A satellite-ground combined ultra-short-term solar radiation prediction method and system

Abstract

The present invention relates to the field of solar radiation prediction technology, specifically to a satellite-ground combined ultra-short-term solar radiation prediction method and system. It includes the following steps: S1: Collect current state data and historical big data based on solar radiation. S2: Preprocess ground-based cloud images and air-based cloud images. S3: Identify cloud state data based on a constructed cloud layer recognition model. S4: Use a solar radiation attenuation prediction model to identify the total attenuation rate of solar radiation, and construct a radiation attenuation table. S5: Use extrapolation to predict cloud state change data and satellite-observed total radiation change data, and then calculate the future short-term ground-observed total radiation prediction data. The present invention combines the observations of geostationary satellites and all-sky imagers to achieve satellite-ground combined observations of clouds at different altitudes, fully considers the impact of clouds in the entire atmospheric column on solar shortwave radiation, and optimizes the calculation process while maintaining the prediction accuracy, thereby improving the accuracy and efficiency of solar radiation attenuation prediction.

Description

Technical Field

[0001] The present invention relates to the field of solar radiation technology, and in particular to a satellite-ground combined ultra-short-term solar radiation prediction method and a satellite-ground combined ultra-short-term solar radiation prediction system. Background Art

[0002] In photovoltaic power generation, solar radiation has a direct impact, which is mainly reflected in the following aspects: 1. The influence of light intensity. The power generation of photovoltaic cells directly depends on the intensity of incident light. The higher the light intensity, the greater the output power of photovoltaic cells. 2. The influence of spectrum: Light of different wavelengths has an impact on the conversion efficiency of photovoltaic cells. The spectral distribution of sunlight changes with time and weather conditions. 3. The influence of temperature: The conversion efficiency of photovoltaic cells decreases with increasing temperature. High temperature will cause the battery voltage to drop, thereby reducing power output. When the sunlight is strong, the surface temperature of the battery rises, which will reduce the power generation efficiency. 4. The influence of seasonal changes: The change of solar altitude angle in different seasons will affect the incident light intensity and spectral distribution of photovoltaic cells. In winter, the light duration is shorter and the light intensity is weaker, and the efficiency of photovoltaic power generation will decrease. 5. The influence of clouds, aerosols, water vapor content in the air, etc. Among them, clouds are the most important factor affecting the solar irradiance reaching the surface. The generation of evaporation and movement changes are the fundamental reasons for the uncertainty of solar irradiance on the ground. Therefore, in photovoltaic power generation, the prediction of solar radiation is particularly important.

[0003] A Chinese invention patent with publication number "CN112989709A" discloses an ultra-short-term forecasting method for solar energy resources based on ground-based cloud maps, the ultra-short-term forecasting method for solar energy resources comprising: determining historical cloud information based on a plurality of original ground-based cloud maps within a predetermined time period in the past obtained by an all-sky imager; establishing a cloud amount forecasting model according to the historical cloud information; determining a forecast cloud amount according to the cloud amount forecasting model, the forecast cloud amount being the cloud amount within a predetermined time period in the future; establishing a clear sky radiation forecasting model according to historical clear sky radiation data within a predetermined time period in the past; establishing a cloud attenuation model according to the historical cloud information; determining a cloud-sky radiation model according to the clear sky radiation forecasting model and the cloud attenuation model; and determining the solar radiation output within the predetermined time period in the future using the cloud-sky radiation model according to the forecast cloud amount.

[0004] Although the above forecasting methods solve the interference of some clouds on solar radiation, the ground-based cloud images can only identify the status information of low-altitude clouds, and it is difficult to fully represent the interference of clouds in the sky on solar radiation. Therefore, the forecast accuracy is not high. In addition, in cloud recognition, the cloud type and the coverage interference of low, medium and high-altitude clouds are not taken into account. Therefore, the accuracy and timeliness of solar radiation prediction are not strong. In addition, since cloud prediction uses model prediction, the training and calculation are complicated, and the overall prediction efficiency is low. Summary of the invention

[0005] Based on this, it is necessary to propose a satellite-ground combined ultra-short-term solar radiation prediction method and system to address the problems of low accuracy and low efficiency in existing solar radiation prediction.

[0006] The present invention is implemented by the following technical solution: A satellite-ground combined ultra-short-term solar radiation prediction method comprises the following steps:

[0007] S1: Collect ground-based cloud images, air-based cloud images, and satellite observation total radiation data, and obtain historical big data based on solar radiation.

[0008] S2: Preprocessing of ground-based cloud images and air-based cloud images.

[0009] S3: Obtain cloud status data based on the established cloud recognition model. The cloud status data includes the types and sizes of low-altitude clouds, mid-altitude clouds, and high-altitude clouds.

[0010] S4: Use the constructed SVM-LSTM-based solar radiation attenuation prediction model to identify the current total solar radiation attenuation rate according to the cloud state data. Construct a radiation attenuation table based on the total solar radiation attenuation rate and cloud state data.

[0011] S5: Use the extrapolation method to predict the cloud state change data and the satellite observation total radiation change data. According to the cloud state change data, query and obtain the corresponding total solar radiation attenuation rate in the radiation attenuation table. According to the satellite observation total radiation change data, calculate the future short-term ground observation total radiation forecast data.

[0012] The above-mentioned satellite-ground combined ultra-short-term solar radiation prediction method realizes satellite-ground combined observation of clouds at different altitudes by combining observations from geostationary satellites and all-sky imagers, and fully considers the impact of clouds in the entire atmospheric column on solar shortwave radiation. At the same time, target detection and semantic segmentation are used to identify ground-based cloud images and air-based cloud images respectively, which can not only perform pixel-by-pixel image recognition to obtain more accurate cloud layer identification data, but also the VGG-16 network model fused with the residual network can improve the performance and generalization ability of the model, enhance the robustness of the model, support deeper structures, and have stronger scalability and migration. The partial least squares method is used to extract the features of cloud data in the cloud map, which not only retains the important information in the data, but also simplifies the input data as a whole while maintaining the prediction accuracy, optimizes the calculation process, and improves the efficiency of solar radiation attenuation prediction.

[0013] Furthermore, the preprocessing methods include geometric correction, radiation correction, atmospheric correction, image stitching, and enhancement processing.

[0014] Geometric correction is used to correct the geometric distortion of ground-based cloud images and air-based cloud images. Radiation correction is used to correct radiation distortion and adjust the image grayscale value of ground-based cloud images and air-based cloud images. Atmospheric correction is used to eliminate the influence of atmospheric particle scattering and absorption on ground-based cloud images and air-based cloud images. Image stitching is used to stitch ground-based cloud images or air-based cloud images to generate image products covering the entire observation area. Enhancement processing is used to enhance the visual quality and information discernibility of ground-based cloud images and air-based cloud images.

[0015] Furthermore, in step S3, a deep learning-based object detection and semantic segmentation model is used to identify the actual type and size of the cloud layer. The specific steps are as follows:

[0016] S31: Randomly select M historical data from the historical big data, and mark the types and sizes of the clouds therein to form a labeled cloud data set.

[0017] S32: Build a deep learning-based object detection and semantic segmentation model. Use the cloud layer dataset to train the object detection and semantic segmentation model, and retain the model parameters that meet the test accuracy.

[0018] S33: Run the trained target detection and semantic segmentation models on the ground-based cloud image and the air-based cloud image to be tested to obtain pixel-level segmentation results based on cloud type and size.

[0019] S34: Compare the ground-based cloud image and the air-based cloud image, and integrate the cloud type and size at each pixel point one by one.

[0020] Furthermore, in step S4, the construction and training method of the solar radiation attenuation prediction model is as follows:

[0021] S41: Establish a solar radiation attenuation prediction model based on SVM-LSTM network.

[0022] S42: Randomly select N historical data from the historical big data, and run the target detection and semantic segmentation model to obtain N cloud state data. The ratio of the ground observation total radiation data to the satellite observation total radiation data in the N historical data is used as the attenuation rate data, and the attenuation rate data and the cloud state data constitute the attenuation rate data set.

[0023] S43: Divide the attenuation rate data set into a training set and a test set according to a preset ratio, use the training set to train the solar radiation attenuation prediction model, use the test set to verify the solar radiation attenuation prediction model, and retain the model parameters that meet the test accuracy.

[0024] S44: Inputting the cloud state data collected and identified in real time into the trained solar radiation attenuation prediction model to obtain corresponding attenuation rate data.

[0025] Furthermore, in step S41, the method for constructing the solar radiation attenuation prediction model is as follows:

[0026] S411: Use partial least squares method to extract features from the historical data sets of air-based cloud images and ground-based cloud images.

[0027] S412: Construct a support vector machine.

[0028] S413: Using LSTM network to optimize the parameters of support vector machine.

[0029] S414: Add a residual network to the solar radiation attenuation prediction model and perform skip-layer connection on the LSTM network.

[0030] Furthermore, in step S5, the predicted value of the ground-observed total radiation is expressed as:

[0031] O t =∑φ ti B t . i=1,2,…,mn.

[0032] φ ti =α ti *β ti *γ ti

[0033] In the formula, O t is the total ground-observed radiation at time t, B t is the total satellite radiation corresponding to each pixel at time t, mn is the total number of pixels, α ti is the solar radiation attenuation rate of the i-th pixel in the low-altitude cloud layer at time t, β ti is the solar radiation attenuation rate of the ith pixel in the hollow cloud layer at time t, γ ti is the solar radiation attenuation rate of the i-th pixel in the high-altitude cloud layer at time t.

[0034] The present invention also provides a satellite-ground combined ultra-short-term solar radiation prediction system, comprising a collection module, a preprocessing module, a cloud layer recognition model, a solar radiation attenuation prediction model and a solar radiation prediction module.

[0035] The acquisition module is used to collect ground-based cloud images, air-based cloud images, satellite-observed total radiation data, and obtain historical big data based on solar radiation. The preprocessing module is used to preprocess ground-based cloud images and air-based cloud images. The cloud recognition model is used to obtain cloud state data based on ground-based cloud images and air-based cloud images. The solar radiation attenuation prediction model is used to identify the current total solar radiation attenuation rate based on the cloud state data. And a radiation attenuation table is constructed based on the total solar radiation attenuation rate and the corresponding cloud state data. The solar radiation prediction module is used to predict cloud state change data and satellite-observed total radiation change data, and query and obtain the corresponding total solar radiation attenuation rate in the radiation attenuation table based on the cloud state change data, and calculate the future short-term ground observation total radiation prediction data based on the satellite-observed total radiation change data.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention combines the observations of geostationary satellites and all-sky imagers to achieve satellite-ground combined observations of clouds at different altitudes, fully considering the impact of clouds in the entire atmospheric column on solar shortwave radiation, and at the same time, using target detection and semantic segmentation to identify ground-based cloud images and air-based cloud images respectively, which can not only perform pixel-by-pixel image recognition to obtain more accurate cloud layer recognition data, but also integrate the VGG-16 network model of the residual network to improve the performance and generalization ability of the model, enhance the robustness of the model, support deeper structures, and have stronger scalability and mobility. The partial least squares method is used to extract the features of cloud layer data in the cloud map, which not only retains important information in the data, but also simplifies the input data as a whole while maintaining the prediction accuracy, optimizes the calculation process, and improves the efficiency of solar radiation attenuation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a step diagram of the satellite-ground combined ultra-short-term solar radiation prediction method in Example 1 of the present invention;

[0039] Figure 2 for Figure 1 The basic flow chart of the satellite-ground combined ultra-short-term solar radiation prediction method;

[0040] Figure 3 It is a schematic diagram of the framework of the satellite-ground combined ultra-short-term solar radiation prediction system in Example 2 of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component at the same time. When a component is considered to be "fixed to" another component, it may be directly fixed on the other component or there may be a central component at the same time.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0044] Example 1: Please refer to Figure 1-Figure 2 This embodiment provides a satellite-ground combined ultra-short-term solar radiation prediction method, comprising the following steps:

[0045] S1: Collect ground-based cloud images, air-based cloud images, satellite observation of total radiation data, and obtain historical big data based on solar radiation. Among them, the ground-based cloud images mainly observe the low-altitude clouds (water clouds) observed from the ground. The ground-based cloud images can be obtained using the observation results of the Huayun All-Sky Imager (HY-WP1A).

[0046] The air-based cloud images mainly observe the mid- and high-level cloud amounts from high altitudes downward. The air-based cloud images and the shortwave radiation B from high altitudes downward can be obtained with the help of the Fengyun-4 geostationary satellite cloud products. The information on ice clouds observed by geostationary satellites is a supplement to the ice clouds observed from the sky downwards. By combining the cloud observations of satellites and all-sky imagers, the conditions of clouds at different altitudes can be obtained, fully considering the impact of clouds in the entire atmospheric column on solar shortwave radiation, thereby improving the prediction accuracy of solar radiation.

[0047] Historical big data includes cloud status data, satellite observation total radiation data, and ground observation total radiation data. Historical big data can be obtained by filtering the observation data in the past year, while removing duplicate data and erroneous data.

[0048] S2: Preprocessing the ground-based cloud image and the air-based cloud image. In this embodiment, the preprocessing method includes:

[0049] Geometric correction: Correct the geometric distortion of ground-based cloud images and air-based cloud images to ensure that the image spatial position corresponds to the geographic coordinate system.

[0050] Radiation correction: Correct the radiation distortion caused by satellite sensor drift, detection sensitivity and other factors to make the image grayscale values ​​of ground-based cloud images and air-based cloud images closer to the actual radiation values ​​of the ground objects.

[0051] Atmospheric correction: eliminate the influence of atmospheric particle scattering and absorption on ground-based cloud images and air-based cloud images, and improve the physical information content of the images.

[0052] Image stitching: stitching multiple ground-based or air-based cloud images to generate image products covering a larger area. In the stitching process, the geometric and radiometric consistency between images needs to be considered.

[0053] Enhancement processing: including histogram equalization, edge sharpening and other methods, used to improve the visual quality and information discernibility of ground-based cloud images and air-based cloud images.

[0054] The above preprocessing method can improve the accuracy of cloud state recognition in cloud maps, and at the same time make the areas in ground-based cloud maps and air-based cloud maps correspond one to one. By comparing and combining the two cloud maps, cloud information can be obtained more comprehensively, thereby improving the accuracy of cloud recognition.

[0055] S3: Obtain cloud status data based on the established cloud recognition model. The cloud status data includes the types and sizes of low-altitude clouds, mid-altitude clouds, and high-altitude clouds.

[0056] In this embodiment, deep learning target detection and semantic segmentation technology are used to estimate the actual type and size of the cloud layer. The specific steps are as follows:

[0057] S31: Data preparation: Randomly select M historical data from the historical big data, and mark the type and size of the cloud layers to form a labeled cloud layer dataset. The initial historical data can be expressed as a(p, b, o), where p is a ground-based cloud image or an air-based cloud image, b is the total radiation observed by satellite, and o is the total radiation observed on the ground. In the cloud layer dataset, only the ground-based cloud image or the air-based cloud image is marked, so the data in the cloud layer dataset can be expressed as a p (p, c), where c is the cloud state data, expressed in terms of cloud position, type and size, denoted as c(e, k, s), where e is the cloud position, including low altitude, middle altitude and high altitude, k is the cloud type, including cumulus, stratus, stratocumulus, cirrus, cirrocumulus, cirrostratus, cumulonimbus, etc., and s is the size, expressed in terms of cloud area.

[0058] S32: Training target detection and semantic segmentation model: Using the annotated cloud dataset, training the target detection model, and then accurately identifying the cloud state data in the ground-based cloud image and the air-based cloud image. In this embodiment, the VGG-16 network model fused with the residual network is used.

[0059] At the same time, a semantic segmentation model is trained to perform pixel-level segmentation on the image. In this embodiment, U-Net is used for pixel segmentation.

[0060] S33: Run the trained object detection and semantic segmentation model on the image to be tested to obtain the type information of the cloud layer in the image and the pixel-level segmentation result.

[0061] Combined with the established cloud size database, the corresponding size information is found according to the cloud category detected in the image. The actual size of the cloud in the real world is calculated by using the pixel size of the cloud in the image and the pre-trained size information.

[0062] In this embodiment, cloud recognition is performed on ground-based cloud images and air-based cloud images respectively. Specifically, a low-altitude cloud size dataset can be constructed using the historical dataset of ground-based cloud images, and then a target detection model and a semantic segmentation model for low-altitude clouds are trained. At the same time, a medium- and high-altitude cloud size dataset is constructed using the historical dataset of air-based cloud images, and then a target detection model and a semantic segmentation model for medium- and high-altitude clouds are trained. Finally, the type and size of low-altitude clouds and the type and size of medium- and high-altitude clouds are output respectively.

[0063] In other embodiments, the calculated actual size information of the cloud layer may also be output together with the original image to intuitively display the type and size of the cloud layer in the image.

[0064] S34: Compare the ground-based cloud image and the air-based cloud image, and integrate the cloud type and size at each pixel point one by one. For example, the cloud state data of each pixel point in the ground-based cloud image is represented by c li (e li , k li ,s li ), the cloud state data of each pixel in the airborne cloud image is represented by c si (e si , k si ,s si ), then the cloud state data of each pixel after integration is expressed as c i (c li , c si ).

[0065] In other embodiments, the precipitation rate for the current period and future periods may also be predicted based on the collected historical cloud data.

[0066] Using target detection and semantic segmentation to identify ground-based cloud images and air-based cloud images respectively can not only perform pixel-by-pixel image recognition to obtain more accurate cloud recognition data, but also the VGG-16 network model integrated with the residual network can improve the performance and generalization ability of the model, enhance the robustness of the model, support deeper structures, and have stronger scalability and migration.

[0067] S4: Use the constructed SVM-LSTM-based solar radiation attenuation prediction model to identify the current total solar radiation attenuation rate according to the cloud state data. Construct a radiation attenuation table based on the total solar radiation attenuation rate and cloud state data.

[0068] Among them, the construction and training methods of the solar radiation attenuation prediction model are as follows:

[0069] S41: Establish a solar radiation attenuation prediction model based on SVM-LSTM network. The construction method is as follows:

[0070] S411: Use partial least squares method to extract features from the historical data set of air-based cloud images and the historical data set of ground-based cloud images. For example, the historical data set is used as the input set X = {a1, a2, a3, ..., a q}, the corresponding features are used as the output set Y, and the corresponding feature data Y = {(x1, y1), (x2, y2), (x3, y3), ..., (x N ,y N )}, respectively extract the principal component t1 of the input set and the principal component u1 of the output set, and maximize the covariance of t1 and u1, establish the regression equation of X for t1 and the regression equation of Y for u1, and use the residual matrix of the two regression equations to continue to extract the principal component t1 of the input set k And the principal component u of the output set k , and establish the corresponding regression equation. When the extracted principal component meets the accuracy requirement, establish the input set X and output set Y respectively about the principal component t k and principal component u k The final regression equation is:

[0071]

[0072] Where M is the rank of X, w l is the lth eigenvector, P0 is the value of X for t k The regression equation, Q0 is Y for u k The regression equation, r l is the regression coefficient vector, P m is the residual matrix of the mth regression equation.

[0073] S412: Construct a support vector machine. Input the feature data Y into the support vector machine for global optimal solution, and then obtain the solar radiation attenuation prediction model. The final classification decision function of the support vector machine is:

[0074]

[0075] In the formula, α i is the Lagrange multiplier, N is the number of sample points, x i ,y i are the input features, τ is the hyperparameter, and b* is the optimal solution for displacement.

[0076] S413: Use LSTM network to optimize the parameters of support vector machine. In this embodiment, the final output result h t Expressed as:

[0077] h t =o t *tanh(C t )

[0078] o t =σ(W o [h t-1 , x t ]+b o )

[0079] Where c(t) is the state vector, W o is the parameter matrix, x t is the input, b o is the bias term.

[0080] S414: Add a residual network to the solar radiation attenuation prediction model and perform skip-layer connection on the LSTM network. The residual network is used to solve the gradient vanishing problem of the LSTM network and optimize the smoothness of the forward and backward propagation of the LSTM network.

[0081] S42: Randomly select N historical data from the historical big data, and run the target detection and semantic segmentation model to obtain N cloud state data. The ratio of the ground observation total radiation data to the satellite observation total radiation data in the N historical data is used as the attenuation rate data, and the attenuation rate data and the cloud state data constitute the attenuation rate data set. The calculation formula of the attenuation rate can be expressed as: φ = O / B, where O is the ground observation total radiation data and B is the satellite observation total radiation data.

[0082] S43: Divide the attenuation rate data set into a training set and a test set according to a preset ratio, use the training set to train the solar radiation attenuation prediction model, use the test set to verify the solar radiation attenuation prediction model, and retain the model parameters that meet the test accuracy. In this embodiment, the attenuation rate data set is divided into a training set and a test set at a ratio of 8:2.

[0083] S44: Input the cloud state data collected and identified in real time into the trained solar radiation attenuation prediction model to obtain the corresponding attenuation rate data. t Then, the attenuation rate φ t Substitute the satellite observation total radiation data into the attenuation rate calculation formula to calculate the ground observation total radiation data.

[0084] In addition, the trained solar radiation attenuation prediction model is run on historical big data together with the target detection and semantic segmentation models to obtain the cloud status data and corresponding attenuation degree in each ground-based cloud map and air-based cloud map, and then integrated to obtain the radiation attenuation table.

[0085] S5: Use the extrapolation method to predict the cloud state change data and the satellite observation total radiation change data. According to the cloud state change data, query and obtain the corresponding total solar radiation attenuation rate in the radiation attenuation table. According to the satellite observation total radiation change data, calculate the future short-term ground observation total radiation forecast data.

[0086] In this embodiment, low-altitude cloud data, mid-altitude cloud data, high-altitude cloud data and satellite-observed total radiation data based on time series are recorded respectively, and the cloud state change data and satellite-observed total radiation change data of the future period are obtained by using the optical flow method. According to the cloud change data, the solar radiation attenuation rate of the future period is queried and obtained in the radiation attenuation table, and the predicted value of the ground-observed total radiation is obtained in combination with the corresponding satellite-observed total radiation change data. The formula is expressed as follows:

[0087] O t =∑φ ti B t . i=1,2,…,mn.

[0088] φ ti =α ti *β ti *γ ti

[0089] In the formula, O t is the total ground-observed radiation at time t, B t is the total satellite radiation corresponding to each pixel at time t. Since the total satellite radiation in the same area is basically the same, the total satellite radiation corresponding to each pixel is set to the same value. mn is the total number of pixels, and α tiis the solar radiation attenuation rate of the i-th pixel in the low-altitude cloud layer at time t, β ti is the solar radiation attenuation rate of the ith pixel in the hollow cloud layer at time t, γ ti is the solar radiation attenuation rate of the high-altitude cloud layer at the ith pixel at time t, where, when there is no cloud, α ti , β ti , γ ti =1.

[0090] In this embodiment, since the time resolution of the air-based cloud image is 15 minutes, the time resolution of the ground-based cloud image is 5 minutes, and the spatial resolution is 4KM, the satellite-ground combined ultra-short-term solar radiation prediction method of this embodiment can achieve a rolling forecast of 15 minutes.

[0091] In summary, the satellite-ground combined ultra-short-term solar radiation prediction method of this embodiment realizes satellite-ground combined observation of clouds at different altitudes by combining the observations of geostationary satellites and all-sky imagers, and fully considers the influence of clouds in the entire atmospheric column on solar shortwave radiation. At the same time, target detection and semantic segmentation are used to identify ground-based cloud images and air-based cloud images respectively, which can not only perform pixel-by-pixel image recognition to obtain more accurate cloud layer identification data, but also the VGG-16 network model integrating the residual network can improve the performance and generalization ability of the model, enhance the robustness of the model, support deeper structures, and have stronger scalability and migration. The partial least squares method is used to extract the features of cloud data in the cloud map, which not only retains important information in the data, but also simplifies the input data as a whole while maintaining the prediction accuracy, optimizes the calculation process, and improves the efficiency of solar radiation attenuation prediction.

[0092] Example 2: Please refer to Figure 1-Figure 3 This embodiment is based on the satellite-ground combined ultra-short-term solar radiation prediction method of Example 1, and proposes a satellite-ground combined ultra-short-term solar radiation prediction system, including a collection module, a preprocessing module, a cloud recognition model, a solar radiation attenuation prediction model and a solar radiation prediction module.

[0093] The acquisition module is used to collect ground-based cloud images, air-based cloud images, satellite observation total radiation data, and obtain historical data based on solar radiation. Specifically, the acquisition module may include geostationary satellites, all-sky imagers, ground observation stations, and big data collectors. Among them, the ground-based cloud images are obtained using the observation results of the Huayun all-sky imager, and the air-based cloud images and satellite observation total radiation data can be observed with the help of the Fengyun-4 geostationary satellite cloud product. The historical big data based on solar radiation includes ground-based cloud image historical data, air-based cloud image historical data, satellite observation total radiation historical data, and ground observation total radiation data, all of which can be obtained through the big data collector. Moreover, the satellite observation total radiation data can be observed and stored with the help of the Fengyun-4 geostationary satellite cloud product, and the ground observation total radiation historical data can be observed and stored through the ground radiation observation station.

[0094] The preprocessing module is used to preprocess ground-based cloud images and air-based cloud images. The preprocessing methods include geometric correction, radiation correction, atmospheric correction, image stitching, and enhancement processing.

[0095] The cloud recognition model is used to identify the types and sizes of low-altitude clouds, mid-altitude clouds, and high-altitude clouds based on ground-based cloud images and air-based cloud images. In this embodiment, the target detection and semantic segmentation model of deep learning is used to estimate the actual type and size of the cloud layer. The specific method is as described in Example 1.

[0096] The solar radiation attenuation prediction model is used to identify the current total solar radiation attenuation rate based on the preprocessed low, medium and high cloud conditions and ground-observed total radiation data, and to construct a radiation attenuation table based on the total solar radiation attenuation rate and the corresponding cloud condition data.

[0097] The solar radiation prediction module is used to predict the cloud state change data and the satellite observation total radiation change data, and to query and obtain the corresponding total solar radiation attenuation rate in the radiation attenuation table based on the cloud state change data, and to calculate the future short-term ground observation total radiation prediction data based on the satellite observation total radiation change data.

[0098] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A satellite-ground combined ultra-short-term solar radiation prediction method, characterized in that: The steps include: S1: Collect ground-based cloud images, air-based cloud images, satellite observation total radiation data, and obtain historical big data based on solar radiation; S2: preprocessing the ground-based cloud image and the air-based cloud image; S3: obtaining cloud state data based on the constructed cloud recognition model; the cloud state data includes the types and sizes of low-altitude clouds, mid-altitude clouds, and high-altitude clouds; S4: Using the constructed solar radiation attenuation prediction model based on SVM-LSTM, the current total solar radiation attenuation rate is identified according to the cloud state data; a radiation attenuation table is constructed according to the total solar radiation attenuation rate and the cloud state data; the construction and training method of the solar radiation attenuation prediction model is as follows: S41: Establish a solar radiation attenuation prediction model based on SVM-LSTM network; the construction method of the solar radiation attenuation prediction model is as follows: S411: Use partial least squares method to extract features from the historical data sets of air-based cloud images and ground-based cloud images; S412: construct support vector machine; S413: Using LSTM network to optimize the parameters of support vector machine; S414: Add a residual network to the solar radiation attenuation prediction model and perform skip-layer connection on the LSTM network; S42: randomly selecting N historical data from the historical big data, and running a target detection and semantic segmentation model to obtain N cloud state data; taking the ratio of the ground observation total radiation data to the satellite observation total radiation data in the N historical data as attenuation rate data, and the attenuation rate data and the cloud state data constitute an attenuation rate data set; S43: dividing the attenuation rate data set into a training set and a test set according to a preset ratio, using the training set to train the solar radiation attenuation prediction model, using the test set to verify the solar radiation attenuation prediction model, and retaining model parameters that meet the test accuracy; S44: inputting the cloud state data collected and identified in real time into the trained solar radiation attenuation prediction model to obtain corresponding attenuation rate data; S5: Use the extrapolation method to predict the cloud state change data and the satellite observation total radiation change data, query and obtain the corresponding total solar radiation attenuation rate in the radiation attenuation table according to the cloud state change data, and calculate the future short-term ground observation total radiation forecast data according to the satellite observation total radiation change data.

2. The satellite-ground combined ultra-short-term solar radiation prediction method according to claim 1, characterized in that: In step S2, the preprocessing method includes: Geometric correction, used to correct the geometric distortion of ground-based cloud images and air-based cloud images; Radiation correction is used to correct radiation distortion and adjust the image grayscale values ​​of ground-based cloud images and air-based cloud images; Atmospheric correction is used to eliminate the effects of atmospheric particle scattering and absorption on ground-based and air-based cloud images; Image stitching, which is used to stitch ground-based cloud images or air-based cloud images to generate image products covering the entire observation area; Enhancement processing is used to enhance the visual quality and information discernibility of ground-based cloud images and air-based cloud images.

3. The satellite-ground combined ultra-short-term solar radiation prediction method according to claim 1, characterized in that: In step S3, a deep learning-based object detection and semantic segmentation model is used to identify the actual type and size of the cloud layer. The specific steps are as follows: S31: randomly selecting M historical data from the historical big data, and marking the types and sizes of the clouds therein to form a cloud data set with labels; S32: constructing a target detection and semantic segmentation model based on deep learning; using the cloud layer dataset to train the target detection and semantic segmentation model, and retaining model parameters that meet the test accuracy; S33: Run the trained object detection and semantic segmentation models on the ground-based cloud image and the air-based cloud image to be tested to obtain pixel-level segmentation results based on cloud type and size; S34: Compare the ground-based cloud image and the air-based cloud image, and integrate the cloud type and size at each pixel point one by one.

4. The satellite-ground combined ultra-short-term solar radiation prediction method according to claim 1, characterized in that: In step S5, the predicted value of the ground-observed total radiation is expressed as: ABOUT t =∑φ ti B t ;i=1,2,…,mn; f ti =a ti *b ti *c ti In the formula, O t is the total ground-observed radiation at time t, B t is the total satellite radiation corresponding to each pixel at time t, mn is the total number of pixels, α ti is the solar radiation attenuation rate of the i-th pixel in the low-altitude cloud layer at time t, β ti is the solar radiation attenuation rate of the ith pixel in the hollow cloud layer at time t, γ ti is the solar radiation attenuation rate of the i-th pixel in the high-altitude cloud layer at time t.

5. A satellite-ground combined ultra-short-term solar radiation prediction system, which adopts a satellite-ground combined ultra-short-term solar radiation prediction method as claimed in any one of claims 1 to 4, characterized in that: include: The acquisition module is used to collect ground-based cloud images, air-based cloud images, satellite observation total radiation data, and obtain historical big data based on solar radiation; A preprocessing module, used for preprocessing the ground-based cloud image and the air-based cloud image; A cloud layer recognition model, used to obtain cloud layer status data according to the ground-based cloud image and the air-based cloud image; A solar radiation attenuation prediction model, used to identify the current total solar radiation attenuation rate based on the cloud state data; And constructing a radiation attenuation table according to the total solar radiation attenuation rate and the corresponding cloud state data; The solar radiation prediction module is used to predict the cloud state change data and the satellite observation total radiation change data, and query and obtain the corresponding solar radiation total attenuation rate in the radiation attenuation table according to the cloud state change data, and calculate the future short-term ground observation total radiation prediction data according to the satellite observation total radiation change data.

Images