Method and device for determining an albedo of the ground and / or an irradiance of radiation reflected by the ground, in a solar installation

Abstract

The invention relates to a method and a device for determining an albedo of the ground and / or an irradiance of radiation reflected by the ground (14), in a solar installation (12), by means of at least one device (30, 31, 54) for capturing ground-based measurement data, comprising at least one camera (32) which captures an image or a plurality of images of the ground (14) and an evaluation unit (42) which evaluates the ground-based measurement data using satellite image data. The invention also relates to a computer program for determining an irradiance of radiation reflected by the ground, to a data processing device, and to trained machine-learning models for extracting features.

Description

[0001] Description

[0002] title

[0003] Method and device for determining a ground albedo and / or an irradiance of radiation reflected from the ground in a solar installation

[0004] State of the art

[0005] The invention relates to a method and a device for determining a ground albedo and / or an irradiance of radiation reflected from the ground in a solar energy system. The invention further relates to a computer program, a data processing device, and machine learning models for determining a ground albedo and / or an irradiance of radiation reflected from the ground in a solar energy system.

[0006] For yield forecasting, control, and monitoring of solar power systems, knowledge of the solar resource at a specific location is of great interest. Examples of solar power systems include photovoltaic power plants, abbreviated to PV power plants below, which use fixed or tracking flat-plate collectors or photovoltaic modules as their collector surface. The solar resource is determined by the global irradiance of solar radiation and its components, direct radiation and diffuse radiation, as well as the radiation reflected from the ground. To increase yield efficiency, PV power plants with bifacial modules are typically used, in which the radiation striking both the front and back of the module is utilized and converted. Knowing the component of the radiation reflected from the ground is of great interest, especially for bifacial PV power plants.The solar resource can be determined through measurements from ground-based monitoring systems and satellite-based data. Monitoring systems for solar power plants, typically ground-based measurement systems for determining the solar resource at the planned site of a solar power plant, are often equipped with sky-facing pyranometers and, in some cases, additionally with ground-facing pyranometers to measure the incoming radiation. These measurements are used to evaluate the performance of a solar power plant or to assess the solar resource at the site.

[0007] From the publication “Intercomparison of Surface Albedo Retrievals from MISR, MODIS, CGLS Using Tower and Upscaled Tower Measurements”, Rui Song 1 , Jan-Peter Muller, Said Kharbouche and William Woodgate, Remote Sensing, 2019, 11 ,pp. 644 - 664 it is known to determine a temporal and spatial distribution of the albedo of the Earth’s surface using conventional satellite observations.

[0008] Disclosure of the invention

[0009] The object of the invention is to provide an improved method for determining a ground albedo and / or an irradiance of radiation reflected at the ground in a solar installation.

[0010] A further object is to provide a device for carrying out such a method.

[0011] A further object is to provide a computer program for determining a ground albedo and / or an irradiance of radiation reflected from the ground in a solar energy system. A further object is to provide a data processing system for determining a ground albedo and / or an irradiance of radiation reflected from the ground in a solar energy system.

[0012] Another task is to provide a trained machine learning model for extracting features and / or structures from camera images.

[0013] Another task is to provide a trained machine learning model for determining features and / or structures from measured data of satellite images.

[0014] Another task is to provide a trained machine learning model for determining features and / or structures from global irradiance measurement data.

[0015] Another task is to provide a trained machine learning model for determining the irradiance of radiation reflected at the ground.

[0016] The objects are achieved by the features of the independent claims. Advantageous embodiments and advantages of the invention emerge from the further claims, the description, and the drawings.

[0017] According to one aspect of the invention, a method is proposed for determining a ground albedo and / or an irradiance of radiation reflected from the ground in a solar energy system, in particular on a rear module side of at least one solar energy module of the solar energy system. The method is carried out using at least one device for acquiring ground-based measurement data, comprising at least one camera that captures one or more images of the ground, and an evaluation unit that evaluates the ground-based measurement data using satellite image data.

[0018] It is advantageous to suitably combine measurements and / or image information from ground-based and satellite-based data sources in order to exploit the advantages of the respective data sources, in particular large spatial coverage and high accuracy. In this case, the ground-based data sources are at least one camera.

[0019] Furthermore, it may be advantageous to compare features and / or structures from measurement data of satellite images and features and / or structures from images of a camera and to determine a spatially high-resolution albedo for the entire solar power plant, in particular an entire power plant, by a suitable assignment of the features and / or structures from a spatially high-resolution albedo determined in the limited field of view of the camera.

[0020] The camera image or images can be evaluated to determine at least a local ground albedo and / or an irradiance of radiation reflected by the ground in the camera's field of view. Using the satellite data and the ground albedo determined using the camera image, the spatially resolved ground albedo and / or the irradiance of radiation reflected by the ground for the area of ​​the solar energy system can be generated. The camera image or images can be evaluated to determine at least the local ground albedo and / or the irradiance of radiation reflected by the ground for a surface type in the camera's field of view. Using the satellite image data, each point on the solar energy system can be assigned to a surface type.Based on the surface type, the ground albedo determined by a camera and / or the irradiance of radiation reflected from the ground can be assigned to at least one point in the solar energy system for the surface type present at that point. Such an assignment is also referred to as "correlating" in this process.

[0021] In this process, either the same surface types can be detected in the camera image and the satellite image, or a known association between surface types in the camera image and surface types in the satellite image can be used. In the method, a surface type can refer to a set of visual, particularly textural and / or color, features that are suitable for recognizing areas with similar ground albedo between different points in time and at different locations.

[0022] The solar technology system can be a solar power plant, in particular a PV power plant with photovoltaic modules, in particular a bifacial PV power plant, wherein typically a module front and a module rear can be used to convert solar radiation into electrical energy. Advantageously, the ground albedo and / or the irradiance of the radiation reflected by the ground can be determined throughout the entire solar technology system. The solar radiation incident on the front and / or rear of the photovoltaic modules comprises in particular the components direct radiation, diffuse radiation, and radiation reflected by the ground. The rear module side can be a module side facing away from the sun. The rear module side can be a module side oriented essentially vertically. The rear module side can be a module side facing the ground.The rear side of the module can be arranged in a horizontal plane.

[0023] A typical device design in the simplest embodiment may comprise a camera with an approximately hemispherical field of view, in particular an approximately hemispherical fisheye camera, which takes images of the ground. Such a hemispherical field of view is also referred to as a field of view of approximately 180°. The camera, in particular the fisheye camera, can be positioned at a location where as many of the surface types encountered at the site as possible are within the camera's field of view. The camera, in particular the fisheye camera, can advantageously be leveled approximately horizontally.

[0024] Satellite imagery can originate from satellite-based data sources and is referred to as satellite-based imagery. Satellite imagery can be images captured from space, typically by satellites. In addition to satellite imagery, aerial imagery can optionally be used.

[0025] Satellite imagery is also known as remote sensing data and can be acquired from a satellite stationed in low Earth orbit (LEO). The satellite imagery can advantageously be high-resolution satellite images with a resolution significantly better than 10 m x 10 m, especially better than 5 m x 5 m, and especially better than 1 m x 1 m. It is advantageous if the resolution is significantly finer than the spacing between the modules and the length and width of the photovoltaic module rows. This allows the satellite imagery to distinguish between the ground and the modules.

[0026] Satellite image data can be used to determine the surface characteristics of the soil types visible in the solar energy system, particularly around the modules. Since the surface characteristics of the soil on which the modules are arranged can influence the intensity of the radiation reflected by the ground and / or the soil albedo, it is important to know these characteristics. It is advantageous that satellite image data can be used to distinguish between the soil on which the modules are arranged and components of the solar energy system, such as modules, mounting brackets, etc.

[0027] In an advantageous embodiment, at least one radiation sensor unit can additionally be used in the method. The radiation sensor unit can comprise a radiation sensor or be a pyranometer for measuring the global irradiance. The radiation sensor unit can detect the global horizontal irradiance (GHI). An extension of the typical design of the device can additionally comprise at least one sky-facing radiation sensor unit or a sky-facing pyranometer.

[0028] Since the proportion of solar resources, particularly the irradiance, on the rear side of the module typically accounts for at least 5%–20% of the annual yield in bifacial PV power plants, it can be advantageous to measure the radiation reflected from the ground that hits the rear side of the module (rPOA: rear-side plane-of-array). The irradiance on the rear side of the module (rPOA irradiation) can be the combined solar irradiance with the two components diffuse radiation, which hits the rear side of the module directly from the sky, and radiation reflected from the ground. An error in determining the radiation hitting the rear side of the module can therefore have a significant impact on the modeling of the yield forecast, and it is advantageous to measure this directly.

[0029] Using state-of-the-art methods, determining the temporal and spatial variation of ground-reflected radiation within the partially built-up area of ​​a PV power plant has so far been difficult and subject to significant measurement errors. Current methods are problematic because they either only detect ground-reflected radiation point by point and / or are compromised by structures in the field of view of the respective measuring system and / or can only capture rough spatial averages and / or are compromised by atmospheric influences on the measurement.Since the radiation reflected from the ground can typically vary over time, often due to the non-uniform soil conditions over a large power plant area and / or due to the lack of information on the soil conditions or changing soil conditions with the seasons, a current determination of the ground albedo or ground-reflected radiation is necessary for each evaluated period.

[0030] Soil albedo is a useful measure of soil reflectivity. To illustrate this, some examples of the reflectivity of solar radiation by different soil types are given here. For example, dry grassland has a different reflectivity and soil albedo than wet grassland; grassland can generally have a lower reflectivity and a lower soil albedo than sandy soil. Dark arable soil, in turn, can have an even lower soil albedo. In addition, reflectivity and soil albedo differ depending on the wavelength range being measured (e.g., visible wavelength range versus infrared wavelength range). Spectral reflectivity varies not only in magnitude but also in shape depending on soil properties.

[0031] State-of-the-art methods, summarized, for example, in IEC 61724-1 (IEC 61724-1:2021 Photovoltaic system performance - Part 1: Monitoring), propose the use of twelve radiometers to determine the reflected irradiance (rPOA) incident on the rear of the module for a 100 MW PV power plant. Even more radiometers are recommended for uneven soil conditions. Thus, the proposed method, using a measurement system with two sensors—the radiation sensor and the camera—can offer the advantage of lower capital expenditure, significantly reduced measurement and data analysis effort, and lower costs.

[0032] If pyranometers are used in the state-of-the-art method, they can measure the global irradiance from the half-space above a pyranometer sensor plane quite accurately. If the pyranometer is aligned with the ground, the ground-reflected radiation can thus be measured directly. However, this measured value is specific to the exact conditions at the sensor location. Determining the ground-reflected radiation for each module back in a PV power plant requires inaccurate conversions. For such conversions, for example, the composition of the radiation reflected from the ground is estimated using visibility factors and a measured GHI using models. In this process, it has proven difficult to distinguish between the radiation reflected from the ground in shaded and unshaded areas, as well as from areas with different ground conditions, in the field of view of the back of a PV module.Because the proposed method can use both satellite image data and camera images, it is possible to distinguish between shaded and unshaded areas and different soil conditions.

[0033] The measuring system of the device used in the method according to the invention is also much easier to handle. Furthermore, the proposed method can be more cost-effective than using drones equipped with a pyranometer and a camera to measure the radiation reflected from the ground.

[0034] A further advantage of the method according to the invention, compared to the satellite data used in the prior art for determining the radiation reflected by the ground, as described, for example, in NASA (2023) MODIS: Moderate Resolution Imaging Spectroradiometer, 2023, is that a distinction can be made between the structures of the PV modules and the ground. This is because satellite image data with a resolution of better than 10 mx 10 m, in particular better than 5 mx 5 m and, advantageously, better than 1 mx 1 m can be used. Only in this way can a ground albedo be determined in partially built-up areas, such as PV power plants. In the comparatively coarse-resolution satellite data used to date, both ground surfaces and surfaces of structures, in particular PV modules, are included in the measured albedo.Unmixing techniques could be used to distinguish between ground albedo and the albedo of the observed structures, but significant errors are to be expected. In contrast, our method can exclude built-up areas from the evaluation. A ground albedo is thus determined for the unshaded spaces between structures. This avoids the strong interference described. Furthermore, a ground albedo with significantly better resolution can be obtained, thus determining the ground reflectivity more accurately. This also allows the irradiance of the radiation reflected from the ground to be determined more accurately.

[0035] According to an advantageous embodiment of the method, the following steps can be included:

[0036] (i) capturing one or more images of the ground with the camera in a field of view of the camera directed towards the ground;

[0037] (ii) extracting features and / or structures and / or different surface types from the image or multiple images of the ground using a first machine learning model and generating a first result data set;

[0038] (iii) determining a ground albedo for each of the features and / or structures and / or surface types in the camera image or images;

[0039] (iv) extracting features and / or structures and / or different surface types from the provided satellite image data, in particular image data of a satellite, by means of a second machine learning model and generating a second result data set;

[0040] (v) generating a joint data set from the second result data set of the satellite image data and the first result data set of the analyzed camera images; (vi) assigning at least one feature and / or structure and / or surface type from the image or multiple images of the camera to a feature and / or structure and / or surface type determined from satellite image data;

[0041] (vii) Determining the ground albedo for each point in the power plant by correlating the ground albedo corresponding to the feature and / or structure and / or surface type in the camera image or images with features and / or structures and / or surface types determined from the satellite imagery at each point in the satellite imagery.

[0042] A ground albedo can be determined for each of the features and / or structures and / or surface types in the camera image(s). Tabular values ​​can be used as fallback values.

[0043] The first machine learning model can advantageously use the same algorithm as the second machine learning model.

[0044] The first machine learning model can be trained with first training data sets, in particular with reference data. The second machine learning program can be trained with first or second training data sets.

[0045] According to an advantageous embodiment of the method, the additional steps may include:

[0046] (i) Determining measurement data of the global irradiance with the at least one radiation sensor of the radiation sensor unit, wherein the global irradiance comprises diffuse radiation, and / or radiation reflected on the ground and / or direct radiation,

[0047] (ii) Determining the ground albedo for at least one of the surface types and / or features and / or structures in the camera image by a combined evaluation of measured global irradiance and image information from the camera.

[0048] The measurement data of the global irradiance of the at least one radiation sensor unit, in particular the at least one radiation sensor of the radiation sensor unit, can be determined in particular in a field of view in an approximately horizontal plane directed towards the sky, wherein the global irradiance comprises diffuse radiation and / or direct radiation and / or radiation reflected on the ground.

[0049] Measurement data of a hemispheric irradiance of at least one radiation sensor unit, in particular of at least one radiation sensor of the radiation sensor unit, can be determined, in particular, in a field of view in a plane of a rear module side, wherein the global irradiance comprises diffuse radiation, and / or radiation reflected from the ground, and / or direct radiation. Such measurement data can advantageously be combined with the image information from the camera. The measurement data can be used, in particular, for radiometric corrections of the image information from the camera.

[0050] According to an advantageous embodiment, the method may additionally comprise the step:

[0051] Determining the radiation reflected from the ground for at least one point of the solar installation from the ground albedo and the global irradiance measured by the radiation sensor unit. In an advantageous embodiment of the method, the following additional step may be included:

[0052] Assigning a ground albedo to surface types in the satellite image data to which no surface type could be assigned in the camera image, in particular by relying on empirical values.

[0053] Alternatively or additionally, table values ​​can be used.

[0054] In an advantageous embodiment of the method, the following step may additionally be included:

[0055] Assigning surface types to areas in the satellite image data for which no surface type could be assigned, based on surface types of surrounding image areas, particularly using interpolation or inpainting techniques. This step can be used in addition to or as an alternative to empirical values ​​or tabulated values.

[0056] According to an advantageous embodiment of the method, features and / or structures and / or surface types can be extracted from the measurement data of the radiation sensor unit by means of a third machine learning model and a further result data set can be generated, wherein the third machine learning model is trained with third training data sets, in particular with reference data.

[0057] According to an advantageous embodiment of the method, the camera's result data set and the satellite image data can be combined into a single data set. According to an advantageous embodiment of the method, the radiation sensor's result data set, the camera's result data set, and the satellite image data can be combined into a single data set.

[0058] According to an advantageous embodiment of the method, the satellite image data can be preprocessed and / or analyzed using a machine learning model, or raw data. The raw data can be analyzed in the evaluation unit using an additional machine learning model.

[0059] According to an advantageous embodiment of the method, a convolutional neural network (CNN) algorithm can be used in the first machine learning model and / or the second machine learning model and / or the machine learning model for the satellite image data.

[0060] According to an advantageous embodiment of the method, the radiation sensor unit can have at least one pyranometer and / or a reference cell.

[0061] According to an advantageous embodiment of the method, the ground-based measurement data can be determined by a radiation sensor unit directed towards the sky, in particular a global radiation sensor directed towards the sky, and the camera which is directed essentially towards the ground.

[0062] According to an advantageous embodiment of the method, the ground-based measurement data can be acquired by a global radiation sensor directed toward the ground and the camera, which is essentially directed toward the ground. According to an advantageous embodiment of the method, the camera can cover a spatial field of view of approximately 180° and can capture images of the ground.

[0063] The spatial field of view can be a hemispherical field of view and cover an almost circular section of the ground.

[0064] According to an advantageous embodiment of the method, areas in the camera images and / or the satellite image data can be classified into surface types, whereby surface types are differentiated between areas on the ground and elements of the solar installation, whereby at least two of the following categories are distinguished:

[0065] (iii) unshaded ground,

[0066] (iv) shaded ground,

[0067] (v) soil surface,

[0068] (vi) Superstructures.

[0069] The ground surface can, in turn, be classified into different surface types, in particular vegetation types and / or sand and / or snow and / or bare soil, for example, arable land. The structures can be the PV modules themselves, PV module mounts, buildings, fences, and / or sensor mounts.

[0070] According to an advantageous embodiment of the method, image pixels in the satellite image data can be assigned to marker points or points in the solar power plant; in particular, the georeferencing method can be used for this purpose. According to an advantageous embodiment of the method, image pixels in the satellite image data can be assigned to marker points or points in the solar power plant. The accuracy of the georeferencing is increased by image transformations that reduce the deviations between known positions of objects in the PV power plant and their detected positions in the satellite image.

[0071] Georeferencing is a well-known process by which maps and satellite images can be assigned to a geospatial coordinate system by assigning coordinate values. This can be particularly necessary when digital maps, for example, captured by a satellite, lack a reference to a geospatial coordinate system. Typically, the spatial information, the georeference, can be assigned to a dataset. Positions of structures or marker points identified in the satellite image (i.e., image coordinates) can also be compared with the expected positions of these objects (i.e., geographical coordinates, for example, from the technical documentation of the facility) to increase the accuracy of the georeferencing.

[0072] According to an advantageous embodiment of the method, a surface type, in particular a surface type determined from the satellite image data, can be assigned a ground albedo that corresponds to a ground albedo determined from the ground-based measurement data or that was determined from the ground-based measurement data. According to an advantageous embodiment of the method, a spatially resolved ground albedo and its distribution in the solar energy system and / or an average value for the solar energy system can be determined from the satellite image data and camera images, in particular by means of the evaluation unit. The ground albedo can have an areal resolution.

[0073] According to an advantageous embodiment of the method, intensity values ​​for each color channel in the camera image can be evaluated in order to derive the ground albedo of at least one surface type, in particular using a radiometric camera model to derive radiation received by the camera and reflected by the ground from the intensity values.

[0074] According to an advantageous embodiment of the method, intensity values ​​for each color channel in the camera image can be evaluated in combination with measured values ​​from the radiation sensor unit in order to derive the ground albedo of at least one surface type.

[0075] According to an advantageous embodiment of the method, a demixing method can be used in the evaluation unit in order to assign a ground albedo to at least one surface type determined from the image of the camera and / or from the satellite image data from the ground albedo, which is calculated in particular from one or more measured values ​​of a global radiation sensor directed towards the ground and from one or more measured values ​​of the radiation sensor unit;According to an advantageous embodiment of the method, a demixing method can be applied in the evaluation unit, which uses an image from the camera and / or surface types recognized from an image from the camera in order to assign a ground albedo to at least one surface type determined from the image from the camera and / or from the satellite image data from the ground albedo, which is calculated in particular from one or more measured values ​​from a global radiation sensor directed towards the ground and from one or more measured values ​​from the radiation sensor unit.

[0076] According to an advantageous embodiment of the method, a spectral correction of the image data can be carried out based on color channels of the camera, whereby a broadband and / or spectrally weighted value of the ground albedo can be determined.

[0077] The broadband albedo can correspond to the ratio of the broadband ground-reflected horizontal radiation, which can be measured with a horizontally leveled thermopile pyranometer directed towards the ground, to the broadband global radiation in the horizontal plane (GHI), which can be measured with a horizontally leveled thermopile pyranometer directed towards the sky. Compared to measurements with two pyranometers, the method allows a broadband albedo to be measured for each surface type present in the camera image. Furthermore, a spectrally weighted ground albedo can be determined according to a user-defined spectral weighting. The spectral response of a PV module, in particular, can be used as the spectral weighting.The correction of the image data to determine a broadband and / or spectrally weighted value of the ground albedo can be determined, in particular, by experimentally comparing the measurement data of one or more ground-directed spectroradiometers (e.g., a combination of EKO MS-711 and MS-712 from the source: https: / / www.eko-instruments.com / eu / categories / products / spectroradiometers / wiser-i-spectroradiometer) with the image data from the camera. In particular, a ground-reflected horizontal spectrally weighted or broadband radiance calculated from the spectroradiometer data can be compared with the corresponding measured value based on the data from each color channel of the camera. In particular, an empirical spectral correction function can be determined, which uses the ratios of the measured values ​​based on the data from each color channel of the camera as input variables and outputs a multiplicative correction factor.This correction function can in particular be applied as a piecewise linear function.

[0078] According to an advantageous embodiment of the method, spatially high-resolution satellite images with a resolution significantly finer than 10 mx 10 m can be used as satellite image data.

[0079] It is advantageous that the satellite image data are sufficiently high-resolution to distinguish structures such as photovoltaic modules from the subsurface. According to an advantageous refinement of the method, temporal, for example, seasonal, changes in the ground albedo for each surface type can be taken into account by modeling the influence of the sun's position and the proportion of diffuse radiation in global radiation on the ground albedo.

[0080] According to an advantageous embodiment of the method, the growth cycle of plants and / or weather-related processes can be modeled in order to model temporal changes in the surface types detected in the satellite image data.

[0081] According to an advantageous embodiment of the method, the camera can take a picture continuously, at least once a day.

[0082] This allows changes in surface types to be recorded promptly. Even more advantageously, the camera can take images several times a day. This allows changes in surface albedo depending on the position of the sun to be recorded.

[0083] According to an advantageous embodiment of the method, at least one of the variables from a ratio of broadband radiation to the portion of radiation registered by the camera, and / or an intensity of RGB channels of the camera, and / or an internal and / or external calibration of the camera, and / or an inclination and orientation of the camera sensor, and / or an inclination and orientation of an inclined plane, and / or the position of the sun during the radiation measurement and / or a camera sensitivity, which is determined from an illuminance of the camera and / or the spectral sensitivity of the RGB channels and / or recording settings and / or the RGB camera image and / or the internal and / or external calibration of the camera, can be used to convert measured values ​​from the camera. The red-green-blue (RGB) color channels of the camera image can be summed in a weighted manner.The weighting of the channels ensures the camera's sensitivity is as uniform as possible in the visible wavelength range. This gray value can be multiplied by a broadband correction to account for radiation at wavelengths outside the camera's measurement range.

[0084] According to a further aspect of the invention, a device for carrying out a method according to the invention is proposed, which device comprises at least one camera and an evaluation unit, wherein the evaluation unit is configured to carry out an evaluation of ground-based measurement data of the camera using satellite image data.

[0085] This allows the ground albedo in a solar energy system to be determined using simple means. Reflected radiation on the rear side of a module can be measured cost-effectively. The use of satellite-based data, especially satellite data, allows for the determination of a better-resolved ground albedo distribution and the differentiation between structures and the ground. The advantages listed for the method also apply to the device.

[0086] According to an advantageous embodiment of the device, the camera can be designed such that at least the following properties are present: a single image is recorded in a fixed time grid, in particular every half and full minute; the at least one sensor of the camera has a constant color temperature; the camera has a constant exposure time for each single image; for exposure control of the camera, a predetermined minimum value of an average image brightness is set, wherein a signal amplification applied by the camera remains unchanged at a higher image brightness, in particular wherein the predetermined minimum value of an average image brightness is preferably at most 10%, particularly preferably at most 8%, very particularly preferably at least 5%.

[0087] The signal amplification applied by the camera can be described by the ISO value applied.

[0088] According to an advantageous embodiment of the device, the camera can be designed to record the ground in the field of view, wherein the camera can in particular comprise a fisheye camera and / or a surveillance camera. The camera can in particular be designed to capture the entire field of view in one shot.

[0089] According to an advantageous embodiment of the device, the camera can be designed to capture the entire field of view in one shot. In particular, the camera can be designed as a surveillance camera and / or a fisheye camera.

[0090] According to an advantageous embodiment of the device, at least one radiation sensor unit with at least one sensor can additionally be provided, wherein the evaluation of the measurement data from the radiation sensor unit can be carried out using satellite images, in particular satellite images. According to an advantageous embodiment of the device, at least one sensor of at least one radiation sensor unit and the at least one sensor of the camera can each be arranged in a horizontal plane such that the field of view of the at least one sensor lies below the horizontal plane and is flush with the horizontal plane, so that a field of view directed towards the ground can be captured. The field of view is advantageously a spatial 180° field of view.

[0091] According to an advantageous embodiment of the device, the at least one radiation sensor unit and the camera can be coupled so that measurement data recording by the radiation sensor unit and the camera is synchronized in time.

[0092] According to an advantageous embodiment of the device, the at least one radiation sensor unit can comprise at least one of a pyranometer, in particular a thermopile pyranometer, also known as a thermopile pyranometer, a photodiode, or a photovoltaic reference cell.

[0093] According to an advantageous embodiment of the device, the at least one radiation sensor unit can be designed such that measurement data is recorded by the radiation sensor unit with a high temporal resolution, in particular with a temporal resolution of less than 10 see, preferably less than 5 see, particularly preferably less than or equal to 1 see.

[0094] According to an advantageous embodiment of the device, the radiation sensor unit can be designed to detect solar radiation in a wavelength range from 0.3 pm to 3 pm. According to an advantageous embodiment of the device, the camera can be designed to capture the entire field of view in one shot, in particular, the camera is designed as a surveillance camera and / or a fisheye camera.

[0095] According to a further aspect, a computer program for determining a ground albedo and / or irradiance of radiation reflected at the ground is proposed, which comprises instructions which, when the program is executed by a computer, cause the computer to carry out the steps of a method for determining the irradiance of solar radiation reflected at the ground.

[0096] According to a further aspect, a data processing device for a device for determining a ground albedo and / or irradiance of the solar radiation reflected on the ground is proposed, which comprises at least one measurement data acquisition unit, an evaluation unit and a computer.

[0097] According to a further aspect, a trained, in particular first machine learning model for extracting features, in particular structures and / or surface types, from an image taken by a camera and comprising information on the soil condition is proposed, which is used in the method according to the invention.

[0098] According to a further aspect, a trained, in particular second, machine learning model is proposed for extracting features and / or structures and / or surface types from at least one image acquired from satellite image data and containing information on the soil composition, which is used in the method according to the invention. According to a further aspect, a trained, in particular third, machine learning model is proposed for determining features, in particular structures, from measurement data of the global irradiance, which is used in the method according to the invention.

[0099] According to a further aspect, a trained, in particular fourth, machine learning model for determining an irradiance of solar radiation reflected on the ground is proposed, which has been trained in a supervised manner, using in particular input data and reference data from satellite image data, and which is used in the method according to the invention.

[0100] In summary, the method and / or the device and / or the computer programs with the device arranged in the area of ​​the solar power plant can have the technical configurations and advantages listed below in addition to the advantages mentioned above.

[0101] Advantageously, the fisheye camera can be located at a location where as many of the surface types encountered at the site as possible are within the camera's field of view. The camera can be leveled approximately horizontally. The sky-facing pyranometer can advantageously be constructed and maintained according to the specifications for horizontal global irradiance (GHI) measurements, which can be found, for example, in Task 16 Solar Resource - Best Practices Handbook for the Collection and Use of Solar Resource Data - 3rd Edition, Chapter 3.3, "MEASURING SOLAR RADIATION." The camera and pyranometer can advantageously be located close to each other, and both measurements can be synchronized.Additionally, a ground-facing pyranometer can be installed at the camera location so that the fields of view of the pyranometer and camera coincide as closely as possible. The pyranometer can be leveled approximately horizontally. For example, the setup can be realized by mounting the pyranometer and camera on a horizontal support at a height of 2 m above the ground, approximately 30 cm apart (center to center). A fisheye camera can also be installed at the pyranometer location, with the cloud camera being leveled approximately horizontally and having a clear field of view.The satellite image data can advantageously be high-resolution satellite data with a resolution significantly finer than the spacing between module rows, which are usually spaced 5 meters or less apart, whereby the satellite image data can be related to the area of ​​the power plant.

[0102] Using the sky-facing camera, the direct irradiance (DNI) and the diffuse irradiance (DHI) can be derived from the GHI. A decomposition model or another method can be used for this purpose. If a sky-facing camera is available at the same location, the method described in the publication by Blum, N.B., et al. (2022). "Measurement of diffuse and plane of array irradiance by a combination of a pyranometer and an all-sky imager." Solar Energy 232: 232-247, can be used to determine DNI and DHI. Furthermore, the sky radiance (radiance) can be used. Himmet can be estimated according to the usual calculation of so-called white-sky and black-sky albedo, as described in the publication LUCHT, Wolfgang; SCHAAF, Crystal Barker; STRAHLER, Alan H. "An algorithm for the retrieval of albedo from space using semiempirical BRDF models", . IEEE Transactions on Geoscience and Remote sensing, 2000, 38th year, no. 2, pp. 977-998.

[0103] The camera looking towards the sky can be used to conveniently determine the diffuse radiation and the radiance of the radiation coming from the sky.

[0104] The measurement of diffuse irradiance (DHI) via the sky-facing camera is more accurate than competing systems commonly used in practice.

[0105] Most importantly, the camera can be used to determine the radiance of the radiation originating from the sky from the camera image. This allows comparatively accurate information about the radiance of the sky to be obtained. This can be a significant advantage over a rather rough estimate of the radiance based on a GHI measurement or measurements of DNI and DHI. The advantage here is that it can increase the overall accuracy and robustness of the method.

[0106] The images from the ground-facing camera can be evaluated in combination with the sky's radiance to determine the ground albedo. All of the above-mentioned measuring devices can be connected to the evaluation device, which may have suitable data acquisition. The evaluation device can be connected to a server or incorporate the server as an integral component. For example, the radiation sensor unit, in particular the pyranometer, can be connected to a data logger. The camera and the data logger can independently upload the ground-based measurement data and / or the satellite image data to the server, for example, via the so-called SFTP protocol. The ground-based measurement data, the satellite image data, and the combined data files of partially evaluated or fully analyzed data can be stored on the server pending further processing.Ideally, the evaluation unit can have access to the usual resources of an edge computer. The evaluation unit can have a data connection to retrieve the satellite image data, such as remote sensing data, for example, via an internet connection.

[0107] drawing

[0108] Further advantages will become apparent from the following description of the drawings. The figures illustrate exemplary embodiments of the invention. The figures, the description, and the claims contain numerous features in combination. Those skilled in the art will also expediently consider the features individually and combine them into useful further combinations.

[0109] Examples include:

[0110] Fig. 1 shows a schematic diagram of a solar energy system with different soil types and soil conditions; Fig. 2 shows a schematic diagram of a section of the solar energy system with a structure of a measuring device for determining the radiation reflected from the ground according to an exemplary embodiment;

[0111] Fig. 3 shows schematically a section of the solar installation with a structure of a measuring device for determining the radiation reflected on the ground according to a further embodiment;

[0112] Fig. 4 shows schematically a section of the solar installation with a structure of a measuring device for determining the radiation reflected on the ground according to a further embodiment;

[0113] Fig. 5 shows schematically the structure of the measuring device according to the further embodiment;

[0114] Fig. 6 shows schematically the structure of the measuring device according to the further embodiment;

[0115] Fig. 7 is a flowchart of a method for determining the radiation reflected from the ground on a module;

[0116] Fig. 8 is a flowchart of an extended method for determining the ground reflected radiation on a module;

[0117] Fig. 9 is a more detailed flow diagram of a method for determining the ground reflected radiation on a module;

[0118] Fig.10 shows a more detailed flowchart of the procedure for determining the radiation reflected from the ground.

[0119] Embodiments of the invention

[0120] In the figures, components of the same type or with the same function are designated by the same reference numerals. The figures merely show examples and are not to be understood as limiting. Directional terminology used below, including terms such as "left," "right," "top," "bottom," "before," "behind," "after," and the like, serves only to improve understanding of the figures and is in no way intended to limit the scope of the invention. The components and elements illustrated, as well as their design and use, may vary according to the considerations of a person skilled in the art and may be adapted to the respective applications.

[0121] Figure 1 shows a schematic representation of modules 10, in particular photovoltaic modules 10 of a solar installation 12, which are installed on a substrate 14 or ground 14 with different soil conditions 16, 18, 20. The

[0122] Soil conditions 16, 18, 20 are shown geometrically as examples. The arrangement of the modules 10 is shown as an example. A satellite 22 records satellite image data of the entire solar energy system 12. The satellite image data contains data on the soil conditions 16, 18, 20. The rectangle 24 represents the size of a pixel 24, which corresponds to a resolution of the satellite image data. The resolution is significantly better than a distance 26 between two spaced-apart modules 10. A substantially circular area 28 represents a section of an image of the soil 14, which is recorded by a measuring device 30 shown in Figure 2 for determining radiation reflected on the soil 14 using a camera 32. The circle 28 represents the usable area of ​​the field of view of the camera 32 imaged onto the soil 14.

[0123] The modules 10 are typically photovoltaic modules 10 that have a front side 5 and a back side 11 (see Figure 2). The back side 11 is also referred to as the rear module side or module back. The front side 5 of the module 10 is also referred to as the module front. The modules 10 are, for example, bifacial photovoltaic modules 10, in which both the solar radiation striking the front side 5 and the radiation striking the back side 11 can be converted into electrical energy. The modules are spaced apart, typically at a distance 26 of 5 m from each other, mounted on the ground 14. The distance 26 can also be larger or smaller. The modules 10 can be fixed modules 10 or modules 10 whose angle to the respective incident solar radiation can be variably adjusted and thus trackable.

[0124] Figure 2 shows a schematic representation of a section of the solar system 12 with four modules 10 and a measuring device 30.

[0125] The measuring device 30 is a device for determining a ground albedo and / or an irradiance reflected from the ground. The measuring device 30 comprises a camera 32. The camera 32 has a field of view 34 directed toward the ground 14. The camera 32 is arranged on a mounting device 36 with a mounting arm 35 and is detachably connected thereto. The field of view 34 of the camera 32 covers the essentially circular section 28 and designates the usable field of view of the camera 32.

[0126] The measuring device 30 further comprises a radiation sensor unit 38, which is mounted on a second mounting device 40, which is in particular detachably connected to a mounting arm 39. The radiation sensor unit 38 typically comprises a pyranometer or a reference cell, by means of which a global irradiance comprising the components direct radiation, ground-reflected radiation, and diffuse radiation can be measured. The radiation sensor unit 38 has a field of view directed towards the sky. The global irradiance is measured in a field of view of approximately 180° above the plane in which the radiation sensor unit 38 is arranged. The camera 32 and the radiation sensor unit 38 are connected to an evaluation device 42 such that measurement data from the camera 32 and the radiation sensor unit 38 can be stored and analyzed there.Furthermore, the evaluation device 42 is configured to control the camera 32 and the radiation sensor unit 38.

[0127] The evaluation device 42 is further configured to receive and analyze satellite image data of the ground 14 acquired by the satellite 22. A machine learning model in the form of a computer program can run on the evaluation device 42, which analyzes the measurement data from the camera 32, for example, for structures. The measurement data from the camera 32 and the radiation sensor unit 38 are transmitted to the evaluation device 42 via a data transfer device 46.

[0128] The measurement data from the radiation sensor unit 38 are transmitted to the evaluation device 42 by means of a data transfer device 48. The measurement data from the camera 32 and the radiation sensor unit 38 are referred to as ground-based measurement data 49.

[0129] The radiation sensor unit 38 is provided for determining the irradiance of solar radiation in a field of view of 180° above the plane in which it is arranged, which is referred to as the mounting plane. The camera 32 is provided for capturing a field of view of 180° below a mounting plane of the camera 32, wherein the field of view is in each case a hemispherical field of view which extends below the mounting plane as a three-dimensional hemispherical hemisphere. The satellite images of the satellite 22 are typically stored in a data storage device, for example in a cloud storage device 50, and are transmitted to the evaluation device 42 by means of a data transfer device 52. The data storage device 50 is an external data storage device 50, for example a cloud storage device, which digitally stores the satellite image data in distributed individual storage devices.

[0130] The camera 32 is designed to capture the entire field of view in a single shot. The camera 32 is designed as a surveillance camera and / or a fisheye camera. For example, the setup can be realized by mounting the radiation sensor unit 38 and the camera 32 on the mounting arm 35 of the mounting device 36 at a height of 2 m above the floor 14 at a distance of approximately 30 cm from each other (center to center).

[0131] The radiation sensor unit 38 has at least one of a pyranometer, in particular a thermopile pyranometer, also known as a thermopile pyranometer, a photodiode, a photovoltaic reference cell and is designed to record measurement data of the radiation sensor unit 38 with a high temporal resolution, in particular with a temporal resolution of less than 10 see, preferably less than 5 see, particularly preferably less than or equal to 1 see.

[0132] Furthermore, the radiation sensor unit 38 is designed to detect solar radiation in a wavelength range from 0.3 pm to 3 pm. The evaluation unit 42 has a computer unit (not shown) that is configured to execute the computer program for evaluating the measurement data recorded by the camera 32 using a machine learning program. Furthermore, the evaluation unit 42 is configured to execute a machine learning program for evaluating the measurement data from the radiation sensor unit 38.

[0133] Figure 3 shows a schematic representation of a portion of the solar system 12 with four modules 10 and a measuring device 31. The measuring device 31 is a simplified measuring device 30 and has only the camera 32. The camera 32 has the field of view 34 directed toward the ground 14. The camera 32 is arranged on the mounting device 36 with the mounting arm 35 and is detachably connected thereto. The field of view 34 of the camera 32 covers the essentially circular section 28. The measuring device 31 is a device for determining the ground albedo and / or the global irradiance.

[0134] Figure 4 shows a schematic representation of a measuring device 54 that can be used as a supplement to the measuring devices 30 and 31 or as an alternative measuring device 54 for detecting the ground albedo and / or the global irradiance and solar radiation reflected on the ground in the solar energy system 12. The measuring device 54 has the camera 32 and the radiation sensor unit 38. The camera 32 is a first camera 32, and the radiation sensor unit 38 is a first radiation sensor unit 38. The first camera 32 and the first radiation sensor unit 38 are arranged on a mounting device 56. Additionally, a second camera 58 and a second radiation sensor unit 60 are arranged on the mounting device 56. A field of view 62 of the second camera 58 is directed towards the sky. A field of view 64 of the second radiation sensor unit 60 is directed towards the ground.The measurement data from camera 32, camera 58, radiation sensor unit 38, and radiation sensor unit 60 are transmitted to the evaluation unit 42 via a common data transfer device 66. The satellite image data are transmitted from the external memory 50 to the evaluation unit 42 via the data transfer device 52.

[0135] Figures 5 and 6 show the measuring device 54 in a schematic side view. The first camera 32, the second camera 58, the first radiation sensor unit 38, and the second radiation sensor unit 60 are arranged on the holder 56. The first camera 32 and the first radiation sensor unit 38 are arranged at a distance 68 from one another. The holder 56 has a substantially vertical mounting post 70 and a mounting arm 72 adjustable relative to a horizontal axis 74. In the embodiment shown in Figure 5, the mounting arm 72 is adjusted and fixed parallel to the horizontal axis 74. The camera 32, the camera 58, the radiation sensor 40, and the radiation sensor 60 are arranged and fastened on the mounting arm 72. The first camera 32 and the first radiation sensor unit 60 have a field of view 62 and field of view 64 directed downwards toward the floor 14.The second camera 58 and the first radiation sensor unit 38 have a field of view 63 and field of view 65 directed upwards towards the sky.

[0136] In the embodiment of the measuring device 54 shown in Figure 6, the mounting arm 72 is aligned along an inclined axis 76, which is inclined relative to the horizontal axis 74 and forms an angle 78 therewith. The camera 58 and the radiation sensor unit 60 are elements of a measuring device 80 referred to as a Pyranocam, which is configured to determine a global irradiance and its components: direct radiation and diffuse radiation, as well as the radiation reflected from the ground, in a horizontal and / or inclined plane. The Pyranocam measuring device 80 is configured to determine the components of the global irradiance. The Pyranocam measuring device 80 is described in WO 2021219570 A1, the content of which is hereby fully incorporated herein.

[0137] The measuring device 30, the measuring device 31, and the measuring device 54 are arranged in the area of ​​the solar energy system 12, for example, the PV system 12. The measuring device 30 has, as radiation sensor unit 38, for example, the sky-facing pyranometer for measuring the global irradiance and the camera 32 that observes the ground. The camera 32 is, for example, a fisheye camera 32 with an approximately hemispherical field of view. An approximately hemispherical field of view is understood to mean a hemispherical field of view around the camera 32, which covers the circular section 26. The radius of the covered area is typically at least 8 meters with a camera mounted at a height of approximately 2 meters.

[0138] The fisheye camera 32 is arranged at a location where as many of the surface types encountered at the site as possible are within the field of view of the camera 32. The camera 32 is leveled approximately horizontally. The sky-facing pyranometer 38 is advantageously set up and maintained according to the specifications for measurements of the horizontal global irradiance (GHI), as described, for example, in the reference: Task 16 Solar Resource - Best Practices Handbook for the Collection and Use of Solar Resource Data - 3rd Edition, Chapter 3, 3 MEASURING SOLAR RADIATION. The camera 32 and the pyranometer 38 are advantageously arranged a short distance 26 from each other, and both measurements are synchronized.

[0139] Additionally, as summarized in the measuring device 54, a ground-facing radiation sensor unit 60, which is, for example, a pyranometer, can be installed near the camera 32 so that the fields of view 64 and 62 of the radiation sensor unit 60 and the camera 32 match as closely as possible. The radiation sensor unit 60 is leveled approximately horizontally. Additionally, a fisheye camera 58 viewing the sky can also be installed near the radiation sensor unit 60. The camera 58 observing the sky can be a cloud camera, is leveled approximately horizontally, and has a clear field of view 65.

[0140] The satellite image data are high-resolution satellite data with a resolution significantly finer than the distances between individual modules 10, which are usually arranged at a distance of 5 m or less, and are recorded for the entire area covered by the solar installation 12.

[0141] The camera 32 and the camera 58, as well as the radiation sensor unit 38 and the radiation sensor unit 60, are connected to the evaluation unit 42, which has a suitable data acquisition system. The evaluation unit 42 is connected to a server of the solar energy system 12. For example, the radiation sensor unit 38, in particular the pyranometer 38, and the radiation sensor unit 60, in particular the pyranometer 60, are connected to a data logger. The camera 32 and the data logger then independently upload measurement data to the server (e.g., via SFTP). The measurement data is stored on the server until processing. The evaluation unit 42 has the usual resources of an edge computer. The evaluation unit 42 has a data connection to retrieve the satellite image data, for example, remote sensing data, for example, via an internet connection.

[0142] If the sky-facing pyranometer 38 is used, the direct irradiance (DNI) and diffuse irradiance (DHI) are derived from the GHI. A decomposition model or another method is used for this purpose. If an additional sky-facing camera 58 is present at the same location, the method described in the publication by Blum, N.B., et al. (2022). "Measurement of diffuse and plane of array irradiance by a combination of a pyranometer and an all-sky imager." Solar Energy 232: 232-247 is used to determine DNI and DHI.

[0143] Furthermore, the radiance of the sky is radiance HimmeiEstimated according to the usual calculation of so-called white-sky and black-sky albedo, as described in the publication LUCHT, Wolfgang; SCHAAF, Crystal Barker; STRAHLER, Alan H., "An algorithm for the retrieval of albedo from space using semiempirical BRDF models", IEEE Transactions on Geoscience and Remote sensing, 2000, 38th vol., no. 2, pp. 977-998. The diffuse radiation and the radiance of the radiation originating from the sky can be determined using the sky-facing camera 58.

[0144] The measurement of diffuse irradiance (DHI) via the sky-facing camera 58 is more accurate than competing systems commonly used in practice.

[0145] Most importantly, camera 58 can be used to determine the radiance of the sky radiation from the camera image. This provides comparatively accurate information about the sky's radiance. This can be a significant advantage over a rather rough estimate of the radiance based on a GHI measurement or measurements of DNI and DHI. It is expected that this will increase the overall accuracy and robustness of the method.

[0146] Figure 7 shows a diagram of a method for determining the ground albedo and / or the irradiance of the radiation reflected from the ground 14, which method is used in the solar installation 12, wherein the measuring device 30, the measuring device 31 and the measuring device 54 carry out the determination of the ground albedo and / or the irradiance of the radiation reflected from the ground 14 on the module front side and / or the rear module side 11 facing away from the sun of at least one solar module 10 of the solar installation 12.

[0147] In step S120, one or more images of the ground 14 are captured using the at least one camera 32. Features and / or structures and / or different surface types are extracted from the image(s) using a first machine learning model, and a first result data set is generated. A ground albedo is determined for each of the features and / or structures and / or surface types in the camera image(s).

[0148] In step S130, satellite image data is provided from satellite data. Features and / or structures and / or different surface types are extracted from the provided satellite image data, in particular image data from a satellite, using a second machine learning model, and a second result data set is generated.

[0149] In step S140, the images of the ground 14 are analyzed in the evaluation unit 42 with the aid of the satellite image data from step S130, and at least one feature and / or structure and / or surface type is assigned, and the ground albedo corresponding to the feature and / or structure and / or surface type in the camera image is correlated with features and / or structures and / or surface types determined from the satellite images at each point in the satellite image.

[0150] Figure 8 shows a diagram of an expanded method for determining the ground albedo and / or the irradiance of the radiation reflected by the ground 14, which method is used in the solar installation 12. The measuring device 30, the measuring device 31, and the measuring device 54 determine the irradiance of the radiation reflected by the ground 14 on the rear module side 11 of the at least one solar module 10 of the solar installation 12, facing the ground. In addition to the steps described in Figure 7, in step S110, the global irradiance is recorded using the at least one radiation sensor unit 38, and the data is analyzed.In particular, measurement data of the global irradiance are determined using the at least one radiation sensor of the radiation sensor unit 38, wherein the global irradiance comprises diffuse radiation, and / or direct radiation, and / or radiation reflected by the ground 14. The ground albedo for at least one of the surface types and / or features and / or structures in the camera image is determined through a combined evaluation of the measured global irradiance and image information from the camera. The global irradiance data and the images from the camera 32 are referred to as ground-based measurement data 49.

[0151] Figure 9 shows the process with the listed process steps in a little more detail with more intermediate steps:

[0152] S110: Determining measurement data of the global irradiance with the at least one radiation sensor of the radiation sensor unit 38, in particular in a field of view looking towards the sky in an approximately horizontal plane, wherein the global irradiance comprises diffuse radiation and / or direct radiation and / or radiation reflected on the ground 14.

[0153] S120: Capturing an image of the ground 14 with the at least one camera 32 in a field of view 62 of the camera 32 directed toward the ground 14; S150: Extracting features, in particular structures and / or surface types, from the image of the ground 14 using a first machine learning model and generating a result data set, wherein the first machine learning model is trained with first training data sets with reference data;

[0154] S160: Determine a ground albedo for each of the features;

[0155] In this case, the image information from camera 32, in particular the intensities of the color channels in each pixel, is advantageously evaluated to determine the radiation reflected from the ground. By comparing the radiation reflected from the ground and the global radiation measured by the radiation sensor unit 38, at least an albedo is then determined.

[0156] S130: Providing satellite image data, in particular image data of a satellite 22, wherein the individual method steps 131 to 137 are not explicitly listed in Figure 9 - see Figure 10;

[0157] S170: Creating a joint data set from the satellite image data and the result data set of the analyzed images of camera 32;

[0158] S180: Assigning the ground albedo from step S160 to each feature determined from the satellite image data from the common data set, in particular by means of a second machine learning model and using the features determined in S150. In a further optional method step S190, structures and / or surface types are extracted from the measurement data of the radiation sensor unit 38 by means of a third machine learning model, and a further result data set is generated, wherein the third machine learning model is trained with third training data sets containing reference data.

[0159] In a further optional method step S200, the result data set of the radiation sensor 38, the result data set of the camera 32 and the satellite image data are combined into a common data set.

[0160] In step S140, the radiation reflected on the ground 14 is determined for each location and for each PV module in the power plant from the local ground albedo and the global irradiance measured by the radiation sensor unit 38.

[0161] The satellite image data provided in method step S130 are either image data of the satellite analyzed using a machine learning model or raw data.

[0162] In the first machine learning model and / or the second machine learning model and / or the machine learning model, a convolutional neural network (CNN) algorithm is used for the satellite image data. The ground-based measurement data 49 are determined by the radiation sensor unit 38, the optional radiation sensor unit 60 directed toward the ground 14, and the camera 32, which is directed substantially toward the ground 14. A spatial field of view of approximately 180° is covered, and the camera 32 captures images of the ground 14.

[0163] In method step S150, areas in the images of the camera 32 and / or the satellite image data are classified into surface types, wherein the surface types are differentiated between areas on the ground 14 and elements of the solar installation 12, wherein at least two of the following categories are distinguished:

[0164] (i) unshaded ground 14,

[0165] (ii) shaded ground 14,

[0166] (iii) soil surface, which in turn can be differentiated into different vegetation types and / or sand and / or snow and / or arable soil;

[0167] (iv) Structures, where the structures may be PV modules, brackets, buildings, fences, and / or sensor mounts.

[0168] In a further method step S134 not explicitly shown in Figure 9 (see Figure 10), image pixels in the images of the satellite image data are assigned to marking points in the solar installation 12, in particular the method of georeferencing is used here.

[0169] In method step S165, a reflection value corresponding to a ground albedo determined from the ground-based measurement data 49 is assigned to each surface type, in particular a surface type determined from the satellite image data. In method step S160, a ground albedo with areal resolution and its distribution in the solar installation 12 and / or an average value for the solar installation 12 are determined from the images of the camera 32, in particular by means of the evaluation unit 42.

[0170] In method steps S150 and S160, a demixing method is applied in the evaluation unit 38 in order to determine from the measured and / or averaged ground albedo an assignment to the surface type determined with the camera 32 and / or from the satellite image data.

[0171] Based on color channels of camera 32, a spectral correction of the image data is performed, whereby a broadband and / or spectrally weighted value of the ground albedo is determined.

[0172] In method step S120, the camera 32 continuously takes a picture, at least once per day.

[0173] To convert measured values ​​from camera 32 and / or camera 58, at least one of the following quantities is used:

[0174] - a ratio of broadband radiation to the portion of radiation registered by the camera 32 and / or the camera 58, and / or

[0175] - an intensity of RGB channels of camera 32 and / or camera 58, and / or

[0176] - an internal and / or external calibration of the camera 32 and / or the camera 58, and / or

[0177] - an inclination and orientation of the sensor of the camera 32 and / or the camera 58, and / or

[0178] - an inclination and orientation of an inclined plane, and / or

[0179] - the position of the sun during radiation measurement and / or - a camera sensitivity, which is determined from an illuminance of the camera 32 and / or the camera 58 and / or

[0180] - the spectral sensitivity of the RGB channels and / or - recording settings and / or

[0181] - the RGB camera image and / or

[0182] - the internal and / or external calibration of the camera 32 and / or the camera 58.

[0183] Figure 10 provides an overview diagram summarizing all the applied process steps, sub-steps, alternatives, and configurations. The arrows represent the links between the individual, independently determined measurement data, the analyzed data, the result files, etc.

[0184] The left column concerns the process step S120: capturing an image of the ground 14 using the camera 32. The following process steps are carried out:

[0185] S121 : Transmission of the measurement data to the evaluation unit 42;

[0186] S122: Segmentation, classification of the measurement data with a subdivision of the image areas into clearly defined classes;

[0187] S123: Filtering out shadowed areas of the ground 14 and / or obstacles from the images;

[0188] S124: Determination of the radiance received by camera 32 for each pixel using a radiometric camera model.

[0189] The next column concerns process step S110: Acquisition of a measured value of the global irradiance, abbreviated as GHI (global horizontal irradiance). Process steps S111 and S112 are then performed: S111: Transmission of the measured data from the radiation sensor unit 38 to the evaluation unit 42;

[0190] S112: Estimation of direct irradiance DNI, diffuse irradiance DHI and sky radiance.

[0191] The following column concerns a method step S210: capturing at least one image of the sky. The images of the sky are captured with the second camera 58. The captured images of the sky are transmitted to the evaluation unit 42 in a method step S211. In a computer program executed in the evaluation unit 42, the captured images of the sky are used in method step S112 to estimate the DNI and / or the DHI and / or the radiance of the sky. The PyranoCam method described above is used here.

[0192] Subsequently, a further method step S220 is performed: the detection of the irradiance of the radiation reflected from the ground RHI. This can be done using the radiation sensor unit 60. The detected value is transmitted to the evaluation unit 42 in a method step 221. The transmitted data is used in a method step 222 to correct the radiance detected by the camera 58.

[0193] The right column describes step S130 in detail: provision of satellite image data from remote sensing data and evaluation of the satellite image data in steps S131, S132, S133, S134, and S135. The satellite image data from the satellite images is evaluated to assign a surface type and, in turn, a measured albedo to all power plant areas.

[0194] In step S131, the satellite image data is transmitted to the evaluation unit 42. In step S132, the satellite image data is segmented and classified, whereby image areas are subdivided into clearly defined classes.

[0195] Method step S132 is typically carried out using a machine learning model, for example a convolutional neural network (CNN).

[0196] Method step S132 is typically carried out using a Mode II chain of various machine learning models, for example by applying a machine learning model for the detection of object classes, a machine learning model for the pixel-by-pixel segmentation of image areas that have an object class, and a convolutional neural network for further subdivision, in particular a semantic segmentation, of image areas that were recognized as unshaded ground.

[0197] Subsequently, in method step S133, image areas which cannot be used for albedo determination are filtered out, namely shaded areas of the ground 14 and obstacles, in particular PV modules.

[0198] In process step S134, the satellite images are georeferenced, whereby each pixel in the satellite image is assigned a reference point in the solar energy system 12. In the next process step 135, the area classification from S132 is interpolated over image areas that are shaded and / or built-up areas. As an alternative to interpolation, an inpainting technique is used. In process step S132, unshaded ground is divided into classes pixel by pixel. These classes indicate different albedo values. In particular, the following classes, i.e. surface types, are distinguished: dry, fresh grass, dry, wet soil, dry, wet sand, various types of arable soil, bushes, steppe grass, and agricultural crops. This is typically done using semantic segmentation. For this purpose, a method described in K. He, X. Zhang, S. Ren, and J.Sun, "Deep Residual Learning for Image Recognition," in CVPR, 2016, uses the method ResNet34. ResNet34 is a convolutional neural network (CNN) that is supervised and trained using a large number of images with known class assignments. Alternative convolutional neural networks exist and can also be used. This machine learning model is trained using large datasets of publicly available and self-acquired and labeled images of ground surfaces and scenery, which are possible in PV power plants. Transfer learning can be used to pre-train the machine learning model in an upstream step. Unsupervised training can be used for this. Alternatively, publicly available model weights can be used.

[0199] In process step S132, a machine learning model for detecting object classes is used to identify image regions containing one of the object classes: solar modules, substructures, concrete foundations, paved shaded and / or unshaded ground, as well as shaded ground and other obstacles. Unshaded, unpaved ground is treated as the background, i.e., image regions without any objects present. These image regions are further processed in semantic segmentation. For each image region mentioned above that contains an object class, a rectangular image section is created, with each edge of the image section (top, bottom, left, right) either tangent to the image of the object or a few pixels away from it.

[0200] For this process step, for example, a model composed as follows can be used: a feature-rich backbone network is used. The network is designed to extract hierarchical features from the image and generate deep representations of the image information. A suitable neck structure combines features of different scales. A suitable head structure generates the final predictions of the object classes based on the processed features of the neck structure. This machine learning model is trained using large datasets of publicly available and self-acquired and labeled satellite images and other images of PV modules. Transfer learning can be used to pre-train the machine learning model in an upstream step. Unsupervised training can be used for this.Alternatively, publicly available model weights can be used.

[0201] The segmentation in process step S132 for the precise pixel-by-pixel detection of object classes in image sections in which an object class was previously detected is carried out using the segment-anything method described in Kirillov, Alexander et al., "Segment anything," arXiv preprint arXiv:2394.02643, 2023 (arxiv.org / pdf / 2304.02643.pdf). This method distinguishes the image of an object class detected in the previous step pixel-by-pixel from the background and other object classes. In the next step, the image sections are reassembled into an overall image.Thus, an image is obtained in which image areas representing individual object classes, i.e. solar modules, substructures, concrete foundations, paved shaded and / or unshaded ground as well as shaded ground and other obstacles, are differentiated approximately pixel by pixel and each labeled, and in which unshaded ground background corresponds to unlabeled image areas.

[0202] In process step S132, unshaded ground is subsequently divided pixel by pixel into further classes. These classes indicate different albedo values. In particular, the following classes, i.e. surface types, are distinguished: dry, fresh grass, dry, wet soil, dry, wet sand, various types of arable soil, shrubs, steppe grass, and agricultural crops. This is typically done using semantic segmentation. For this purpose, a method described in K. He, X. Zhang, S. Ren, and J. Sun, "Deep Residual Learning for Image Recognition," in CVPR, 2016, is used: ResNet34. ResNet34 is a convolutional neural network (CNN) that is supervised and trained using a large number of images with known class assignments. Alternative convolutional neural networks exist and can also be used.This machine learning model is trained using large datasets of publicly available and self-acquired and labeled satellite images and / or other images of PV power plants and / or images of terrain or scenery featuring various surface types encountered in PV power plants. Transfer learning can be used to pre-train the machine learning model in an upstream step. Unsupervised training can be used for this. Alternatively, publicly available data can be used.

[0203] Model weights are used. The convolutional neural network is used to further subdivide image areas identified as unshaded ground into different surface types (light, dark sand, dry, wet grass, various rock types, and herbs). For this purpose, the method described in Fabel, Yann et al., "Applying self-spervised learning for sematic cloud segmentation of all-sky images," Atmospheric Measurement Techniques, 2022, Vol. 15, No. 3, pp. 797-809, is used. This architecture is trained according to the procedure referred to in the above-mentioned literature as the "Fabel method," using ground images 14 for training instead of sky images. Alternative models exist for this purpose and can also be used.

[0204] Using the georeferencing used in process step S134, the pixels of the satellite image are assigned to points on the ground 14 with a deviation of a few centimeters. If the georeferencing provided by the data provider is insufficient, cross-correlation is used to compare the position of solar modules 10 in the satellite image with their known position. The image is iteratively translated, rotated, and stretched to maximize the cross-correlation. Using georeferencing, detection, segmentation, classification, and semantic segmentation, a surface type is assigned to each point on the ground 14.

[0205] In order to assign a surface type to ground areas of the power plant that are obscured by structures, a nearest neighbor interpolation (meaning: closest point) is applied in process step S135. Nearest neighbor interpolation is a simple method of multivariate interpolation in one or more dimensions. In process step S230, the bidirectional reflectance distribution function (BRDF) of each surface type is estimated according to the formula

[0206] BRDF (00,0, (p)- 1 / lT { fiso + fvol * Kvol (00,0, (p) + fgeo * Kgeo(0O,0,(p)}

[0207] Here, fiso and fgeo are weighting factors for isotropic reflection (iso) and surface scattering (geo), respectively. Kvoi and K geo are cores, also called kernels. The cores, or kernels, can be found, for example, in the literature: Roujean, Jean-Louis, Leroy, Marc, Deschamps, Pierre-Yves in "A bidirectional reflectance model of the Earth's surface for the correction of remote sensing data," in Journal of Geophysical Research, Atmospheres (1992), 97, Vol. D18, pp. 20455-20468.

[0208] Kvoi (0o0q>) is the LiS Pace core and K ge o(0o0(p) is the Rossthick kernel from the above reference.

[0209] Furthermore, in step S230, a ground albedo of each surface type of the ground 14 observed in the field of view of the camera 32 is estimated. The results from method steps S124, S112, and S222 are incorporated here. In particular, the known assignment of image pixels to surface types from method step 132 is used to arrive at a ground albedo determined for each pixel for each surface type. In particular, for each surface type, the mean value of the ground albedo is calculated across all image pixels assigned to this surface type. These results from S230 are used in method step S240 to assign a ground albedo to surface types that were not observed in the field of view of the camera. No current measurements are available for these surface types. Therefore, previous measurements and values ​​stored in a database are used.Such database values ​​are created, in particular, in a preparatory step from literature values ​​and measurements at various locations. If a surface albedo has been successfully determined for a surface type in process step S230, it is not changed in process step S240.

[0210] The results of S240 and S135 are used in step S136 to assign a ground albedo to each point on the ground 14. For this purpose, each point on the ground is assigned a ground albedo determined for this surface type in step S240 based on the surface type determined for this point in step S135.

[0211] In S137, the ground albedo is calculated over time and space. These averages are used to calculate a ground albedo that, depending on the user's needs, is representative of individual PV modules, all PV modules in a module string, all PV modules in an inverter, or all PV modules in a PV power plant over a specific period of time, e.g., one hour, one day, or one month. When calculating the average, user-specified temporal and spatial weighting factors are taken into account if required. These weighting factors can give greater weight to periods or areas that contribute more significantly to electricity generation in the evaluated area or period. In particular, the existing global radiation in relation to the average global radiation over the area or period can be used as a weighting factor.The ground albedo is determined for different surface types by first determining the surface types of the ground 14 at the location of the camera 32 facing the ground 14. The described AI-based semantic segmentation of the surface types in the image of the camera 32 facing the ground 14 is performed, with each pixel of the image being assigned a label or category. The image is divided into regions or categories.

[0212] This takes into account shaded areas, non-representative objects, and soil types such as green grass, light sand, concrete foundations, etc. Segmentation, and thus the division of the image into regions, is performed using a convolutional neural network.

[0213] The machine learning models are trained accordingly, using images of the ground 14 instead of images of the sky. Detected objects (e.g., PV tables, sensor mounts) and shaded areas are excluded from further analysis. The albedo is determined only for unshaded ground 14. This allows the measuring device 30, 54 to be set up in areas where a significant portion of the ground is covered by buildings, e.g., between the rows of modules 10 of a solar energy system 12.

[0214] The albedo can also be determined from the radiance measured in method steps S110 and / or S210 and / or S120, as described below: The camera 32 receives the radiance in a pixel at a zenith angle 0 . Only in this case, the zenith angle is defined relative to a perpendicular pointing from the camera 32 to the ground. 0 Ois the zenith angle of a point in the sky, <p ist der relative Azimuthwinkel zwischen einem Punkt am Himmel und einem Pixel im Bild der Kamera. Die Winkel 0, O0, p sind über die gegebene Ausrichtung der Kamera 32 bekannt. Somit lässt sich die von der Kamera 32 empfangene Radianzrefiektiert schreiben als Integral von BRDF(0 o , 6 , cp) * Radiance Himme i(0o> <p) über die ganze durch 0 Q , cp described hemisphere above the horizontal £l sky

[0215] The reflected radiance is measured by camera 32. The radiance sky is estimated according to method step S112, either using input data from camera 58 or based on a measurement of the global radiation and / or the direct radiation and / or the diffuse radiation. When using camera 58, a geometric calibration of the camera is used. When using only a measurement of the global radiation and / or the direct radiation and / or the diffuse radiation, the sun's position must be known. For this purpose, the time is measured, the geographical coordinates of the camera location are determined, and a suitable astronomical algorithm is applied to calculate the sun's position. In BRDF(öo, ö, (p)) the three weighting factors f remain as unknowns. By evaluating at least three measurements of reflected radiance, the BRDF can thus be completely determined.The BRDF is determined more reliably when as many radiance values ​​as possible are included. Therefore, at least measurements from the previous 24 hours, taken 10 minutes apart, and all pixels of a surface type are evaluated. The three weighting factors f are then selected, minimizing the mean square error between the measured and calculated radiance reflected or radiance reflected corrected. For known DNI and DHI, or when using a camera 32, the radiance reflected and thus the BRDF of the surface type are known. With the BRDF, the albedo of the surface type is always known and can be calculated directly by weighted integration and dividing by the global irradiance present in the horizontal plane (GHI), as follows: 1 describes the lower hemisphere.

[0216] Overall, an albedo measurement is obtained in real time for all surface types present in the image of camera 32.

[0217] To further increase the measurement accuracy, the following is carried out in an optimal implementation of the procedure:

[0218] The reflected radiance captured by camera 32 is weighted integrated to obtain the ground reflected radiance (RHI): where U is the solid angle of the hemisphere below the horizontal. k is a calibration factor, k is chosen at each time point so that the RH of the camera matches the measured RHI of the ground-facing pyranometer 60. This results in an adjusted radiance reflected value, corrected = k * radiance reflected, and the albedo calculation for a soil type is adjusted to: In summary, the method S100 can recognize the same surface types from the images of camera 32 and the satellite image data through the process steps. Additionally, the ground albedo of the respective surface type of the ground 14 is determined from the images of camera 32 and applied to all areas of the solar installation 12 of the same surface type. It is advantageous that the satellite image data are repeated at regular intervals, thus allowing seasonally dependent changes in the ground 14 to be recognized. This allows a relatively accurate determination of the irradiance of the radiation reflected by the ground 14, even in the case of spatial inhomogeneity of the ground 14, and a precise yield forecast can be created.

[0219] Reference symbol

[0220] 5 Front of a module

[0221] 10 module, PV module

[0222] 11 Rear, module back

[0223] 12 solar technology systems

[0224] 14 Floor

[0225] 16 Soil quality, soil type 1

[0226] 18 Soil quality, soil type 2

[0227] 20 Soil condition, soil type 3

[0228] 22 satellites

[0229] 24 pixel resolution of the satellite image data

[0230] 26 distance

[0231] 28 circular area

[0232] 30 measuring device

[0233] 31 Measuring device

[0234] 32 Camera

[0235] 34 Camera field of view 32

[0236] 35 Mounting arm

[0237] 36 bracket

[0238] 38 Radiation sensor unit

[0239] 39 Mounting arm

[0240] 40 second bracket

[0241] 42 Evaluation device, evaluation unit

[0242] 44 Data connection, data transfer device

[0243] 46 Data connection, data transfer device

[0244] 48 Data connection, data transfer device

[0245] 49 ground-based measurement data

[0246] 50 Data storage facility

[0247] 52 Data connection, data transfer device

[0248] 54 Measuring device 56 Bracket

[0249] 58 Camera

[0250] 60 Radiation sensor unit

[0251] 62 Camera field of view 63 Field of view directed upwards

[0252] 64 Field of view of the radiation sensor unit

[0253] 65 Field of view directed upwards

[0254] 66 Data transfer device

[0255] 68 Distance 70 Mounting post of the bracket 56

[0256] 72 Mounting arm of the bracket 56

[0257] 74 horizontal axis

[0258] 76 inclined axis

[0259] 78 angles

[0260] S110 to S190 process steps

[0261] S210 to S240 process steps

Claims

Claims 1. Method for determining a ground albedo and / or an irradiance of radiation reflected on the ground (14) in a solar installation (12), in particular on a rear module side (11) of at least one solar module (10) of the solar installation (12), with at least one device (30, 31, 54) for recording ground-based measurement data, comprising at least one camera (32) which records one or more images of the ground (14), and an evaluation unit (42) which evaluates the ground-based measurement data using satellite image data.

2. Method according to claim 1, wherein in the device (30, 31, 54) additionally at least one radiation sensor unit (38) is provided, which detects a global irradiance.

3. Method according to claim 1 or 2 comprising the following steps: (i) capturing one or more images of the ground (14) with the camera (32) in a field of view of the camera (32) directed towards the ground (14); (ii) extracting features and / or structures and / or different surface types from the image or the plurality of images of the ground (14), in particular by means of a first machine learning model, and generating a first result data set; (iii) determining a ground albedo for each of the features and / or structure and / or surface type in the camera image or images; (iv) Extracting features and / or structures and / or surface types from the provided satellite image data, in particular image data from satellites, in particular by means of a second machine learning model, and generating a second result dataset; (v) generating a joint data set from the second result data set of the satellite image data and the first result data set of the analyzed images of the camera (32); (vi) Assigning at least one feature and / or structure and / or surface type from the image or multiple images of the camera to at least one feature and / or structure and / or surface type determined from the satellite image data; (vii) determining a ground albedo for at least one point in the power plant by correlating the ground albedo corresponding to the feature and / or structure and / or surface type in the image or multiple images from the camera with features and / or structures and / or surface types determined from the satellite image data at each point in the satellite image.

4. The method according to claim 3, comprising the additional steps: (i) determining measurement data of the global irradiance with the at least one radiation sensor of the radiation sensor unit (38), wherein the global irradiance comprises diffuse radiation and / or radiation reflected on the ground (14) and / or direct radiation, (ii) Determining the ground albedo for at least one feature and / or structure and / or one of the surface types in the camera image by a combined evaluation of measured global irradiance and image information from the camera.

5. Method according to claim 3 or 4, comprising the additional step of: determining the radiation reflected on the ground (14) for at least one point of the solar installation from the ground albedo and the global irradiance measured by means of the radiation sensor unit (38).

6. A method according to any one of claims 4 or 5, comprising the additional step: Assigning a ground albedo to surface types in the satellite image data to which no surface type could be assigned in the camera image, in particular by relying on empirical values.

7. A method according to any one of claims 3 to 6, comprising the additional step: Assigning surface types to areas in the satellite image data to which no surface type could be assigned, based on surface types of surrounding image areas, in particular by means of interpolation or inpainting methods.

8. The method according to claim 2 to 7, wherein features and / or structures and / or surface types are extracted from the measurement data of the radiation sensor unit (38) by means of a third machine learning model and a further result data set is generated, wherein the third machine learning model is trained with third training data sets, in particular with reference data.

9. Method according to one of the preceding claims, wherein the ground-based measurement data (49) are determined by a global radiation sensor (60) directed towards the ground (14) and the camera (32) which is directed substantially towards the ground (14).

10. Method according to one of the preceding claims, wherein a spatial field of view of approximately 180° can be covered by means of the camera (32) and the camera (32) captures images of the ground (14).

11. Method according to one of the preceding claims, wherein areas in the images of the camera (32) and / or the satellite image data are classified into surface types, wherein surface types are differentiated between areas on the ground (14) and elements of the solar installation (12), wherein at least two of the following categories are distinguished: (i) unshaded ground (14), (ii) shaded ground, (iii) soil surface, (iv) Superstructures.

12. Method according to one of the preceding claims, wherein image pixels in the images of the satellite image data are assigned to marking points or points in the solar installation (12), in particular wherein the method of georeferencing is used.

13. Method according to one of the preceding claims, wherein a surface type, in particular a surface type determined from the satellite image data, is assigned a ground albedo which corresponds to a ground albedo determined from the ground-based measurement data (49).

14. Method according to one of the preceding claims, wherein a spatially resolved ground albedo and its distribution in the solar installation (12) and / or an average value for the solar installation (12) are determined from the satellite image data and the images of the camera (32), in particular by means of the evaluation unit (42).

15. Method according to one of the preceding claims, wherein intensity values for each color channel in the camera image are evaluated in order to derive therefrom the ground albedo of at least one surface type, in particular wherein a radiometric camera model is used to derive radiation received by the camera and reflected by the ground from the intensity values.

16. Method according to one of the preceding claims, wherein intensity values for each color channel in the image of the camera are evaluated in combination with measured values of the radiation sensor unit (38) in order to derive therefrom the ground albedo of at least one surface type.

17. Method according to one of the preceding claims 2 to 16, wherein a demixing method is used in the evaluation unit (42) in order to assign a ground albedo to at least one surface type determined from the image of the camera (32) and / or from the satellite image data from a ground albedo which is calculated from one or more measured values of the global radiation sensor (60) directed towards the ground (14) and from one or more measured values of the radiation sensor unit (38).

18. Method according to one of the preceding claims, wherein a spectral correction of the image data is carried out based on color channels of the camera (32, 58), whereby a broadband and / or spectrally weighted value of the ground albedo is determined.

19. Method according to one of the preceding claims, wherein spatially high-resolution satellite images with a resolution of significantly finer than 10 m x 10 m are used as satellite image data.

20. Method according to one of the preceding claims, wherein the camera (32, 58) continuously takes an image, at least once per day.

21. Method according to one of the preceding claims, wherein, for converting measured values from the camera (32, 58), at least one of the variables selected from the group consisting of a ratio of broadband radiation to the portion of radiation registered by the camera (32, 58), an intensity of RGB channels of the camera (32, 58), an internal and / or external calibration of the camera (32, 58), an inclination and orientation of the sensor of the camera (32, 58), an inclination and orientation of an inclined plane, and / or the position of the sun during the radiation measurement, and / or a camera sensitivity determined from an illuminance of the camera (32, 58), the spectral sensitivity of the RGB channels, and / or recording settings, the RGB camera image, and / or the internal and / or external calibration of the camera (32, 58) is used.

22. Device (30, 31, 54) for carrying out a method according to one of the preceding claims, comprising at least one camera (32, 58) and an evaluation unit (42) which is configured to carry out an evaluation of ground-based measurement data and / or the camera (32, 58) using satellite image data.

23. The device according to claim 22, wherein the camera (32, 58) is configured to have at least the following properties: a single image is captured at a fixed time interval, in particular every half and full minute; the at least one sensor of the camera (32, 58) has a constant color temperature; the camera (32, 58) has a constant exposure time for each single image; for exposure control of the camera (32, 58), a predetermined minimum value of an average image brightness is set, wherein a signal amplification applied by the camera remains unchanged at a higher image brightness, in particular wherein the predetermined minimum value of an average image brightness is preferably at most 10%, particularly preferably at most 8%, very particularly preferably at least 5%.

24. Device according to one of claims 22 to 23, wherein the camera (32, 58) is designed to record the ground (14) in the field of view, wherein the camera (32, 58) in particular comprises a fisheye camera.

25. Device according to claim 22 to 24, wherein a radiation sensor unit (38, 60) with at least one sensor is provided, wherein the evaluation of the ground-based measurement data of the radiation sensor unit (38, 60) is carried out using satellite images.

26. Device according to claim 22 to 25, wherein the at least one sensor of the radiation sensor unit (38, 60) and at least one sensor of the camera (32, 58) are each arranged in a horizontal plane such that the field of view of the at least one sensor lies below the horizontal plane and terminates with the horizontal plane, so that a field of view (62, 64) directed towards the ground (14) can be detected.

27. Device according to one of claims 22 to 26, wherein the at least one radiation sensor unit (38, 60) and the at least one camera (32, 58) are coupled so that measurement data are recorded by the radiation sensor unit (38, 60) and the camera (32, 58) in a time-synchronized manner.

28. Device according to one of claims 22 to 27, wherein the radiation sensor unit (38, 60) is designed such that a measurement data recording of the radiation sensor unit takes place with a high temporal resolution, in particular with a temporal resolution of less than 10 see, preferably less than 5 see, particularly preferably less than or equal to 1 see.

29. Device according to one of claims 22 to 28, wherein the radiation sensor unit (38, 60) is designed to detect solar radiation in a wavelength range of 0.3 pm to 3 pm.

30. Device according to one of claims 22 to 29, wherein the camera (32, 58) is designed to capture the entire field of view in one recording, in particular wherein the camera (32, 58) is designed as a surveillance camera and / or as a fisheye camera.

31. A computer program for determining an irradiance of radiation reflected at the ground, comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of a method for determining the irradiance of solar radiation reflected at the ground (14) according to one of claims 1 to 21.

32. Data processing device for a device (30, 31, 54) for determining a ground albedo and / or irradiance of the solar radiation reflected on the ground (14), at least comprising a measurement data acquisition unit, an evaluation unit (42) and a computer.

33. A trained, in particular first, machine learning model for extracting features and / or structures and / or surface types (16, 18, 20) from at least one image taken by a camera (32) and comprising information on the soil condition, which is used in a method according to one of claims 1 to 21.

34. A trained, in particular second, machine learning model for determining a ground albedo and / or an irradiance of solar radiation reflected on the ground (14), which has been trained in a supervised manner, in particular using input data and reference data from satellite image data.

35. A trained, in particular third, machine learning model for determining features and / or structures from global irradiance measurement data.

36. A trained, in particular fourth, machine learning model for determining an irradiance of solar radiation reflected at the ground is proposed, which was trained in a supervised manner, in particular using input data and reference data from Satellite image data can be used.

Images