Distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching

Abstract

The present application relates to the technical field of photovoltaic power prediction, in particular to a distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching, which comprises the following steps: firstly, a cloud-ground mapping model is constructed based on photovoltaic sites and solar positions to determine a satellite cloud map feature area; subsequently, a three-dimensional convolutional neural network is used to predict a future satellite cloud map sequence; then, the similarity score of the predicted satellite cloud map sequence and the satellite cloud map of each historical date in the prediction period is calculated, and a temperature constraint is introduced to screen the historical date with the highest score under the constraint as a similar day, and the power data in the prediction period in the similar day is extracted as similar historical power; finally, the similar historical power is combined with the measured power of the target day's adjacent period as the neural network input, and the future 4-hour regional photovoltaic cluster power prediction result is output. The present application integrates spatial meteorological information and photovoltaic power mode, thereby effectively improving the prediction accuracy of the distributed photovoltaic cluster photovoltaic power.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power prediction technology, and specifically to a method for ultra-short-term power prediction of distributed photovoltaic clusters based on cloud-to-ground mapping matching. Background Technology

[0002] With the continuous growth of distributed photovoltaic (PV) installed capacity, the significant uncertainty in its power output poses a severe challenge to the safe and stable operation of the power system. Accurate PV power forecasting is considered an effective and economical means to mitigate this adverse effect. In practical applications, due to the small capacity of individual distributed PV sites and their limited impact on the overall power grid, power system operators are more concerned with the total power forecasting of distributed PV clusters within a region.

[0003] Current mainstream photovoltaic (PV) power forecasting methods typically rely on surface meteorological observation data or numerical weather prediction data to achieve high forecast accuracy. However, most distributed PV sites are not equipped with dedicated meteorological observation equipment, and acquiring high spatiotemporal resolution weather forecast data covering a wide area is costly. These factors significantly limit the application of traditional forecasting methods in regional distributed PV power forecasting.

[0004] In reality, the core factor influencing photovoltaic (PV) power output is the solar irradiance received by the PV modules, which is directly affected by cloud distribution. Therefore, satellite cloud imagery, which reflects the dynamic changes of cloud cover over a wide area, has become an important source of information for improving the accuracy of regional distributed PV power prediction. Satellite cloud imagery possesses advantages such as wide coverage and high temporal resolution, and has received widespread attention in PV power prediction research in recent years.

[0005] Most existing prediction methods directly use the entire satellite cloud image as model input, without fully considering the spatiotemporal correspondence between cloud distribution and ground-based photovoltaic (PV) sites. Due to the dispersed layout and small capacity of distributed PV sites, their power output is primarily influenced by local cloud clusters surrounding the sites. The entire satellite cloud image often covers a large amount of meteorological areas unrelated to the target site, introducing redundant information. This redundancy not only interferes with the model's learning of key features, reducing prediction performance, but also significantly increases the model's computational burden. Therefore, effectively extracting key cloud features from satellite cloud images that are relevant to changes in PV power in the target area and improving the prediction accuracy of distributed PV power is a critical problem that urgently needs to be solved. Summary of the Invention

[0006] The main objective of this invention is to provide a method for ultra-short-term power prediction of distributed photovoltaic clusters based on cloud-to-ground mapping matching, aiming to solve the problem of how to improve the prediction accuracy of distributed photovoltaic power.

[0007] The technical solution proposed in this invention is as follows:

[0008] A method for ultra-short-term power prediction of distributed photovoltaic clusters based on cloud-ground mapping matching is applied to an ultra-short-term power prediction system for photovoltaic clusters based on cloud-ground mapping matching; the system includes a data collection module, a calculation module, and a prediction module; the method further includes:

[0009] The data collection module acquires satellite cloud images covering the distributed photovoltaic cluster based on the geographical location of the distributed photovoltaic cluster.

[0010] The prediction module obtains the historical satellite cloud image sequence for each moment within a preset time period, and determines the satellite cloud image set for the future prediction period based on the historical satellite cloud image sequence and the three-dimensional convolutional neural network model, and obtains the feature region cloud image information for the future prediction period based on the satellite cloud image set for the future prediction period.

[0011] The computing module constructs a cloud-to-ground mapping model, and based on the geographical location of the target photovoltaic site and the position of the sun, determines the characteristic regions in the satellite cloud image covering the distributed photovoltaic cluster that are associated with the power output of the target photovoltaic site.

[0012] The calculation module calculates the similarity score of the feature regions in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of the same period on each date within the past preset time period. Under the set temperature constraint, the most similar historical satellite cloud image is selected, and the photovoltaic power data of the most similar historical satellite cloud image in the prediction period is extracted as similar historical power data.

[0013] The prediction module constructs a photovoltaic power prediction model based on artificial neural networks. It uses similar historical power data of the prediction period and measured photovoltaic power data of adjacent periods before the prediction period as input features of the model to obtain the output photovoltaic power of the target photovoltaic site in the future prediction period.

[0014] Preferably, the prediction module acquires historical satellite cloud image sequences for each time period within the past preset time frame, and determines the set of satellite cloud images for the future prediction period based on the historical satellite cloud image sequences and a three-dimensional convolutional neural network model. Based on the set of satellite cloud images for the future prediction period, it acquires feature region cloud image information for the future prediction period, including:

[0015] The prediction module acquires satellite cloud images from the satellite cloud imagery for T consecutive time steps within the preceding prediction period of the future prediction period. And form a collection of historical satellite cloud images. ;

[0016] The prediction module obtains a set of satellite cloud images for the future prediction period based on the historical satellite cloud image set C. Where i represents the satellite cloud image of the i-th sampling point in the historical 3D convolutional neural network model set C arranged in time, i=1,2,…,T, and T represents the total number of satellite cloud images input into the 3D convolutional neural network model; F represents the prediction result of the three-dimensional convolutional network model, where F represents the set of satellite cloud images for the future prediction period obtained by inputting the historical satellite cloud image set C into the three-dimensional convolutional neural network model.

[0017] Preferably, the calculation module constructs a cloud-to-ground mapping model, and based on the geographical location and solar position of the target photovoltaic site, determines the feature regions in the satellite cloud image covering the distributed photovoltaic cluster that are associated with the power output of the target photovoltaic site, including:

[0018] The calculation module calculates the target photovoltaic site's geographical coordinates and solar hour angle. and solar declination angle Calculate the solar zenith angle at the current moment. ;

[0019] The calculation module is based on the solar zenith angle. Calculate the solar azimuth angle using the geographic coordinates of the target photovoltaic site. ;

[0020] The calculation module is based on the solar zenith angle. Sun azimuth And the cloud height is used to calculate the intersection of sunlight rays directed towards the target photovoltaic site and the cloud cluster;

[0021] The calculation module calculates the range of variation of the intersection points based on the general distribution of cloud heights, and merges them into possible regions of the intersection points in space, which are used as the feature regions.

[0022] Preferably, the calculation module calculates the target photovoltaic site based on its geographical coordinates and solar hour angle. and solar declination angle Calculate the solar zenith angle at the current moment. ,include:

[0023] The calculation module obtains the current date n and calculates the solar declination angle based on the current date n. and solar hour angle ;

[0024] The calculation module obtains the geographical coordinates of the target photovoltaic site, and then calculates the coordinates based on the geographical coordinates of the target photovoltaic site and the solar declination angle. and solar hour angle Calculate the solar zenith angle The geographical coordinates of the target photovoltaic site include longitude. and latitude ;

[0025] The calculation module is based on the solar zenith angle. Calculate the solar azimuth angle using the geographic coordinates of the target photovoltaic site. ,include:

[0026] The calculation module is based on the solar declination angle. The geographical coordinates of the target photovoltaic site and the solar zenith angle. Calculate the solar azimuth angle .

[0027] Preferably, the calculation module is based on the solar zenith angle. Sun azimuth And the cloud height calculation of the intersection of sunlight rays directed towards the target photovoltaic site and the cloud cluster, including:

[0028] The calculation module sets the theoretical cloud height range. Combined with the solar zenith angle Calculate the points where sunlight intersects with cloud clusters at various altitudes along the direction from the sun to the target photovoltaic site, mark these points as cloud obstruction points, and derive the distance range between each cloud obstruction point and the target photovoltaic site. ;

[0029] The calculation module is based on the solar azimuth angle. and the median of the distance interval Calculate the east-west horizontal distance x and the north-south vertical distance y from the center of the cloud-blocking point to the target photovoltaic site;

[0030] The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site And the vertical distance y, calculate the latitude of the center of the cloud-obscured point. ;

[0031] The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site ,longitude And the horizontal distance x, calculate the longitude of the center of the cloud-obscured point. ;

[0032] The calculation module uses the center of the cloud obscuration point as the center and the distance interval as the reference. The length is the radius A circular region is constructed on the satellite cloud image, and this circular region is used as the characteristic region that actually affects the power output of the target photovoltaic site.

[0033] Preferably, the solar declination angle The calculation formula is:

[0034] ,

[0035] The solar hour angle The calculation formula is:

[0036] ,

[0037] In the formula, n is a digital date, which represents what day of the year it is; Pi; Let h represent the sine function; h is the hourly time, and the formula for calculating h is:

[0038] ,

[0039] In the formula, This indicates the number of hours at the current time. This indicates the number of minutes at the current time.

[0040] Preferably, the calculation module obtains the geographical coordinates of the target photovoltaic site, and based on the geographical coordinates of the target photovoltaic site and the solar declination angle... and solar hour angle Calculate the solar zenith angle The calculation formula is:

[0041] ,

[0042] In the formula, Indicates the latitude of the target photovoltaic site; Indicates the solar declination angle; Indicates solar hour angle; Represents the sine function. Represents the cosine function. Represents the arcsine function;

[0043] The calculation module is based on the solar declination angle. The geographical coordinates of the target photovoltaic site and the solar zenith angle. Calculate the solar azimuth angle The calculation formula is:

[0044] ,

[0045] In the formula, Represents the inverse cosine function;

[0046] The calculation module sets the theoretical cloud height range. Combined with the solar zenith angle Calculate the points where sunlight intersects with cloud clusters at various altitudes along the direction from the sun to the target photovoltaic site, mark these points as cloud obstruction points, and derive the distance range between each cloud obstruction point and the target photovoltaic site. ,include:

[0047] The calculation module calculates the distance L from the cloud obstruction point to the target photovoltaic site:

[0048] ,

[0049] In the formula, H represents the height of the cloud cluster. Indicates the zenith angle of the sun. Represents the tangent function;

[0050] The calculation module is based on the solar azimuth angle. and the median of the distance interval The formulas for calculating the horizontal distance x (east-west direction) and the vertical distance y (north-south direction) from the center of the cloud-blocking point to the target photovoltaic site are as follows:

[0051] ,

[0052] ,

[0053] The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site And the vertical distance y, calculate the latitude of the center of the cloud-obscured point. The calculation formula is:

[0054] ,

[0055] The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site ,longitude And the horizontal distance x, calculate the longitude of the center of the cloud-obscured point. The calculation formula is:

[0056] .

[0057] Preferably, the calculation module calculates the similarity score of feature regions between the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of the same time period on each date within a preset time period. Under the set temperature constraint, the most similar historical satellite cloud image is selected, and the photovoltaic power data of the most similar historical satellite cloud image in the prediction period is extracted as similar historical power data, including:

[0058] The calculation module obtains the feature region of each time step in the future prediction period through the cloud-ground mapping model;

[0059] The calculation module calculates the similarity score between the satellite cloud image of the future prediction period and the feature regions of the historical satellite cloud image sequence of the same time period for each historical date within the past preset time period;

[0060] The calculation module selects the historical dates with the highest similarity scores under the set temperature constraints. Extract the historical dates The photovoltaic power data for the forecast period is used as the similar historical power data.

[0061] Preferably, the calculation module calculates the similarity score of feature regions in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of the same time period for each historical date within the past preset time period using the following formula:

[0062] ,

[0063] in, This represents the similarity score for the k-th historical date. These represent the historical satellite cloud image sequences for the current date within the predicted time period. This represents the sequence of satellite cloud images for the k-th historical date within the predicted time period; This represents the pixels located in the feature region. MAE This represents the mean absolute error.

[0064] Preferably, the calculation module selects the historical date with the highest similarity score under the set temperature constraint conditions. Extract the historical dates The photovoltaic power data for the forecast period, as the similar historical power data, includes:

[0065] The calculation module introduces the regional average temperature as a temperature constraint condition, and applies this temperature constraint condition to the calculation of the similarity score. The calculation method is as follows:

[0066] ,

[0067] Where m represents the total number of days of the preset duration. The temperature constraint function is expressed as follows:

[0068] ,

[0069] in, The allowable temperature difference threshold; Let be the regional average temperature on the k-th historical date. The average temperature of the region for the current date;

[0070] The calculation module selects the historical date with the highest similarity score. As similar dates, photovoltaic power data for the predicted period is extracted from similar dates as similar historical power data.

[0071] The above technical solution can achieve the following beneficial effects:

[0072] The proposed method for ultra-short-term power prediction of distributed photovoltaic (PV) clusters based on cloud-ground mapping matching utilizes the close relationship between the output curve of distributed PV power plants and cloud shading to construct a cloud-ground mapping model. This model calculates the location and magnitude of cloud shading that will affect the target PV site during the future prediction period, enabling the target PV site to obtain reliable future weather data and effectively improving the prediction accuracy of distributed PV power. This, in turn, contributes to the operation and scheduling of distributed PV power plants. Specifically, this application combines satellite cloud images as weather features with the historical power curves of the power plants and inputs them into a PV power prediction model based on artificial neural networks for prediction, obtaining the predicted PV power value of the target PV site. This effectively improves the prediction accuracy of PV power in distributed PV clusters, helps to adjust the operating power of peak-shaving units in advance, reduces scheduling pressure, and promotes the application of PV power generation. Attached Figure Description

[0073] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0074] Figure 1 This is a flowchart illustrating the steps of the first embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention. Detailed Implementation

[0075] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0076] This invention proposes a method for ultra-short-term power prediction of distributed photovoltaic clusters based on cloud-ground mapping matching.

[0077] As attached Figure 1 As shown, in the first embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, the method is applied to a photovoltaic cluster power ultra-short-term prediction system based on cloud-ground mapping matching; the system includes a data collection module, a calculation module, and a prediction module; this embodiment also includes the following steps:

[0078] Step S110: The data collection module acquires satellite cloud images covering the distributed photovoltaic cluster based on the geographical location of the distributed photovoltaic cluster;

[0079] Step S120: The prediction module obtains the historical satellite cloud image sequence for each moment within a preset time period (e.g., the past month, i.e., 30 days), and determines the satellite cloud image set for the future prediction period (e.g., the next 4 hours) based on the historical satellite cloud image sequence and the three-dimensional convolutional neural network model, and obtains the feature area cloud image information for the future prediction period based on the satellite cloud image set for the future prediction period.

[0080] Step S130: The calculation module constructs a cloud-to-ground mapping model, and based on the geographical location of the target photovoltaic site and the position of the sun, determines the feature regions in the satellite cloud map covering the distributed photovoltaic cluster that are associated with the power output of the target photovoltaic site;

[0081] Step S140: The calculation module calculates the similarity score of the feature regions in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of the same time period on each date within the past preset time period. Under the set temperature constraint, the most similar historical satellite cloud image is selected, and the photovoltaic power data of the most similar historical satellite cloud image in the prediction period is extracted as similar historical power data.

[0082] Step S150: The prediction module constructs a photovoltaic power prediction model based on artificial neural networks. It uses the similar historical power data of the prediction period and the measured photovoltaic power data of the adjacent period before the prediction period (the duration of the adjacent period is consistent with the duration of the prediction period, that is, the first 4 hours of the future prediction period) as the input features of the model to obtain the output photovoltaic power of the target photovoltaic site in the future prediction period (the next 4 hours).

[0083] The proposed method for ultra-short-term power prediction of distributed photovoltaic (PV) clusters based on cloud-ground mapping matching utilizes the close relationship between the output curve of distributed PV power plants and cloud shading to construct a cloud-ground mapping model. This model calculates the location and magnitude of cloud shading that will affect the target PV site during the future prediction period, enabling the target PV site to obtain reliable future weather data and effectively improving the prediction accuracy of distributed PV power. This, in turn, contributes to the operation and scheduling of distributed PV power plants. Specifically, this application combines satellite cloud images as weather features with the historical power curves of the power plants and inputs them into a PV power prediction model based on artificial neural networks for prediction, obtaining the predicted PV power value of the target PV site. This effectively improves the prediction accuracy of PV power in distributed PV clusters, helps to adjust the operating power of peak-shaving units in advance, reduces scheduling pressure, and promotes the application of PV power generation.

[0084] In the second embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-to-ground mapping matching proposed in this invention, based on the first embodiment, step S120 includes the following steps:

[0085] Step S210: The prediction module acquires satellite cloud images from the satellite cloud image for T consecutive time steps within the preceding prediction period of the future prediction period. And form a collection of historical satellite cloud images. .

[0086] Specifically, for example, in this embodiment, the current time is 12 noon, the future prediction period is from 12 noon to 4 pm, then the previous prediction period of the future prediction period is from 8 am to 12 pm on the same day.

[0087] Step S220: The prediction module obtains a set of satellite cloud images for the future prediction period based on the historical satellite cloud image set C. Where i represents the satellite cloud image of the i-th sampling point in the historical 3D convolutional neural network model set C arranged in time, i=1,2,…,T, and T represents the total number of satellite cloud images input into the 3D convolutional neural network model; F represents the prediction result of the three-dimensional convolutional network model, where F represents the set of satellite cloud images for the future prediction period obtained by inputting the historical satellite cloud image set C into the three-dimensional convolutional neural network model.

[0088] In the third embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-to-ground mapping matching proposed in this invention, based on the first embodiment, step S130 includes the following steps:

[0089] Step S310: The calculation module calculates the target photovoltaic site's geographical coordinates and solar hour angle. and solar declination angle Calculate the solar zenith angle at the current moment. .

[0090] Specifically, the solar zenith angle here It is the angle between the sunlight shining on the target photovoltaic site and the vertical line on the ground.

[0091] Step S320: The calculation module is based on the solar zenith angle. Calculate the solar azimuth angle using the geographic coordinates of the target photovoltaic site. .

[0092] Step S330: The calculation module calculates based on the solar zenith angle. Sun azimuth And the cloud height is used to calculate the intersection of sunlight rays directed towards the target photovoltaic site and the cloud cluster.

[0093] Step S340: The calculation module considers the uncertainty of cloud height in actual applications, calculates the range of variation of the intersection point according to the general distribution of cloud height, and merges them into the possible regions of the intersection point in space as the feature region.

[0094] Specifically, this embodiment provides a specific scheme for determining the characteristic regions associated with the power output of the target photovoltaic site in a satellite cloud image covering a distributed photovoltaic cluster.

[0095] In the fourth embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, based on the third embodiment, step S310 includes the following steps:

[0096] Step S410: The calculation module obtains the current date n and calculates the solar declination angle based on the current date n. and solar hour angle .

[0097] Step S420: The calculation module obtains the geographical coordinates of the target photovoltaic site, and based on the geographical coordinates of the target photovoltaic site and the solar declination angle... and solar hour angle Calculate the solar zenith angle The geographical coordinates of the target photovoltaic site include longitude. and latitude .

[0098] Step S320 includes the following steps:

[0099] Step S430: The calculation module calculates based on the solar declination angle. The geographical coordinates of the target photovoltaic site and the solar zenith angle. Calculate the solar azimuth angle .

[0100] Specifically, this embodiment provides a detailed scheme for calculating the solar altitude angle and solar azimuth angle.

[0101] In the fifth embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, based on the fourth embodiment, step S330 includes the following steps:

[0102] Step S510: The calculation module sets the theoretical cloud height range. Combined with the solar zenith angle Calculate the points where sunlight intersects with cloud clusters at various altitudes along the direction from the sun to the target photovoltaic site, mark these points as cloud obstruction points, and derive the distance range between each cloud obstruction point and the target photovoltaic site. .

[0103] Step S520: The calculation module calculates based on the solar azimuth angle. and the median of the distance interval Calculate the east-west horizontal distance x and the north-south vertical distance y from the center of the cloud-shading point to the target photovoltaic site.

[0104] Step S530: The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site And the vertical distance y, calculate the latitude of the center of the cloud-obscured point. .

[0105] Step S540: The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site ,longitude And the horizontal distance x, calculate the longitude of the center of the cloud-obscured point. .

[0106] Step S550: The calculation module uses the center of the cloud obscuring point as the center and the distance interval as the calculation range. The length is the radius (Specifically:) A circular region is constructed on the satellite cloud image, and the constructed circular region is used as the characteristic region that actually affects the power output of the target photovoltaic site.

[0107] Specifically, considering the continuity of the cloud cluster at the boundary of the feature region, the diameter of the constructed circular region is expanded by an additional 20% on the original basis to more accurately cover the area that has a potential occlusion effect on the site.

[0108] In the sixth embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-to-ground mapping matching proposed in this invention, based on the third embodiment, the solar declination angle... The calculation formula is:

[0109] ,

[0110] The solar hour angle The calculation formula is:

[0111] ,

[0112] In the formula, n is a digital date, which represents what day of the year it is; Pi; Let h represent the sine function; h is the hourly time, and the formula for calculating h is:

[0113] ,

[0114] In the formula, This indicates the number of hours at the current time. This indicates the number of minutes at the current time.

[0115] In the seventh embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, based on the fourth embodiment, the calculation module obtains the geographical coordinates of the target photovoltaic site, and based on the geographical coordinates of the target photovoltaic site and the solar declination angle... and solar hour angle Calculate the solar zenith angle The calculation formula is:

[0116] ,

[0117] In the formula, Indicates the latitude of the target photovoltaic site; Indicates the solar declination angle; Indicates solar hour angle; Represents the sine function. Represents the cosine function. This represents the arcsine function.

[0118] The calculation module is based on the solar declination angle. The geographical coordinates of the target photovoltaic site and the solar zenith angle. Calculate the solar azimuth angle The calculation formula is:

[0119] ,

[0120] In the formula, This represents the inverse cosine function.

[0121] Step S510 includes the following steps:

[0122] Step S710: The calculation module calculates the distance L from the cloud obstruction point to the target photovoltaic site.

[0123] ,

[0124] In the formula, H represents the height of the cloud cluster. Indicates the zenith angle of the sun. Represents the tangent function;

[0125] Specifically, as a default setting, the theoretical cloud height range here is... The range is set to 2km to 10km, which can be adjusted based on detailed local measurements or actual meteorological observation data of the target area.

[0126] The calculation module is based on the solar azimuth angle. and the median of the distance interval The formulas for calculating the horizontal distance x (east-west direction) and the vertical distance y (north-south direction) from the center of the cloud-blocking point to the target photovoltaic site are as follows:

[0127] ,

[0128] ,

[0129] The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site And the vertical distance y, calculate the latitude of the center of the cloud-obscured point. The calculation formula is:

[0130] ,

[0131] The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site ,longitude And the horizontal distance x, calculate the longitude of the center of the cloud-obscured point. The calculation formula is:

[0132] .

[0133] In the eighth embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, based on the second embodiment, step S140 includes the following steps:

[0134] Step S810: The calculation module obtains the feature region of each time step in the future prediction period through the cloud-ground mapping model.

[0135] Step S820: The calculation module calculates the similarity score between the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of the same time period for each historical date within the past preset time period, and the feature regions therein.

[0136] Step S830: The calculation module selects the historical date with the highest similarity score under the set temperature constraint conditions. Extract the historical dates The photovoltaic power data for the forecast period is used as the similar historical power data.

[0137] In the ninth embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-to-ground mapping matching proposed in this invention, based on the eighth embodiment, the calculation formula for the similarity score of the feature regions in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of the same period on each historical date within the past preset duration is as follows:

[0138] ,

[0139] in, This represents the similarity score for the k-th historical date. These represent the historical satellite cloud image sequences for the current date within the predicted time period. This represents the sequence of satellite cloud images for the k-th historical date within the predicted time period; This represents the pixels located in the feature region. MAE This represents the mean absolute error.

[0140] In the tenth embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, based on the ninth embodiment, step S830 includes the following steps:

[0141] Step S1010: The calculation module introduces the Regional Average Temperature (RAT) as a temperature constraint condition to calculate the similarity score. The calculation method is as follows:

[0142] ,

[0143] Where m represents the total number of days of the preset duration (30 in this embodiment). The temperature constraint function is expressed as follows:

[0144] ,

[0145] in, The allowable temperature difference threshold is set to 5°C in this embodiment; Let be the regional average temperature on the k-th historical date. The average temperature for the region on the current date.

[0146] Specifically, the temperature constraint mentioned above means that a sample is considered usable for matching only when the temperature difference is within a reasonable range.

[0147] Step S1020: The calculation module selects the historical date with the highest similarity score. As similar dates, photovoltaic power data for the predicted period is extracted from similar dates as similar historical power data.

[0148] Specifically, the similar historical power data here is used as input for subsequent models.

[0149] In the eleventh embodiment of the distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching proposed in this invention, based on the first embodiment, the photovoltaic power prediction model based on artificial neural network includes, but is not limited to, convolutional neural network model, recurrent neural network model and long short-term memory network model.

[0150] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0151] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A distributed photovoltaic cluster power ultra-short-term prediction method based on cloud-ground mapping matching, characterized in that, The application relates to a photovoltaic cluster power ultra-short-term prediction system based on cloud-ground mapping matching; the system comprises a data collection module, a calculation module and a prediction module; the method further comprises the following steps: The data collection module obtains satellite cloud images covering the distributed photovoltaic cluster according to the geographical position of the distributed photovoltaic cluster; The prediction module obtains a historical satellite cloud image sequence of each time point in a preset time period in the past, determines a satellite cloud image set of a future prediction period based on the historical satellite cloud image sequence and a three-dimensional convolutional neural network model, and obtains feature region cloud image information of the future prediction period based on the satellite cloud image set of the future prediction period; The calculation module constructs a cloud-ground mapping model, determines a feature region associated with the power output of a target photovoltaic station in the satellite cloud image covering the distributed photovoltaic cluster based on the geographical position of the target photovoltaic station and the position of the sun, and calculates the similarity score of the feature region in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of each date in the same period in the past within a preset time period, filters out the most similar historical satellite cloud image under the set temperature constraint condition, extracts photovoltaic power data of the most similar historical satellite cloud image in the prediction period as similar historical power data, and constructs a photovoltaic power prediction model based on an artificial neural network. The prediction module obtains a historical satellite cloud image sequence of each time point in a preset time period in the past, determines a satellite cloud image set of a future prediction period based on the historical satellite cloud image sequence and a three-dimensional convolutional neural network model, and obtains feature region cloud image information of the future prediction period based on the satellite cloud image set of the future prediction period, comprising: The calculation module constructs a cloud-ground mapping model, determines a feature region associated with the power output of a target photovoltaic station in the satellite cloud image covering the distributed photovoltaic cluster based on the geographical position of the target photovoltaic station and the position of the sun, and calculates the similarity score of the feature region in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of each date in the same period in the past within a preset time period, filters out the most similar historical satellite cloud image under the set temperature constraint condition, extracts photovoltaic power data of the most similar historical satellite cloud image in the prediction period as similar historical power data, and constructs a photovoltaic power prediction model based on an artificial neural network.

2. The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method according to claim 1, characterized in that, The calculation module calculates the variation range of the intersection point according to the universal distribution of the cloud height, and combines the intersection point into a possible region of the intersection point in space as the feature region. The prediction module acquires satellite cloud images of T continuous time steps in a previous prediction period in a future prediction period in the satellite cloud images , and forms a historical satellite cloud image set ; The prediction module obtains a satellite cloud image set of a future prediction period based on a historical satellite cloud image set C Wherein, i represents the satellite cloud image of the i th sampling point arranged in time in the historical three-dimensional convolutional neural network model set C, i = 1, 2, …, T, and T represents the total number of satellite cloud images input into the three-dimensional convolutional neural network model. F represents the satellite cloud image set of the future prediction period obtained by inputting the historical satellite cloud image set C into the three-dimensional convolutional neural network model. 3.The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method according to claim 1, characterized in that, The calculation module calculates the distance L of the cloud shading point to the target photovoltaic station: The calculation module calculates the target photovoltaic site's geographical coordinates and solar hour angle. and solar declination angle Calculate the solar zenith angle at the current moment. ; The calculation module calculates the solar azimuth angle based on the solar zenith angle and the geographic coordinates of the target photovoltaic site ; The calculation module calculates the intersection of the sun rays directed to the target photovoltaic site with the cloud mass, depending on the solar zenith angle , the solar azimuth angle , and the cloud height. The calculation module calculates the similarity score of the feature region in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of each date in the same period in the past within a preset time period, filters out the most similar historical satellite cloud image under the set temperature constraint condition, extracts photovoltaic power data of the most similar historical satellite cloud image in the prediction period as similar historical power data, comprising:

4. The method of claim 3, wherein the cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method is characterized in that, The calculation module calculates the solar zenith angle at the current time according to geographical coordinates, a solar hour angle, and a solar declination angle of a target photovoltaic station , comprising:​​ The calculation module acquires a current date n, calculates a solar declination angle according to the current date n and a solar hour angle ; The calculation module obtains geographical coordinates of a target photovoltaic station, calculates a solar zenith angle based on the geographical coordinates of the target photovoltaic station, a solar declination angle , and a solar hour angle , and calculates the solar zenith angle , wherein the geographical coordinates of the target photovoltaic station include a longitude and a latitude . The calculation module calculates the solar azimuth angle based on the solar zenith angle and the geographic coordinates of the target photovoltaic site , comprising: The calculation module calculates the solar azimuth angle from the solar declination angle , the geographical coordinates of the target photovoltaic site, and the solar zenith angle .

5. The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method according to claim 4, characterized in that, The calculation module calculates the intersection of the sun rays directed to the target photovoltaic site with the cloud mass according to the solar zenith angle , the solar azimuth angle , and the cloud height, including: The computing module sets a theoretical cloud height range , in combination with the solar zenith angle , the intersection of the sunlight and each height cloud cluster in the direction from the sun to the target photovoltaic site is marked as a cloud blocking point, and the distance interval between each cloud blocking point and the target photovoltaic site is derived ; The computing module calculates the east-west horizontal distance x and the north-south vertical distance y from the center of the cloud blocking point to the target photovoltaic station according to the solar azimuth angle and the median of the distance interval , the computing module calculates the latitude of the center of the cloud-occluded point based on the Earth equatorial radius , the latitude of the target photovoltaic station , and the vertical distance y ; the calculation module calculates the longitude of the center of the cloud-occluded point based on the Earth's equatorial radius , the latitude of the target photovoltaic station , the longitude , and the horizontal distance x ; The calculation module uses the center of the cloud obscuration point as the center and the distance interval as the reference. The length is the radius A circular region is constructed on the satellite cloud image, and this circular region is used as the characteristic region that actually affects the power output of the target photovoltaic site.

6. The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method according to claim 3, characterized in that, The solar declination angle The calculation formula is: ， The solar hour angle The formula for calculating the solar hour angle is: ， where n is the digitized date expressed as the day of the year; is the ratio of the circumference of a circle to its diameter; represents the sine function; h is the digitized hour time, calculated as ， In the formula, represents the number of hours of the current time, represents the number of minutes of the current time.

7. The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method according to claim 4, characterized in that, The calculation module obtains the geographical coordinates of the target photovoltaic site, and then calculates the coordinates based on the geographical coordinates of the target photovoltaic site and the solar declination angle. and solar hour angle Calculate the solar zenith angle The calculation formula is: ， wherein denotes the latitude of the target photovoltaic site; denotes the solar declination angle; denotes the solar hour angle; denotes the sine function, denotes the cosine function, denotes the arcsine function; The calculation module is based on the solar declination angle. The geographical coordinates of the target photovoltaic site and the solar zenith angle. Calculate the solar azimuth angle The calculation formula is: ， In the formula, denotes the inverse cosine function; The computing module sets a theoretical cloud height range , in combination with the solar zenith angle , the intersection of the sunlight and each height cloud cluster in the direction from the sun to the target photovoltaic station is marked as a cloud blocking point, and the distance interval between each cloud blocking point and the target photovoltaic station is derived , comprising: The calculation module obtains the feature region of each time step in the future prediction period through the cloud-ground mapping model; ， where H denotes the cloud height, denotes the solar zenith angle, denotes the tangent function; The computing module calculates the east-west horizontal distance x and the south-north vertical distance y of the center of the cloud blocking point to the target photovoltaic station according to the sun azimuth angle and the median of the distance interval The calculation formula is as follows: ， ， The computing module calculates the latitude of the center of the cloud-shading point based on the Earth equatorial radius , the latitude of the target photovoltaic station , and the vertical distance y The calculation formula is as follows: ， The calculation module is based on the Earth's equatorial radius. Latitude of the target photovoltaic site ,longitude And the horizontal distance x, calculate the longitude of the center of the cloud-obscured point. The calculation formula is: 。 8.The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short term prediction method according to claim 2, characterized in that, The calculation module calculates the similarity score of the feature region in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of each date in the same period in the past within a preset time period; The calculation module calculates the similarity score of the feature region in the satellite cloud image of the future prediction period and the historical satellite cloud image sequence of each date in the same period in the past within a preset time period; ​ the computing module selects a historical date with the highest similarity score under a set temperature constraint extracts the historical date photovoltaic power data of a prediction period as the similar historical power data. 9.The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short term prediction method according to claim 8, characterized in that, The calculation module calculates the similarity score of the satellite cloud image of the future prediction period and the feature region in the historical satellite cloud image sequence of the same period of each historical date in the past preset time length, and the calculation formula is as follows: ， wherein, denotes a similarity score for the kth historical date, denotes a historical sequence of satellite cloud images for the current date at the prediction period, denotes a sequence of satellite cloud images for the kth historical date at the prediction period; denotes a pixel located at a feature region, MAE is the mean absolute error.

10. The cloud-to-ground mapping matching based distributed photovoltaic cluster power ultra-short-term prediction method according to claim 9, characterized in that, the computing module selects a historical date with the highest similarity score under a set temperature constraint extracts the historical date photovoltaic power data of a prediction period as the similar historical power data, including: The calculation module introduces the regional average temperature as a temperature constraint condition, and introduces the temperature constraint condition to the calculation of the similarity score, and the calculation method is as follows: ， wherein m represents the total number of days in the past preset time length, represents a temperature constraint function, which is described as follows: ， wherein, is a temperature difference threshold value; is a regional average temperature for the kth historical date, is a regional average temperature for the current date; The computing module selects the highest similarity score of the historical dates As a similar date, the photovoltaic power data of the prediction period is extracted from the similar dates as similar historical power data.

Images