the numerical weather prediction, ground meteorological observation data and satellite channel cloud image are fused into a three-channel pseudo-color cloud image, the time dimension and the space dimension of the three are unified. In this way, the dynamic changes of the atmosphere and the cloud layer can be more accurately reflected, which is conducive to improving the accuracy of subsequent solar radiation degree prediction.

[0067] In some embodiments, according to the three-channel pseudo-color cloud image, a layered cloud image set with time dimension information is constructed, including: continuously obtaining multiple frames of satellite TBB channel images according to a set time interval from the three-channel pseudo-color cloud image, constructing a time sliding window, and determining a brightness temperature change sequence; according to the brightness temperature change sequence, performing histogram division and threshold cutting on each frame of satellite TBB channel image to extract a brightness temperature distribution area representing a cloud body at different height levels; and stacking the images formed by the brightness temperature distribution areas in time order to construct an image sequence tensor to obtain the layered cloud image set.

[0068] Specifically, the satellite TBB (Brightness Temperature) channel image is an image observed by a satellite in an infrared band. The satellite TBB channel image can reflect the radiation temperature at the top of the cloud layer.

[0069] Specifically, satellite TBB channel images at corresponding time points are extracted from the three-channel pseudo-color cloud image according to a set time interval (for example, 10 minutes apart, and three frames of satellite TBB channel images are continuously extracted), so that continuous TBB image frames taken at different time points can be obtained. These image frames contain brightness temperature information of the cloud layer at different times.

[0070] Specifically, the time sliding window is a fixed-length window that moves in a time sequence. In the constructed time sliding window, each frame of satellite TBB channel image in the window is analyzed, and the brightness temperature difference value of the corresponding pixel points between adjacent frames of images in the window is calculated, so that a brightness temperature change sequence can be obtained.

[0071] Specifically, according to the brightness temperature change sequence, the brightness temperature values of all pixel points in each frame of satellite TBB channel image can be counted, and a brightness temperature histogram can be constructed. The histogram can show the distribution of different brightness temperature values in the image. The horizontal axis represents the brightness temperature value, and the vertical axis represents the number of pixel points corresponding to the brightness temperature value.

[0072] Specifically, according to the result of histogram division, the brightness temperature threshold can be determined. These brightness temperature thresholds divide the brightness temperature range into different intervals, and each interval corresponds to a specific cloud feature or height level. For example, lower brightness temperature values can correspond to thicker and higher cloud bodies, and higher brightness temperature values can correspond to thinner and lower cloud bodies or clear sky regions. By comparing the brightness temperature value of each pixel in the image with the set threshold, the image is divided into different regions, each region representing the brightness temperature distribution of a specific height level or type of cloud body, so that the brightness temperature distribution region corresponding to each frame of satellite TBB channel image can be determined.

[0073] Specifically, the image layers representing the brightness temperature distribution regions of cloud bodies at different height levels are stacked in time sequence, that is, a three-dimensional data structure is obtained, in which two dimensions represent the spatial coordinates (such as rows and columns) of the image, and the third dimension represents time. This three-dimensional data structure is called a tensor (i.e. image sequence tensor). The image sequence tensor contains the brightness temperature distribution information of the cloud layer at different spatial positions, as well as the change of the brightness temperature distribution information over time. Therefore, a layered cloud atlas with time dimension information is formed.

[0074] Optionally, after the image sequence tensor is constructed, the prediction method further includes: pre-processing the image sequence tensor by using difference filtering and local standard deviation enhancement.

[0075] Specifically, after the image sequence tensor is constructed, by pre-processing the image sequence tensor by using difference filtering and local standard deviation enhancement, the quality of the image sequence tensor can be further improved, and the dynamic trend of the key region can be highlighted, so that the subsequent cloud layer analysis and prediction can be more accurately performed.

[0076] In this embodiment, based on the three-channel pseudo-color cloud image, a layered cloud atlas with time dimension information is constructed, which contains the brightness temperature distribution information of the cloud layer at different spatial positions, as well as the change of the brightness temperature distribution information over time. This provides more accurate data basis for subsequent analysis of the influence of the cloud layer on the solar radiation, and is beneficial to improve the accuracy of subsequent prediction of the solar radiation on the ground.

[0077] In some embodiments, the pre-designed U-Net network structure includes an up-sampling block, a down-sampling block and an FPN module; the pre-designed U-Net network structure is used to extract multi-scale features of the layered cloud atlas, and a global satellite feature map is constructed, including: the up-sampling block and the down-sampling block are used to extract multi-scale features from the layered cloud atlas, to obtain a plurality of feature maps of different scales; the FPN module is used to pre-process the plurality of feature maps of different scales; and the 2D convolution operation is used to connect the plurality of pre-processed feature maps of different scales, to obtain the global satellite feature map.

[0078] Exemplarily, the pre-designed U-Net network structure is as shown in FIG. 1, which includes up-sampling blocks, down-sampling blocks and FPN modules. The multiple up-sampling blocks and down-sampling blocks constitute the U-Net backbone network in the upper half of the figure, and the FPN modules constitute the FPN network in the lower half of the figure. Figure 3 Figure 3 As CNN has an advantage in processing high-resolution satellite channel cloud sets, a classic CNN network ResNet is adopted as the up-sampling block and the down-sampling block to improve the backbone network of the U-Net, i.e., each up-sampling block is composed of a deconvolution layer, and each down-sampling block is composed of a Res-block. Figure 3 Figure 3 The upper right corner of the figure shows different operators, and the lower right corner shows the details of the Res-block and the deconvolution layer.

[0079] Specifically, the down-sampling block is used to gradually reduce the size of the feature map while increasing the number of channels of the feature map, so as to extract abstract features of different scales. The up-sampling block has the opposite effect of the down-sampling block, and is used to gradually increase the size of the feature map while reducing the number of channels of the feature map.

[0080] Specifically, local features are extracted from the hierarchical cloud set by the down-sampling block to obtain feature maps of different scales. The size of the feature maps of different scales is increased by the up-sampling block through deconvolution or interpolation operation, and then the feature maps are spliced or added with the feature maps of the same scale in the down-sampling process, so as to realize the fusion of features and obtain multiple feature maps of different scales.

[0081] Specifically, the FPN module is a network structure for building a multi-scale feature pyramid. In the feature extraction process of the hierarchical cloud set, the feature maps of different scales contain cloud layer information of different levels. The FPN module can further preprocess these feature maps to enhance the relevance and complementarity between features. For example, the high-level feature map may contain the overall distribution and category information of the cloud layer, while the low-level feature map contains the edge and detail information of the cloud layer. Through the processing of the FPN module, these information can be effectively combined.

[0082] Specifically, the multiple feature maps of different scales preprocessed by the FPN module are stacked in a set order (such as from low to high or from high to low), and then a 2D convolution (Conv2D) operation is performed on the stacked feature maps using one or more 2D convolution kernels, so as to obtain a global satellite feature map.

[0083] Optionally, the 2D convolution operation is performed through multiple convolution kernels, and each convolution kernel is composed of a linear function and a Sigmoid activation function.

[0084] ​​In this embodiment, by cooperating the up-sampling block and the down-sampling block, the multi-scale features are extracted from the hierarchical cloud set, which can ensure that the proposed features cover the cloud layer information from local to global. By preprocessing the multi-scale feature maps through the FPN module, the relevance and complementarity between the features can be enhanced. Therefore, the global satellite feature map obtained by connecting the preprocessed feature maps of different scales through 2D convolution operation can provide more representative data for subsequent prediction of ground surface solar radiation, which is conducive to improving the accuracy of subsequent prediction of ground surface solar radiation.

[0085] In some embodiments, the FPN module is used to preprocess the feature maps of different scales, including: determining the largest scale feature map in the plurality of feature maps of different scales; performing 2D convolution operation and skip connection on the largest scale feature map; performing 2D convolution operation and spatial pyramid pooling on the feature maps other than the largest scale feature map, and using a bilinear interpolation algorithm to enlarge the feature maps other than the largest scale feature map to the same scale as the largest scale.

[0086] Specifically, the largest scale feature map in the plurality of feature maps of different scales contains edge, texture and other detailed information of the cloud layer, which belongs to the basis for processing feature maps of other scales. Therefore, the largest scale feature map in the plurality of feature maps of different scales needs to be determined first.

[0087] Specifically, the 2D convolution operation slides the convolution kernel on the feature map, calculates the dot product of the convolution kernel and the local region of the feature map, and extracts the features in the largest scale feature map. For example, the features of the cloud layer edge in the largest scale feature map are extracted through convolution operation, so that the boundary of the cloud layer is clearer, which is convenient for subsequent analysis and processing.

[0088] Specifically, the skip connection is a technology of directly connecting features of different layers. In the FPN module, the skip connection is performed on the largest scale feature map, which can retain the detailed information in the feature map and combine it with the subsequent fused features.

[0089] Specifically, the principle of performing 2D convolution operation on the largest scale feature map is the same as described above. Convolution operation is performed on the feature maps of other scales, which can extract the features in these feature maps and enhance the expression ability of the features.

[0090] Specifically, the spatial pyramid pooling (SPP) is a method of converting feature maps of different scales into fixed size feature vectors. In the FPN module, spatial pyramid pooling is performed on the feature maps of other scales, which can extract multi-scale context information in these feature maps.

[0091] Specifically, the bilinear interpolation is an image interpolation algorithm that can estimate the value of a new pixel point by performing linear interpolation between adjacent pixel points. In the feature map upsizing process, the bilinear interpolation can calculate the value of each pixel point in the upsized feature map according to the values of the surrounding pixel points, thereby realizing the size upsizing of the feature map. Since subsequent fusion of feature maps of different scales is required, the bilinear interpolation algorithm is used to upsize the feature maps of other scales to the same size as the largest scale.

[0092] In this embodiment, when the FPN module is used to preprocess multiple feature maps of different scales, the maximum scale feature map is first determined, and 2D convolution operation and jump connection are performed thereon to preserve key information. For other scale feature maps, 2D convolution operation and spatial pyramid pooling are first performed to extract multi-scale context information. Finally, the bilinear interpolation algorithm is used to upsize them to the same size as the maximum scale, which can realize the size unification of the feature maps. In this way, the feature maps of different scales have the same spatial resolution, which enhances the correlation between features and facilitates the subsequent fusion of multiple feature maps of different scales into a global satellite feature map.

[0093] In some embodiments, inputting the local feature map into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station comprises: performing gray scale standardization and scale resampling processing on the local feature map; inputting the processed local feature map into the convolution layer network of the photovoltaic prediction model to extract the texture structure features and brightness distribution patterns of the local feature map layer by layer; inputting the texture structure features and brightness distribution patterns into the fully connected layer network of the photovoltaic prediction model to determine the ground surface solar radiation of the target photovoltaic power station.

[0094] Specifically, since different local feature maps have inconsistent gray scale ranges and scales, the prediction results obtained by the photovoltaic prediction model based on local feature maps with standard gray scale and standard size are more accurate. Therefore, the local feature map needs to be processed by gray scale standardization (e.g., scaling the gray scale value to [0, 1] or [-1, 1]) and scale resampling (e.g., scaling to 64x64 pixels).

[0095] Specifically, the photovoltaic prediction model is constructed by using a shallow convolutional regression network, which includes a convolution layer network and a fully connected layer network.

[0096] Specifically, the convolution layer network is composed of multiple convolution layers. Therefore, inputting the processed local feature map into the convolution layer network of the photovoltaic prediction model can extract the texture structure features and brightness distribution patterns of the local feature map layer by layer.

[0097] Specifically, after the convolutional layer network extracts the texture structure features and the brightness distribution patterns, the features are input into the fully connected layer network for integration and mapping, and the high-dimensional feature vector is converted into a low-dimensional output, that is, the prediction value of the ground surface solar radiation of the target photovoltaic power station is obtained.

[0098] In this embodiment, the consistency of the input data is ensured through grayscale standardization and scale resampling processing. The texture structure features and the brightness distribution patterns are extracted layer by layer through the convolutional layer network, which can comprehensively capture the key information in the local feature map and improve the accuracy of the obtained ground surface solar radiation.

[0099] Another method for predicting the ground surface solar radiation is provided in the embodiments of the present disclosure, as shown in the following formula (1). Figure 4 The prediction method comprises the following steps.

[0100] S401, the processor acquires numerical weather prediction, ground meteorological observation data and satellite channel cloud images, and constructs a three-channel pseudo-color cloud image based on cloud top temperature.

[0101] S402, the processor constructs a layered cloud image set with time dimension information according to the three-channel pseudo-color cloud image.

[0102] S403, the processor extracts multi-scale features of the layered cloud image set by using a pre-designed U-Net network structure, and constructs a global satellite feature map.

[0103] S404, the processor extracts the local feature map corresponding to the target photovoltaic power station from the global satellite feature map, and inputs the local feature map into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station.

[0104] S405, the processor calculates the flow field between the continuous cloud images in the layered cloud image set by using a Dual TV-L1 optical flow algorithm, and performs short-time cloud layer extrapolation on the layered cloud image set.

[0105] Specifically, the optical flow algorithm is a method for estimating the motion information of pixels in an image sequence. The Dual TV-L1 optical flow algorithm is an optical flow calculation method based on variational method, which combines total variation (TV) regularization and L1 norm data items. When calculating the flow field between the continuous cloud images in the layered cloud image set, the Dual TV-L1 optical flow can regard the pixels in the layered cloud image set as moving entities, and solve the motion vector of each pixel by minimizing the energy function, so as to obtain the motion flow field between the cloud layers.

[0106] Specifically, according to the calculated flow field, the cloud layers in the layered cloud image set can be short-time predicted. That is, according to the motion vector of each pixel in the current time cloud image, the cloud layer is moved in time to obtain the cloud layer distribution in the future short time.

[0107] S406, the processor extracts the features of different cloud layers in the cloud layer extrapolated cloud atlas set by using the multi-layer residual network, calculates the contribution of each cloud layer to the total radiation of the ground surface by a dynamic weight mechanism, and corrects the predicted ground surface solar radiation.

[0108] Specifically, the cloud layer extrapolated cloud atlas set is input into the multi-layer residual network, and the multi-layer residual network performs layer-by-layer convolution operation on the cloud atlas by convolution layer, so as to extract the features of different cloud layers.

[0109] Specifically, the dynamic weight mechanism can dynamically calculate the contribution weight of each cloud layer to the total radiation of the ground surface according to different features of the cloud layer (such as cloud thickness, cloud amount, cloud type, etc.). These features can be represented by the feature vector extracted by the multi-layer residual network. For example, the cloud layer with larger cloud thickness has stronger shielding effect on solar radiation, and its contribution weight is likely to be higher; while the contribution weight of thin cloud layer is relatively low.

[0110] Optionally, the dynamic weight mechanism adopts a machine learning model (such as neural network, support vector machine, etc.) to realize.

[0111] Specifically, a correction coefficient can be calculated by the contribution weight of each cloud layer, and the predicted ground surface solar radiation is multiplied by the correction coefficient to obtain the corrected ground surface solar radiation. Alternatively, the shielding amount of each cloud layer to the ground surface solar radiation can be calculated by the contribution weight of each cloud layer, and the total shielding amount of all cloud layers to the ground surface solar radiation is obtained, and the predicted ground surface solar radiation is subtracted by the total shielding amount to obtain the corrected ground surface solar radiation.

[0112] In this embodiment, after predicting the ground surface solar radiation of the target photovoltaic power station, the contribution of each cloud layer to the total radiation of the ground surface is calculated, and the predicted ground surface solar radiation is corrected. In this way, the prediction accuracy and precision of the ground surface solar radiation are further improved.

[0113] In combination with Figure 5As shown, the embodiment of the present disclosure provides a prediction device 500 for ground surface solar radiation, which comprises a first construction module 501, a second construction module 502, a third construction module 503 and a prediction module 504. The first construction module 501 is configured to acquire numerical weather prediction, ground meteorological observation data and satellite channel cloud images, and construct a three-channel pseudo-color cloud image based on cloud top temperature. The second construction module 502 is configured to construct a hierarchical cloud image set with time dimension information according to the three-channel pseudo-color cloud image. The third construction module 503 is configured to extract multi-scale features of the hierarchical cloud image set by using a pre-designed U-Net network structure, and construct a global satellite feature map. The prediction module 504 is configured to cut a local feature map corresponding to a target photovoltaic power station from the global satellite feature map, and input the local feature map into a pre-trained photovoltaic prediction model to predict the ground surface solar radiation of the target photovoltaic power station.

[0114] In combination Figure 6 As shown, the embodiment of the present disclosure provides a prediction device 600 for ground surface solar radiation, which comprises a processor 601 and a memory 602. Optionally, the device can further comprise a communication interface 603 and a bus 604. Wherein the processor 601, the communication interface 603 and the memory 602 can complete mutual communication through the bus 604. The communication interface 603 can be used for information transmission. The processor 601 can invoke the logic instructions in the memory 602 to execute the prediction method for ground surface solar radiation of the above-mentioned embodiment.

[0115] In addition, the logic instructions in the memory 602 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium.

[0116] The memory 602 as a kind of computer readable storage medium can be used to store software programs, computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor 601 executes the program instructions / modules stored in the memory 602, thereby executing function application and data processing, i.e. realizing the prediction method for ground surface solar radiation in the above-mentioned embodiment.

[0117] The memory 602 can include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function; the data storage area can store data created according to the use of the terminal device, etc. In addition, the memory 602 can include a high-speed random access memory, and can also include a non-volatile memory.

[0118] The embodiment of the present disclosure provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are configured to execute the above-mentioned prediction method for ground surface solar radiation.

[0119] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes one or more instructions to make a computer device (which can be a personal computer, a server, or a network device) execute all or part of the steps of the method disclosed in the embodiment of the present disclosure. The foregoing storage medium can be a non-transitory storage medium, for example, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0120] The above description and drawings sufficiently show the embodiments of the present disclosure to enable a person skilled in the art to practice them. Other embodiments can include structural, logical, electrical, process, and other changes. The embodiments only represent possible changes. Unless explicitly required, individual components and functions are optional, and the order of operations can be changed. Parts and features of some embodiments can be included in or replace parts and features of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and not to limit the claims. As used in the description of the embodiments and the claims, unless the context clearly requires otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms as well. Similarly, as used in this application, the term "and / or" refers to any and all possible combinations of one or more associated listed items. In addition, when used in this application, the term "comprise" and its variants "comprises" and / or comprises" and the like mean the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, or device that includes the stated element. In this document, each embodiment focuses on the differences from other embodiments, and the same or similar parts between various embodiments can be referred to each other. For the method, product, and the like disclosed in the embodiments, if it corresponds to the method part disclosed in the embodiments, the relevant part can be referred to the description of the method part.

[0121] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods for each specific application to realize the described functions, but such implementation should not be considered beyond the scope of the embodiments of the present disclosure. The skilled person can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0122] In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units can only be a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms. The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to implement the embodiments. In addition, each functional unit in the embodiments of the present disclosure can be integrated in one processing unit, or each unit can be a physically independent unit, or two or more units can be integrated in one unit.

[0123] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

Claims

1. A method for predicting surface solar radiance, characterized in that, include: Acquire numerical weather prediction, surface meteorological observation data and satellite channel cloud images to construct a three-channel pseudo-color cloud image based on cloud top temperature; Based on the three-channel pseudo-color cloud image, a hierarchical cloud atlas with time dimension information is constructed. A pre-designed U-Net network structure is used to extract multi-scale features from hierarchical cloud atlases and construct a global satellite feature map. The local feature map corresponding to the target photovoltaic power station is extracted from the global satellite feature map, and the local feature map is input into the pre-trained photovoltaic prediction model to predict the surface solar irradiance of the target photovoltaic power station.

2. The prediction method according to claim 1, characterized in that, Acquire numerical weather prediction, surface meteorological observation data, and satellite channel cloud images to construct a three-channel pseudo-color cloud image based on cloud top temperature, including: Numerical weather forecasts, ground meteorological station data, and satellite cloud images are uniformly interpolated to a set time frequency; Map numerical weather forecasts and ground meteorological station data to a spatial grid scale consistent with satellite channel cloud images; By stitching together the three-channel pseudo-color data of the satellite cloud image with numerical weather prediction and ground meteorological station data, a three-channel pseudo-color cloud image based on cloud top temperature is obtained.

3. The prediction method according to claim 1, characterized in that, Based on the three-channel pseudo-color cloud image, a hierarchical cloud image atlas with time dimension information is constructed, including: From the three-channel pseudo-color cloud image, multiple frames of satellite TBB channel images are continuously acquired at set time intervals to construct a time sliding window and determine the brightness temperature change sequence. Based on the brightness temperature change sequence, histogram division and threshold cutting are performed on each frame of satellite TBB channel image to extract the brightness temperature distribution area representing cloud bodies at different altitude levels; By stacking image layers composed of brightness temperature distribution regions in chronological order, an image sequence tensor is constructed to obtain a layered cloud atlas.

4. The prediction method according to claim 1, characterized in that, The pre-designed U-Net network structure includes upsampling blocks, downsampling blocks, and FPN modules; A pre-designed U-Net network structure is used to extract multi-scale features from hierarchical cloud atlases, constructing a global satellite feature map, including: Multi-scale features are extracted from the hierarchical cloud map set by using upsampling blocks and downsampling blocks to obtain feature maps of multiple different scales; The FPN module is used to preprocess feature maps of different scales; 2D convolution operations are used to connect multiple preprocessed feature maps of different scales to obtain a global satellite feature map.

5. The prediction method according to claim 4, characterized in that, The FPN module is used to preprocess feature maps of different scales, including: Determine the largest-scale feature map among multiple feature maps of different scales; Perform 2D convolution operations and skip connections on the feature map at the largest scale; 2D convolution and spatial pyramid pooling are performed on the feature maps except for the largest scale in multiple feature maps, and bilinear interpolation algorithm is used to enlarge the feature maps except for the largest scale to the same scale.

6. The prediction method according to claim 1, characterized in that, The local feature map is input into a pre-trained photovoltaic prediction model to predict the surface solar irradiance of the target photovoltaic power station, including: Perform grayscale normalization and scale resampling on the local feature map; The processed local feature map is input into the convolutional network of the photovoltaic prediction model, and the texture structure features and brightness distribution patterns of the local feature map are extracted layer by layer. Texture structure features and brightness distribution patterns are input into the fully connected layer network of the photovoltaic prediction model to determine the surface solar irradiance of the target photovoltaic power station.

7. The prediction method according to any one of claims 1 to 6, characterized in that, After predicting the surface solar irradiance of the target photovoltaic power station based on local feature maps, the prediction method also includes: The Dual TV-L1 optical flow algorithm is used to calculate the flow field between consecutive cloud images in the layered cloud atlas and to perform short-time cloud extrapolation on the layered cloud atlas. A multi-layer residual network is used to extract the features of different cloud layers in the layered cloud map after cloud extrapolation. The contribution of each cloud layer to the total surface radiance is calculated through a dynamic weighting mechanism, and the predicted surface solar radiance is corrected.

8. A device for predicting surface solar radiance, characterized in that, include: The first construction module is configured to acquire numerical weather forecasts, ground meteorological observation data, and satellite channel cloud images to construct a three-channel pseudo-color cloud image based on cloud top temperature. The second construction module is configured to construct a hierarchical cloud map atlas with time dimension information based on the three-channel pseudo-color cloud map; The third building module is configured to extract multi-scale features of hierarchical cloud atlases using a pre-designed U-Net network structure to construct a global satellite feature map. The prediction module is configured to extract a local feature map corresponding to the target photovoltaic power station from the global satellite feature map, and input the local feature map into a pre-trained photovoltaic prediction model to predict the surface solar irradiance of the target photovoltaic power station.

9. A device for predicting surface solar radiance, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to perform the method for predicting surface solar radiance as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Equipment body; The device for predicting surface solar radiance as described in claim 8 or 9 is disposed on the device body.

Images