An ultra-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data

Abstract

The application provides a super-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data, relates to the technical field of solar radiation prediction, and comprises the following steps: collecting a sky image data set, wherein the sky image data comprises ground-based cloud image, solar radiation data and meteorological information data; based on the sun position in the ground-based cloud image, a spherical sky image is constructed, the spherical sky image is projected to a two-dimensional plane to obtain a plane sky image, and through the sun position information and the clear sky irradiance function, the first theoretical total solar radiation is obtained; the meteorological information data and the total solar radiation are pretreated to obtain standard meteorological information data and standard total solar radiation respectively; a sky image fusion prediction model is constructed, the standard meteorological information data, the standard total solar radiation and the first theoretical total solar radiation are input into the sky image fusion prediction model, and predicted total solar radiation and predicted photovoltaic power are obtained. The application is helpful to reduce the limitations possibly caused by a single data source and improve the comprehensiveness and accuracy of solar radiation prediction and photovoltaic power prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solar radiation prediction, and particularly relates to a super-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data. BACKGROUND

[0002] With the increasing demand for renewable energy worldwide, solar energy as a clean and sustainable energy has attracted widespread attention in its development and utilization. Solar power generation, especially photovoltaic power generation, has become an important part of the new energy field due to its environmental friendliness and high energy conversion efficiency. However, the efficiency and stability of solar power generation are directly affected by the conditions of solar radiation, and the spatio-temporal variation of solar radiation has high uncertainty and complexity. However, due to the influence of weather conditions, seasonal changes, geographical location and other factors on solar radiation, the output power of photovoltaic power generation has obvious intermittency and instability, which poses challenges to the stable operation of the power grid and the dispatching management of the power system.

[0003] A super-short-term solar radiation prediction method and system based on ground-based cloud images are disclosed in Chinese Patent No. CN113537561A, which includes the following steps: obtaining a ground-based cloud image time series dataset and a solar radiation time series dataset before the prediction time; inputting the ground-based cloud image time series dataset into a pre-trained convolutional neural network model to extract a cloud feature vector time series dataset; inputting the solar radiation time series dataset and the cloud feature vector time series dataset into a pre-established recurrent neural network model to obtain solar radiation data at the prediction time; and using the solar radiation data at the prediction time as the super-short-term solar radiation prediction result. However, the prediction model based on historical solar radiation and numerical weather prediction data provided by the above-mentioned application is restricted by data accuracy and algorithm principle, and it is difficult to meet the prediction requirements of the super-short-term time scale, resulting in insufficient accuracy of solar radiation prediction. Therefore, it is necessary to provide a wind power equipment fault source tracing method and system based on fault decoupling multi-level model to improve the accuracy of fault source positioning. SUMMARY

[0004] Therefore, the present application provides a super-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data. By combining sky images, solar radiation data and meteorological information data, the method fully utilizes multiple data sources, captures more factors affecting solar radiation, reduces the limitations that may be brought by a single data source, and improves the comprehensiveness and accuracy of solar radiation prediction and photovoltaic power prediction.

[0005] The present application provides a super-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data, which comprises:

[0006] Collecting a sky image dataset, wherein the sky image data includes ground-based cloud image, solar radiation data and meteorological information data;

[0007] Based on the solar position in the ground-based cloud image, a spherical sky image is constructed, a planar sky image is obtained by projecting the spherical sky image to a two-dimensional plane, and a first theoretical total solar radiation is obtained by the solar position information and the clear sky irradiance function;

[0008] The meteorological information data and the total solar radiation are preprocessed to obtain standard meteorological information data and standard total solar radiation, respectively;

[0009] A sky image fusion prediction model is constructed, the standard meteorological information data, the standard total solar radiation and the first theoretical total solar radiation are input into the sky image fusion prediction model, and a predicted total solar radiation and a predicted photovoltaic power are obtained.

[0010] On the basis of the above technical scheme, preferably, the extracting the spherical sky image projection to a two-dimensional plane to obtain a planar sky image specifically includes:

[0011] The simulated solar position in the ground-based cloud image is calculated by the barycentric method, and the spherical sky image is projected onto a unit sphere according to the simulated solar position, the actual solar position and a re-projection function to obtain a unit sphere sky image;

[0012] Based on the cloud color moment function, the texture matrix function and the unit sphere sky image, the color features and the cloud cluster texture features of the unit sphere sky image are calculated, respectively;

[0013] The unit sphere sky image is converted into a planar sky image by unfolding the unit sphere sky image through the latitude and longitude method.

[0014] On the basis of the above technical scheme, preferably, the color features include a first-order matrix of the average value of color distribution in the unit sphere sky image, a second-order matrix of the dispersion degree of color distribution in the unit sphere sky image and a third-order matrix of the skewness degree of color distribution in the unit sphere sky image, and the cloud cluster texture features include cloud cluster texture energy, cloud cluster texture entropy value, cloud cluster texture contrast, cloud cluster texture correlation and cloud cluster texture inverse difference.

[0015] More preferably, the expression of the re-projection function is:

[0016] θ d =k0θ+k1θ 3 +k2θ 5 +k3θ 7 +k4θ 9

[0017] wherein, θ represents a non-distorted camera angle, θ d represents a distorted camera angle, k0 represents a first projection coefficient, k1 represents a second projection coefficient, k2 represents a third projection coefficient, k3 represents a fourth projection coefficient, and k4 represents a fifth projection coefficient.

[0018] More preferably, the calculating the simulated sun position in the ground-based cloud image by the barycenter method specifically comprises:

[0019] performing screening on the ground-based cloud image to obtain a standard cloud image without cloud layer obstruction, and converting the standard cloud image into a gray-scale cloud image;

[0020] obtaining a cloud pixel value threshold according to the maximum inter-class variance method, the brightness of the gray-scale cloud image, and the contrast of the gray-scale cloud image;

[0021] if a pixel value corresponding to a pixel point in the gray-scale cloud image is greater than the cloud pixel value threshold, determining the pixel point as a sun pixel point, and obtaining a gray value and a pixel coordinate of the sun pixel point;

[0022] calculating a sun barycenter coordinate based on the barycenter method and the pixel coordinate of the sun pixel point to obtain the simulated sun position.

[0023] More preferably, the projecting the spherical sky image onto a unit sphere to obtain a unit sphere sky image specifically comprises:

[0024] generating uniformly distributed pixel points on the unit sphere according to the Fibonacci grid method;

[0025] correspondingly matching each pixel point of the unit sphere with the spherical sky image, and obtaining a pixel value corresponding to the pixel point by a bilinear interpolation function to obtain the unit sphere sky image.

[0026] More preferably, the preprocessing the meteorological information data and the total solar radiation specifically comprises:

[0027] performing an abnormality identification operation on the meteorological information data and the total solar radiation based on a DBSCAN clustering algorithm to obtain data abnormal points in the meteorological information data and the total solar radiation;

[0028] performing a cubic spline fitting on data points around the data abnormal points according to a preset abnormal point interval, so that a spline interpolation replaces an abnormal value corresponding to the data abnormal point.

[0029] More preferably, the expression of the clear sky irradiance function is:

[0030]

[0031] τ = 0.56(e -0.56AM + e -0.09AM )k

[0032] AM = [(r + c e ) 2 cos 2 (θ) + (2r + 1 + c e )(1 - c e )] 1 / 2 - (r + c e )cosθ

[0033]

[0034] wherein E0 represents solar radiation reaching above the atmosphere when the solar radiation has not been attenuated by the atmosphere, E sc represents a solar constant, N represents a serial number of a current day in a year, k represents an adjustment parameter, AM represents a ratio of a path length of solar radiation passing through the atmosphere to a shortest path length, r represents an earth radius, c e represents a cosine value of a solar zenith angle, τ represents a ratio of solar transmitted radiation to incident radiation, E clear represents a first theoretical total solar radiation.

[0035] Further preferably, the sky image fusion prediction model is constructed, the standard meteorological information data, the standard total solar radiation and the first theoretical total solar radiation are input into the sky image fusion prediction model to obtain predicted total solar radiation and predicted photovoltaic power, and specifically comprising:

[0036] The standard meteorological information data, the standard total solar radiation and the first theoretical total solar radiation are divided in a preset proportion to obtain a cloud image training set, a cloud image verification set and a cloud image test set respectively;

[0037] The sky image fusion prediction model is constructed, the cloud image training set is input into the sky image fusion prediction model for training to obtain an optimized sky image fusion prediction model, the cloud image test set is input into the optimized sky image fusion prediction model to obtain a second theoretical total solar radiation;

[0038] A difference between the second theoretical total solar radiation and an actual solar radiation degree in the solar radiation data is taken as a regression target, the second theoretical total solar radiation is compensated to obtain predicted total solar radiation and predicted photovoltaic power.

[0039] Further preferably, the meteorological information data includes temperature, relative humidity, wind speed and wind direction, zenith angle and azimuth angle, cloud amount, air pressure, albedo and broadband turbidity.

[0040] The application provides a kind of based on ground cloud and multi-source data's ultra-short-term photovoltaic power prediction method relative to prior art has the following beneficial effects:

[0041] (1) By combining sky image, solar radiation data and meteorological information data, make full use of multiple data sources, more factors that affect solar radiation can be captured, the limitations that single data source may bring are reduced, the comprehensiveness and accuracy of solar radiation prediction and photovoltaic power prediction are improved, the ground cloud is converted into spherical sky image, and then projected into plane sky image, this process can better preserve the spatial information of cloud distribution, more accurately represent the influence of cloud on solar radiation at different solar elevation angles, and the accuracy of solar total radiation prediction and photovoltaic power prediction is improved by feature screening and outlier processing of multi-source data.

[0042] (2) The barycentric method is used to calculate the position of the sun, which can more accurately locate the sun in the image and adapt to different lighting conditions and cloud distribution, improving the accuracy of subsequent data processing. Projecting the spherical sky image onto the unit sphere effectively standardizes the image space and eliminates geometric distortion caused by shooting angle and equipment difference. By unit sphere projection, the full spatial information of the sky is preserved. By combining cloud color moment function and texture matrix function, color features and cloud texture features are extracted, which can more comprehensively describe the properties of cloud. The extraction of color features helps to distinguish different types of clouds and the clarity of the sky. The calculation of cloud texture features can capture the structure and distribution pattern of the cloud, which helps to predict short-term cloud changes and solar radiation fluctuations.

[0043] (3) The Fibonacci grid method is used to generate uniformly distributed pixel points on the surface of the unit sphere, ensuring that the sampling of the entire sky is consistent and comprehensive, avoiding the problem of excessive sampling near the poles by traditional latitude and longitude grid. By uniformly distributing sampling points on the unit sphere, the spatial information of the original spherical sky image is better preserved. Compared with directly projecting the spherical image to a plane, this method can significantly reduce geometric distortion, especially in the edge region of the sphere, thereby maintaining the overall integrity and accuracy of the image. The Fibonacci grid method allows flexible adjustment of the number of sampling points, so the resolution of the image can be adjusted according to demand, and different computing resources and accuracy requirements can be adapted. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0045] Figure 1 A flowchart of an ultra-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data is provided. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0047] The present application discloses an ultra-short-term photovoltaic power prediction method based on ground-based cloud images and multi-source data, referring to Figure 1 The steps of the method include S1-S4.

[0048] Step S1, collecting a set of sky image data, wherein the sky image data includes ground-based cloud image, solar radiation data and meteorological information data.

[0049] In this step, ground-based cloud images are taken using a ground-based cloud instrument, which records cloud and solar information. Solar radiation energy reaching the ground is measured using a solar radiation meter, including direct radiation (direct sunlight from the sun) and scattered radiation (scattered light from the sky).

[0050] Further, the all-sky imager collects ground-based cloud images and first filters the images to select cloud images during the day. When filtering, images outside the sunrise and sunset times are deleted according to the local sunrise and sunset times. In actual implementation, since the total solar radiation is small near sunrise and sunset, the time period for collection can be reduced, such as 7:00-18:00, which can also meet the algorithm requirements. The collection frequency of cloud image is at least 10 minutes / time. The total solar radiation is collected using a solar radiation meter, with the measurement unit being W / m2. The meteorological information is observed using a meteorological sensor, including air temperature, humidity, wind speed, wind direction, air pressure, etc.

[0051] In one example, an OTT Parsivel2 all-sky imager can be used, which is installed on an open roof platform at an altitude of 50 meters, collects cloud images with a resolution of 1600x1200 pixels every 10 minutes, the collection time is from 7:00 to 18:00 every day, and the image format is JPEG. At the same time, a Kipp&Zonen CMP21 high-precision thermopile radiation meter is used to record total solar radiation data every minute, with an accuracy of ±2% and a measurement range of 0-4000W / m 2The meteorological data collection uses a Vaisala WXT536 multi-parameter weather sensor, recording data every 5 minutes, including air temperature (accuracy ±0.3°C), relative humidity (accuracy ±3%), wind speed (accuracy ±3% or ±0.3 m / s), wind direction (accuracy ±3°), and air pressure (accuracy ±0.5 hPa). All equipment is installed in the same location, ensuring continuous collection throughout the year, 365 days. After data collection, a preliminary screening is performed using an automated script, deleting images before 7:00 and after 18:00, and flagging outliers (such as sudden changes in radiation values or weather data outside the normal range). Data is stored on a Dell PowerEdge R740 local server, equipped with a 20 TB RAID5 storage array, with a timestamp accurate to the second attached to each data point. Equipment checks and cleaning are performed weekly, and the radiometer and weather sensor are calibrated monthly. In addition, the precise GPS coordinates and altitude of the collection site are recorded, and special weather events are recorded by the on-site operator.

[0052] In this embodiment, the meteorological information data includes temperature, relative humidity, wind speed and direction, zenith and azimuth, cloud cover, air pressure, albedo, and broadband turbidity. The solar historical global radiation data is calculated from the collected direct and diffuse horizontal radiation, using the formula:

[0053] GHI = DHI + (cosθ r × DNI)

[0054] where θ r represents the solar zenith angle, GHI represents the solar historical global radiation, DHI represents the diffuse horizontal radiation, and DNI represents the direct radiation.

[0055] Step S2, based on the sun position in the ground-based cloud image, a spherical sky image is constructed, a planar sky image is obtained by projecting the spherical sky image to a two-dimensional plane, and a first theoretical solar total radiation is obtained through the sun position information and the clear sky irradiance function.

[0056] In this step, steps S21-S23 are also included.

[0057] Step S21, the simulated sun position in the ground-based cloud image is calculated by the barycentric method, and the spherical sky image is projected onto the unit sphere to obtain the unit sphere sky image according to the simulated sun position, the actual sun position, and the re-projection function.

[0058] The expression of the re-projection function is:

[0059] θ d = k0θ + k1θ 3 + k2θ 5 + k3θ 7 + k4θ 9

[0060] wherein, θ denotes the non-distorted camera angle, θ d denotes the distorted camera angle, k0 denotes the first projection coefficient, k1 denotes the second projection coefficient, k2 denotes the third projection coefficient, k3 denotes the fourth projection coefficient, and k4 denotes the fifth projection coefficient. By obtaining the sun position in each ground-based cloud image and its corresponding actual sun position, the first projection coefficient to the fifth projection coefficient can be obtained by using the least square method. Thus, the two-dimensional image can be re-projected onto a three-dimensional hemispherical surface to simulate the sky.

[0061] In the embodiment, the ground-based cloud image is screened to obtain a standard cloud image without cloud layer obstruction, and the standard cloud image is converted into a gray-scale cloud image; the cloud image pixel value threshold is obtained according to the maximum inter-class variance method, the brightness of the gray-scale cloud image, and the contrast of the gray-scale cloud image; if the pixel value corresponding to a pixel point in the gray-scale cloud image is greater than the cloud image pixel value threshold, the pixel point is determined as a sun pixel point, and the gray value and the pixel coordinate of the sun pixel point are obtained; the sun barycentric coordinates are calculated based on the barycentric method and the pixel coordinate of the sun pixel point, so as to obtain the simulated sun position, and the actual sun position is calculated through the shooting date.

[0062] The picture with complete sun is screened, and the pixel point position where the sun is located is selected. The occurrence frequency of each pixel value is counted to obtain an image histogram. All possible threshold values are traversed from the minimum pixel value to the maximum pixel value. For each threshold value, the image is divided into foreground and background two parts, and the variance between the two parts is calculated. The threshold value that makes the inter-class variance maximum is selected as the best threshold value, and the sun is segmented.

[0063] The Otsu method is applied to analyze the gray-scale cloud image, an optimal threshold value is automatically determined by maximizing the variance between the foreground (cloud) and the background (sky) two categories, the Otsu method is particularly suitable for bimodal histogram, and the cloud and the sky can be effectively separated. The average value of all pixel values of the gray-scale cloud image is calculated to obtain the overall brightness. This value reflects the overall lighting condition of the image, which is helpful to adjust the threshold value to adapt to different lighting environments. The contrast is estimated by calculating the standard deviation of the pixel values of the gray-scale cloud image. The contrast reflects the difference degree between the cloud and the sky, and high contrast usually means that the boundary between the cloud and the sky is clearer. The three factors are weighted and averaged to obtain the final cloud image pixel value threshold. Specifically, the threshold value obtained by the Otsu method is given a higher weight (such as 60%), because it can usually provide a good initial threshold value. The image brightness and the image contrast are each given a lower weight (such as 20%), for fine-tuning the threshold value. The final threshold value = 0.6*Otsu threshold + 0.2*image brightness + 0.2*image contrast.

[0064] Further, the unit sphere is generated with uniformly distributed pixel points according to the Fibonacci grid method; each pixel point of the unit sphere is matched with the spherical sky image, and a pixel value corresponding to the pixel point is obtained through a bilinear interpolation function to obtain a unit sphere sky image.

[0065] In view of the problem that pixel projection will lead to concentrated pixels on both sides of the hemisphere and sparse pixels on the top of the sphere, an inverse projection method based on spherical uniform sampling is used. The points are generated on the spherical surface by the Fibonacci grid method, which are uniformly distributed, dense and chaotic. The coordinates of the nth point are as follows:

[0066] z n =(2n-1) / L-1

[0067]

[0068] wherein the constant x n represents the x-axis coordinate value of the nth point, y n represents the y-axis coordinate value of the nth point, and z n represents the z-axis coordinate value of the nth point.

[0069] For each point on the spherical surface, a corresponding point on the two-dimensional plane can be obtained, and then a corresponding pixel value is obtained through bilinear interpolation.

[0070] Further, the unit sphere is generated with uniformly distributed pixel points according to the Fibonacci grid method, which specifically includes: determining the total number L of points to be generated on the unit spherical surface, and using the golden angle as the rotation increment to ensure that the distribution of points does not appear obvious regularity aggregation.

[0071] The Fibonacci grid method is used to generate uniformly distributed pixel points on the surface of the unit sphere, which ensures that the sampling of the entire sky is consistent and comprehensive, avoiding the problem of excessive sampling of the traditional latitude and longitude grid near the poles. By uniformly distributing the sampling points on the unit sphere, the spatial information of the original spherical sky image is better preserved. Compared with directly projecting the spherical image to the plane, this method can significantly reduce geometric distortion, especially in the edge region of the sphere, thereby maintaining the overall integrity and accuracy of the image. The Fibonacci grid method allows flexible adjustment of the number of sampling points, so that the resolution of the image can be adjusted according to the requirements, and different computing resources and accuracy requirements can be adapted.

[0072] In step S22, the color features and cloud texture features of the unit sphere sky image are calculated based on the cloud image color moment function, the texture matrix function and the unit sphere sky image.

[0073] In the step, the color features include the first-order matrix of the average value of color distribution in the unit sphere sky image, the second-order matrix of the dispersion degree of color distribution in the unit sphere sky image, and the third-order matrix of the skew degree of color distribution in the unit sphere sky image, and the cloud cluster texture features include cloud cluster texture energy, cloud cluster texture entropy, cloud cluster texture contrast, cloud cluster texture correlation, and cloud cluster texture inverse difference.

[0074] The expression of the cloud image color moment function is:

[0075]

[0076] wherein p i,j represents the i-th color component of the j-th pixel in the unit sphere sky image, m represents the number of pixels in the unit sphere sky image, μ i represents the first-order matrix of the average value of color distribution in the unit sphere sky image, σ i represents the second-order matrix of the dispersion degree of color distribution in the unit sphere sky image, and s i represents the third-order matrix of the skew degree of color distribution in the unit sphere sky image.

[0077] The gray level co-occurrence matrix of the spherical image is calculated, and the texture features of the sky and the cloud layer are extracted. The points after uniform sampling on the sphere also have equidistant points, so the gray level co-occurrence matrix can be obtained as in a two-dimensional image. The cloud cluster texture information can be obtained from the gray level co-occurrence matrix, including cloud cluster texture energy, cloud cluster texture entropy, cloud cluster texture contrast, cloud cluster texture correlation, and cloud cluster texture inverse difference, and the calculation formula is as follows:

[0078]

[0079] Entropy = -∑ i ∑ j C ij logC ij

[0080] Contrast = ∑ i ∑ j (i-j) 2 C ij

[0081]

[0082] wherein Engergy represents cloud cluster texture energy, Entropy represents cloud cluster texture entropy, Contrast represents cloud cluster texture contrast, Correlation represents cloud cluster texture correlation, and IDM represents cloud cluster texture inverse difference, C ij whether represents the normalized gray level co-occurrence matrix.

[0083] Step S23, the unit sphere sky image is unfolded by latitude and longitude method to convert the unit sphere sky image into a planar sky image.

[0084] In this step, the width and height of the planar sky image are determined, usually the width represents longitude (0 to 360 degrees) and the height represents latitude (-90 to 90 degrees). For each pixel (x, y) of the planar image, the corresponding latitude and longitude can be calculated as: longitude = (x / image width)*360°-180°, latitude = 90°-(y / image height)*180°. The latitude and longitude are converted into 3D coordinates (x, y, z) on the unit sphere, x = cos(latitude)*cos(longitude); y = cos(latitude)*sin(longitude); z = sin(latitude).

[0085] In this embodiment, the expression of the clear sky irradiance function is:

[0086]

[0087] AM = [(r + c e ) 2 cos 2 (θ) + (2r + 1 + c e )(1 - c e )] 1 / 2 - (r + c e )cosθ

[0088]

[0089] wherein E0 represents the solar radiation reaching above the atmosphere when the solar radiation has not been attenuated by the atmosphere, E sc represents the solar radiation constant, N represents the serial number of the current day in a year, k represents an adjustment parameter, AM represents the ratio of the path length of the solar radiation through the atmosphere to the shortest path length, r represents the radius of the earth, c e represents the cosine value of the solar zenith angle, τ represents the ratio of the transmitted solar radiation to the incident radiation, E clear represents the first theoretical total solar radiation, E sc = 1367 W / m 2 .

[0090] The use of the center of gravity method to calculate the position of the sun can more accurately locate the sun in the image, and can adapt to different lighting conditions and cloud distribution, improve the accuracy of subsequent data processing, effectively standardize the image space by projecting the spherical sky image onto the unit sphere, eliminate the geometric distortion caused by the difference in shooting angle and equipment, and retain the full spatial information of the sky through the unit sphere projection. Combined with the cloud color moment function and the texture matrix function, the color features and cloud texture features are extracted, which can more comprehensively describe the properties of the cloud layer. The extraction of color features helps to distinguish different types of clouds and the clarity of the sky, and the calculation of cloud texture features can capture the structure and distribution pattern of the cloud layer, which helps to predict the short-term change of the cloud layer and the fluctuation of the solar radiation.

[0091] In step S3, the meteorological information data and the total solar radiation are preprocessed to obtain standard meteorological information data and standard total solar radiation, respectively.

[0092] In this step, steps S31-S32 are also included.

[0093] In step S31, based on the DBSCAN clustering algorithm, the meteorological information data and the total solar radiation are subjected to an anomaly identification operation to obtain data anomaly points in the meteorological information data and the total solar radiation.

[0094] In this step, according to the collected meteorological information data and total solar radiation, it is ensured that the above data contains a time stamp, so as to perform time series analysis, and the data is preliminarily cleaned to handle missing values, which can be filled with mean or median, and the abnormal values are preliminarily screened to remove obviously unreasonable data points. Create new time-related features such as hours of the day, months, seasons, etc. These features help to capture the periodic and seasonal changes of meteorological data. Consider adding lag features such as the previous hour or day of data to capture short-term trends.

[0095] StandardScaler is used to standardize all numerical features to ensure that features of different scales are treated fairly in the DBSCAN algorithm. After standardization, all features will have a zero mean and a unit variance. K-distance plot method is used to select a suitable eps value, and K-distance plot curve is drawn to observe the inflection point. For minPts, generally select about 2 times the number of features, but also consider the size of the data set and the expected cluster size.

[0096] The DBSCAN algorithm is run with selected eps and minPts parameters, which will classify the data points into core points, border points, and noise points. Noise points (points labeled as -1) will be considered as potential abnormal points. Detailed analysis is performed on the abnormal points identified by DBSCAN, checking the distribution of these points in each feature dimension, with special attention to points with abnormal total solar radiation, and analyzing their relationship with other meteorological parameters. The distribution of abnormal points in the time dimension is analyzed to check if there are certain time periods (such as specific months or specific times of the day) where abnormal points are more concentrated. Based on the results of DBSCAN, a reasonable abnormal threshold is set for each feature, which can use the statistical characteristics of normal data points (non-abnormal points), such as mean plus or minus 3 times the standard deviation, as the initial threshold.

[0097] Further, after identifying abnormal points in the data set through DBSCAN or other anomaly detection methods, the indices or timestamps of these abnormal points are saved in a list, and a complete data set containing timestamps and corresponding values (such as temperature, solar radiation, etc.) is prepared. A suitable time window is determined for each abnormal point. For example, for data recorded every hour, 12 hours of data before and after the abnormal point can be selected, for a total of 24 hours of window.

[0098] In one example, the distance between data points is first calculated, and the neighborhood radius of the DBSCAN clustering algorithm is set between 0.5 and 2 times the average distance between data points. The minimum number of points is tried according to the density of the data, between 3 and 10.

[0099] In step S32, the data points around the data abnormal point are fitted by a cubic spline according to the preset abnormal point interval, so that the spline interpolation replaces the abnormal value corresponding to the data abnormal point.

[0100] In this step, for each identified abnormal point, the following steps are performed: determine the exact timestamp of the abnormal point; based on the defined window size, calculate the start and end times of the data to be collected; extract all data points within this time range from the data set; remove the points marked as abnormal from the extracted data, leaving only normal data points; ensure that the number of normal data points collected is sufficient for reliable spline fitting. Use Spearman correlation analysis method to analyze the correlation of the data, and select features with a Spearman correlation coefficient greater than a threshold value with the target variable as important features, for example, a threshold value of 0.75 can be selected.

[0101] Step S4, construct a sky image fusion prediction model, input the standard meteorological information data, the standard total solar radiation and the first theoretical total solar radiation into the sky image fusion prediction model, and obtain the predicted total solar radiation and the predicted photovoltaic power.

[0102] In one example, the collected historical ground-based cloud images and corresponding raw data sets of total solar radiation, meteorological information, etc. and calculated future theoretical total solar radiation data are preprocessed and standardized for training and prediction of a sky image fusion prediction model, which adopts a combination structure of CNN-Transform, CNN extracts image features, and Transform is used to capture time series information; the loss function of the network is MSE, and since there is autocorrelation in the sequence, the difference between the current time and the theoretical radiation degree as the regression target can eliminate the autocorrelation.

[0103] The sky image fusion prediction model is constructed, and the processed ground-based cloud images, meteorological information and theoretical total solar radiation data are input into the model. The CNN part processes image data, the Transformer part processes numerical data, and the two are fused together through the attention mechanism. The model predicts the total solar radiation in the future according to the rules learned from the training data. The CNN neural network model is trained based on the historical ground-based cloud image data set, the Transformer neural network model is trained based on the meteorological information and cloud image feature data set and the cloud image feature data obtained by CNN, and finally the total solar radiation data at the prediction time is obtained.

[0104] The loss function of the sky image fusion prediction model includes the following calculation formula:

[0105]

[0106] Where y represents the true value of the neural network input, and is the difference between the theoretical total solar radiation calculated on the day and the actual total solar radiation. The predicted value of the neural network output is represented by, which is the difference between the predicted day's theoretical total solar radiation and the predicted total solar radiation.

[0107] In this step, steps S41-S42 are also included.

[0108] In step S41, the standard meteorological information data, the standard total solar radiation and the first theoretical total solar radiation are divided by a preset ratio to obtain a cloud image training set, a cloud image validation set and a cloud image test set, respectively.

[0109] In this step, the proportions of the cloud image training set, the cloud image validation set and the cloud image test set are 8:1:1.

[0110] In step S42, the sky image fusion prediction model is constructed, the cloud image training set is input into the sky image fusion prediction model for training to obtain an optimized sky image fusion prediction model, and the cloud image test set is input into the optimized sky image fusion prediction model to obtain the second theoretical total solar radiation.

[0111] In this step, the width and height of the planar sky image are scaled to W*H to meet the input requirements of the CNN neural network, for example, it can be scaled to 224*224. The loss function of the sky image fusion prediction model is set as follows, and the parameters of the model are adjusted according to the experimental effect. The size of the convolution kernel in the CNN network in the sky image fusion prediction model is usually 3x3 or 5x5; the number of attention heads in the Transformer part is usually 4-8, which can be adjusted according to the complexity of the model and the computing resources. The batch size is usually set to 32 according to the memory, and the number of training rounds is between 50 and 100.

[0112] Step S43, taking the difference between the second theoretical total solar radiation and the actual solar radiation in the solar radiation data as the regression target, compensating the second theoretical total solar radiation to obtain the predicted total solar radiation and the predicted photovoltaic power.

[0113] By combining sky images, solar radiation data and meteorological information data, multiple data sources are fully utilized, more factors affecting solar radiation can be captured, the limitations that may be brought by a single data source are reduced, the comprehensiveness and accuracy of solar radiation prediction and photovoltaic power prediction are improved, the ground cloud chart is converted into a spherical sky image, and then projected into a planar sky image, which can better preserve the spatial information of cloud distribution and more accurately represent the influence of clouds on solar radiation at different solar elevation angles. At the same time, the feature selection and outlier processing of multiple data sources are performed, and the accuracy of the total solar radiation prediction is improved.

[0114] It should be understood that in this application, "at least one item" means one or more, and "multiple" means two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B and A and B existing at the same time, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of A, B or C, can represent A, B, C, "A and B", "A and C", "B and C", or "A and B and C", where A, B and C can be single or multiple.

[0115] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting ultra-short-term photovoltaic power based on ground-based cloud maps and multi-source data, characterized in that, The method includes: Collect a dataset of sky images, which includes ground-based cloud images, solar radiation data, and meteorological information data; Based on the sun's position in the ground-based cloud image, a spherical sky image is constructed. The spherical sky image is extracted and projected onto a two-dimensional plane to obtain a planar sky image. The first theoretical total solar radiation is obtained by using the sun's position information and the clear sky irradiance function. The expression for the clear-sky irradiance function is: τ=0.56(e -0.56AM +e -0.09AM )k AM=[(r+c e ) 2 cos 2 (θ)+(2r+1+c e )(1-c e )] 1 / 2 -(r+c e )cosθ Where E0 represents the solar radiation that reaches the upper atmosphere before atmospheric attenuation, E sc Represents the solar radiation constant, N represents the current day's position in the year, k represents the adjustment parameter, AM represents the ratio of the length of the solar radiation path through the atmosphere to the shortest path length, r represents the Earth's radius, and c represents the solar radiation constant. e E represents the cosine of the solar zenith angle, τ represents the ratio of transmitted solar radiation to incident solar radiation, and E represents the solar radiation. clear This represents the total solar radiation according to the first theory. The meteorological information data and the total solar radiation are preprocessed to obtain standard meteorological information data and standard total solar radiation, respectively. A sky image fusion prediction model is constructed by inputting the standard meteorological information data, the standard total solar radiation, and the first theoretical total solar radiation into the sky image fusion prediction model to obtain the predicted total solar radiation, predicted photovoltaic power, and predicted photovoltaic power.

2. The method as described in claim 1, characterized in that, The step of extracting the spherical sky image and projecting it onto a two-dimensional plane to obtain a planar sky image specifically includes: The simulated sun position in the ground-based cloud image is calculated using the centroid method, and the spherical sky image is projected onto a unit sphere based on the simulated sun position, the actual sun position, and the reprojection function to obtain the unit sphere sky image. Based on the cloud image color moment function, texture matrix function, and the unit sphere sky image, calculate the color features and cloud texture features of the unit sphere sky image, respectively. The unit sphere sky image is expanded using the latitude and longitude method, and then transformed into a planar sky image.

3. The method as described in claim 2, characterized in that, The color features include a first-order matrix of the average color distribution in the unit sphere sky image, a second-order matrix of the dispersion of the color distribution in the unit sphere sky image, and a third-order matrix of the skewness of the color distribution in the unit sphere sky image. The cloud texture features include cloud texture energy, cloud texture entropy, cloud texture contrast, cloud texture correlation, and cloud texture inverse difference.

4. The method as described in claim 2, characterized in that, The expression for the reprojection function is: i d =k0θ+k1θ 3 +k2θ 5 +k3θ 7 +k4θ 9 Where θ represents the distortion-free camera angle, θ d The distorted camera angle is represented by k0, k1, k2, k3, and k4.

5. The method as described in claim 2, characterized in that, The calculation of the simulated sun position in the ground-based cloud image using the centroid method specifically includes: The ground cloud image is filtered to obtain a standard cloud image without cloud cover, and the standard cloud image is converted into a grayscale cloud image. The threshold value of the cloud map pixel value is obtained based on the maximum inter-class variance method, the brightness of the grayscale cloud map image, and the contrast of the grayscale cloud map image. If the pixel value corresponding to a pixel in the grayscale cloud image is greater than the cloud image pixel value threshold, then the pixel is identified as a solar pixel, and the grayscale value and pixel coordinates of the solar pixel are obtained. Based on the centroid method and the pixel coordinates of the solar pixel, the centroid coordinates of the sun are calculated to obtain the simulated sun position.

6. The method as described in claim 2, characterized in that, The step of projecting the spherical sky image onto a unit sphere to obtain a unit sphere sky image specifically includes: A uniformly distributed set of pixels is generated on the unit sphere according to the Fibonacci grid method; Each pixel of the unit sphere is matched with the spherical sky image, and the pixel value corresponding to the pixel is obtained by using a bilinear interpolation function to obtain the unit sphere sky image.

7. The method as described in claim 1, characterized in that, The preprocessing of the meteorological information data and the total solar radiation specifically includes: The DBSCAN clustering algorithm is used to perform anomaly identification on the meteorological information data and the total solar radiation to obtain data anomalies in the meteorological information data and the total solar radiation. Based on a preset anomaly range, cubic spline fitting is performed on the data points surrounding the data anomaly point to replace the anomaly value corresponding to the data anomaly point with spline interpolation.

8. The method as described in claim 1, characterized in that, The construction of the sky image fusion prediction model involves inputting the standard meteorological information data, the standard total solar radiation, and the first theoretical total solar radiation into the sky image fusion prediction model to obtain the predicted total solar radiation and predicted photovoltaic power, specifically including: The standard meteorological information data, the standard total solar radiation, and the first theoretical total solar radiation are divided according to a preset ratio to obtain cloud image training set, cloud image verification set, and cloud image test set respectively; A sky image fusion prediction model is constructed. The cloud image training set is input into the sky image fusion prediction model for training to obtain an optimized sky image fusion prediction model. The cloud image test set is input into the optimized sky image fusion prediction model to obtain the second theoretical total solar radiation. The difference between the second theoretical total solar radiation and the actual solar irradiance in the solar radiation data is used as the regression target to compensate for the second theoretical total solar radiation, so as to obtain the predicted total solar radiation and the predicted photovoltaic power.

9. The method as described in claim 1, characterized in that, The meteorological data includes temperature, relative humidity, wind speed and direction, zenith angle and azimuth angle, cloud cover, air pressure, albedo, and broadband turbidity.

Images