Apparatus and method for determining at least one component of hemispherical irradiance of solar radiation in an arbitrary plane

Abstract

The present invention relates to a method (S100, S200, S300, S400, S500) for determining at least one component of hemispheric irradiance (120) of solar radiation in any plane, wherein at least one component has diffuse irradiance (122) and / or direct irradiance (124), and the method comprises the steps of (i) collecting measurement data of hemispheric irradiance (120) using a radiation sensor unit (12) within a field of view (26) on a plane (46) of the radiation sensor unit (12), and (ii) taking an image (110) of the sky (50) using a camera (14) within a field of view (32) on a plane (48) of the camera (14). A method comprising the steps of (iii) capturing an image (50) and generating a resulting dataset by extracting features from an image (110) of the sky (50) using a first machine learning model (S130, S230, S330, S430, S530); (iv) merging measurement datasets to form a common dataset (S140, S240, S340, S440, S540); and (v) determining at least one component of hemispherical irradiance (120) in an arbitrary plane from the dataset using a second machine learning model (S150, S250, S350, S450, S550).

Claims

1. A method (S100, S200, S300, S400, S500) for determining at least one component of hemispherical irradiance (120) of solar radiation in an arbitrary plane, wherein the at least one component includes diffuse irradiance (122) and / or direct irradiance (124), (i) A step of determining the hemispherical irradiance (120) measurement data using the radiation sensor unit (12) within the field of view (26) above the plane (46) of the radiation sensor unit (12), (ii) The step of using the camera (14) to capture an image (110) of the sky (50) within the field of view (32) above the plane (48) of the camera (14), (iii) A first machine learning model extracts features, particularly structure, from the empty (50) image (110) and generates a result dataset (S130, S230, S330, S430, S530), (iv) A step of merging the measurement data of the radiation sensor unit (12) and the result data set of the camera (14) in order to form a common dataset (S140, S240, S340, S440, S540), (v) A second machine learning model determines from the dataset at least one component of the hemispheric irradiance (120), in particular the diffuse irradiance (122) and / or the direct irradiance (124) in any plane (S150, S250, S350, S450, S550), Methods that include...

2. The method according to claim 1, wherein, in image analysis (S235, S335, S535) using the image (110) of the camera (14), features are extracted and a physical camera model is applied.

3. The method according to claim 1 or 2, wherein the image (110) of the camera (14) is evaluated by the physical camera model, a preliminary measurement of the diffuse irradiance (122) is obtained on a horizontal or inclined surface, and this preliminary measurement of the diffuse irradiance or direct irradiance calculated therefrom is incorporated as a feature by the second machine learning model, and / or the direct irradiance is determined using the camera model and the measurement of the hemispherical irradiance, and this parameter is incorporated as a feature by the second machine learning model.

4. The method according to any one of claims 1 to 3, wherein the image (110) of the camera (14) is evaluated by the physical camera model, and by utilizing the camera model, the solar peripheral radiation and / or illuminance or radiative information items for each color channel and / or the ratio of illuminance or radiative information items for two different color channels and / or the number of saturated image pixels are determined, and these parameters are incorporated as features by the second machine learning model.

5. (i) Image pixels are assigned to the observed empty region, and / or (ii) A situation-adapted broadband correction may be applied, the broadband correction being a function of the radiation calculated across the various color channels of the camera (14), in particular the diffuse irradiance. and / or (iii) The radiance is integrated or summed in various regions of the sky, in particular to determine the solar peripheral radiation or diffuse horizontal irradiance from there, and / or (iv) The color channel intensity values ​​or parameters derived therefrom are summed in a weighted manner. and / or (vi) Image metadata is incorporated to compensate for the effects of the exposure control of the camera (14), and / or (vii) The camera (14) is calibrated by using the measured value of the hemispherical irradiance (120) of the radiation sensor unit (12). The method according to any one of claims 1 to 4.

6. (i) Based on the camera model, the estimation of the radiation density of the sky is determined from the image (110) of the camera (14), and / or (ii) The camera model takes into account the spectral sensitivity of the camera (14) for each color channel or color balance function adapted to the color space used, and / or (iii) The intrinsic and external geometric calibration of the camera (14) is used. The method according to any one of claims 1 to 5.

7. The method according to any one of claims 1 to 6, wherein the second machine learning model differs from the first machine learning model in that the first machine learning model is trained using a first training dataset, and the second machine learning model is trained using a second training dataset.

8. The image analysis (S235, S335, S535) using the image (110) of the camera (14) is characterized by being extracted from the measurement data (12) of the radiation sensor unit, according to the method of any one of claims 1 to 7.

9. The method according to any one of claims 1 to 8, wherein in the image analysis (S235, S335, S535) using the image (110) of the camera (14), features are derived from the intensity of image pixels, and a certain number of saturated pixels are incorporated for feature extraction.

10. The method according to any one of claims 1 to 9, wherein at least one sum and / or weighted sum of the intensities of the image pixels, and / or at least one ratio of the sum and / or weighted sum of the intensities of the image pixels are incorporated.

11. The method according to any one of claims 1 to 10, wherein the features extracted from the image (110) of the camera (14) and / or the measurement data of the radiation sensor unit (12) are transmitted to a third machine learning model and processed by the third machine learning model (S538) to generate a dataset of the radiation sensor unit, and the dataset of the radiation sensor unit and the result dataset of the camera (14) are merged (S540) before the second machine learning model is applied (S550).

12. The method according to any one of claims 1 to 11, wherein an image transformation (S315, S415, S515) of the empty image (50) (110) is performed before the application of the first machine learning model.

13. The method according to any one of claims 1 to 12, wherein the first machine learning model and / or the second machine learning model are trained using reference data of at least one component of the hemispheric irradiance (120), particularly the diffuse irradiance (122) on a horizontal and / or inclined plane from a solar tracker, and / or the direct irradiance (124) from a direct pyranometer tracking the sun, and / or using ordinary input data such as the image (110) of the camera (14), the hemispheric irradiance (120) data, and / or intermediate results particularly obtained from the application of a physical model.

14. The first machine learning model described above is pre-trained, (i) A step of using the published weights, (ii) A step of training using at least one unsupervised training method or a self-supervised training method, (iii) a step of augmenting the data using Gaussian blur and / or image color distortion and / or reflection of the image (110), (iv) A step of training using the image (110) of the sky (50) recorded by the camera (14), The method according to any one of claims 1 to 13, wherein at least one of the steps is used.

15. The method according to any one of claims 1 to 14, wherein a convolutional neural algorithm, in particular a convolutional neural network, is used in the first machine learning model.

16. The method according to any one of claims 1 to 15, wherein at least one algorithm from among a multilayer perceptron (MLP), a random forest algorithm, a recurrent neural network, a long- and short-term memory algorithm, a transformer model algorithm, a k-nearest neighbors (k-NN) algorithm, and / or a support vector machine algorithm is used for the second machine learning model.

17. The method according to any one of claims 1 to 16, wherein the first training dataset and / or the second training dataset are filtered so that different atmospheric conditions are represented with similar intensity.

18. The method according to any one of claims 1 to 17, wherein the first machine learning model, the second machine learning model, and / or the third machine learning model are trained in a monitored manner and input data from at least one of the components of the hemispherical irradiance (120) and / or correct reference data are used.

19. The method according to any one of claims 1 to 18, wherein the first machine learning model, the second machine learning model, and the third machine learning model are trained together in a particularly monitored manner, particularly after each individual training.

20. The method according to any one of claims 1 to 19, wherein the image features from the image (110) of the blank (50) are used as additional input data for the third machine learning model or the merge (S140, S240, S340, S440, S540).

21. Apparatus (5) for performing a method (S100, S200, S300, S400, S500) for determining at least one component of hemispherical irradiance (120) of solar radiation in any plane according to any one of claims 1 to 20, wherein the at least one component includes diffuse irradiance (122) and / or direct irradiance (124), The system comprises at least one radiation sensor unit (12), a camera (14), and an evaluation unit (18) provided for evaluating the measurement data of the radiation sensor unit (12) and the camera (14), The radiation sensor unit (12) determines the irradiance of solar radiation (120) in a 180° field of view (26) above the plane (38, 46), and the plane (38, 46) is a horizontal plane or an inclined surface. The camera (14) captures measurement data including an image (110) of the sky (50) in a 180° field of view (32) above the plane (40, 48), and the plane (40, 48) is a horizontal plane or an inclined plane. The evaluation unit (18) comprises at least one first machine learning model and one second machine learning model, The evaluation unit (18) comprises at least, (i) Step (S130) of extracting features, particularly structure, from the blank (50) image (110) using the first machine learning model to generate a result dataset, (ii) A step (S140) of merging the measurement data of the radiation sensor unit (12) and the result data set of the camera (14) in order to form a common dataset, (iii) Step (S150) of determining from the data vector, from the data vector, at least one component of hemispheric irradiance (120), in particular diffuse irradiance (122), and / or direct irradiance (124) in an arbitrary plane, using the second machine learning model. An apparatus designed to perform steps including the following.

22. The apparatus according to claim 21, wherein in image analysis (S235, S335, S535) using the image (110) of the camera (14), features are extracted and a physical camera model is applicable.

23. The apparatus according to claim 21 or 22, wherein the measurement data of the camera (14) includes angle-resolved radiation information.

24. A method for training a first machine learning model to extract features, particularly structure, from an image (110) of the sky (50), (i) the step of determining or using a dataset of images (110) of the sky (50) recorded using a camera (14), in particular a cloud camera, (ii) The step of training the machine learning model using the dataset of the empty (50) image (110), (iii) The step of training the first machine learning model using known images (110) of the sky (50) recorded at another location. A method that includes at least one of the following.

25. (i) A step of using the published weights, (ii) A step of training using at least one unsupervised training method or a self-supervised training method, (iii) Steps to enhance the data using Gaussian blur and / or image color distortion and / or image reflection, (iv) A step of training using the image (110) of the sky (50) recorded by the camera (14) The method according to claim 24, wherein pre-training having at least one of the steps is used.

26. A computer program for determining at least one component of hemispherical irradiance (120) of solar radiation in an arbitrary plane, comprising a command for instructing the computer (19) to perform the steps of a method for determining at least one component of hemispherical irradiance of solar radiation according to any one of claims 1 to 20 (S100, S200, S300, S400, S500) when the computer (19) is executing the program.

27. A data processing device (20) for a device (5) for determining at least one component of hemispherical irradiance (120) of solar radiation in any plane, according to any one of claims 21 to 23, the data processing device comprising at least a measurement data acquisition unit (16), an evaluation unit (18), and a computer (19).

28. A first machine learning model trained for a method (S100, S200, S300, S400, S500) for determining at least one component of hemispherical irradiance (120) of solar radiation in any plane for extracting features, in particular structures, from an image (110) of the sky (50), the first machine learning model trained according to the method of claim 24 or 25.

29. A second machine learning model trained for a method (S100, S200, S300, S400, S500) for determining at least one component of hemispheric irradiance (120) of solar radiation in any plane, according to any one of claims 1 to 20, the second machine learning model trained in a monitored manner, particularly using input data and reference data, and particularly including a multilayer perceptron algorithm, for determining at least one component of hemispheric irradiance (120) of solar radiation in any plane.

Images