Distributed photovoltaic output power accurate prediction method, medium and system

Abstract

The invention provides a distributed photovoltaic output power accurate prediction method, medium and system, and relates to the technical field of distributed photovoltaic power prediction, and the method comprises a meteorological environment influence deviation detection step and a photovoltaic output power prediction step. Obtaining a plurality of typical meteorological days according to the influence of the meteorological environment on the photovoltaic panel, and obtaining power generation deviation vectors corresponding to the typical meteorological days according to the theoretical output power and the actual output power of the typical meteorological days; and calculating prediction theoretical output power according to the meteorological environment of the prediction day, finding a typical meteorological day with the highest similarity according to the meteorological environment of the prediction day, and correcting the prediction theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity to obtain photovoltaic output power of the prediction day. The influence of the meteorological environment on the power generation of the photovoltaic panel is comprehensively considered, a power generation deviation vector is obtained, then the predicted theoretical output power is corrected according to the power generation deviation vector, and the photovoltaic output power of the real prediction day is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of accurate prediction of distributed photovoltaic output power, and specifically relates to an accurate prediction method, medium and system for distributed photovoltaic output power. Background Art

[0002] The green and low-carbon transformation of energy and electricity has become a trend. The distributed photovoltaic system of Guyuan Power Grid has also ushered in a blowout growth, from 2 grid-connected points in 2015 to 1,822 in August 2024. Affected by its small single-unit scale, decentralized access, and strong randomness, large-scale distributed photovoltaics do not have the conditions for real-time monitoring and rolling prediction at the dispatching main station, and are in the blind spot of dispatching agency regulation. With the large-scale access of distributed photovoltaics, the voltage limit and reverse flow of the distribution network have become more and more obvious, and the phenomenon of squeezing the main grid's new energy consumption space has become more and more prominent, seriously restricting the healthy and sustainable development of the clean energy industry; it has also greatly increased the complexity and difficulty of control of the power grid, affecting the reliability of new energy operation and the safe production capacity of the power grid; this has put forward higher requirements for the planning, dispatching and operation management of the power system, and it is necessary to accurately predict the power of photovoltaic power stations to ensure the stability of the operation of the power system. However, due to the complex and changeable meteorological environment, the actual power generation of photovoltaic power generation is inconsistent with the expected power generation. Summary of the invention

[0003] In view of this, the present invention provides a method for accurately predicting distributed photovoltaic output power with improved prediction accuracy.

[0004] It is also necessary to provide an accurate prediction system for distributed photovoltaic output power.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] A distributed photovoltaic output power accurate prediction method includes a meteorological environment impact deviation detection step and a photovoltaic output power prediction step.

[0007] The meteorological environment impact deviation detection step obtains several typical meteorological days according to the impact of the meteorological environment on the photovoltaic panel, and obtains the power generation deviation vector corresponding to the typical meteorological day according to the theoretical output power and the actual output power of the typical meteorological day;

[0008] The photovoltaic output power prediction step calculates the predicted theoretical output power according to the meteorological environment of the prediction day and finds the typical meteorological day with the highest similarity according to the meteorological environment of the prediction day, and corrects the predicted theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity to obtain the photovoltaic output power of the prediction day.

[0009] Preferably, the meteorological environment impact deviation detection step comprises the following steps:

[0010] Step A1: obtaining multiple planar images of each photovoltaic station of the distributed photovoltaic system and obtaining historical meteorological data for one consecutive year;

[0011] Step A2: Establish the three-dimensional spatial position of each photovoltaic panel based on multiple planar images.

[0012] Several typical meteorological days are formed based on the historical meteorological data of one consecutive year;

[0013] Step A3: Calculate the theoretical output power of each typical meteorological day according to the three-dimensional spatial position and physical parameters of each photovoltaic panel;

[0014] Step A4: Obtain the actual output power of each typical meteorological day, and compare the actual output power of each typical meteorological day with the theoretical output power to obtain the power generation deviation vector corresponding to each typical meteorological day.

[0015] Preferably, the photovoltaic output power prediction step comprises the following steps:

[0016] Step B1: Obtain the meteorological forecast data for the forecast day, and calculate the predicted theoretical output power according to the three-dimensional spatial position and physical parameters of each photovoltaic panel;

[0017] Step B2: According to the meteorological forecast data of the forecast day, obtain the typical meteorological day with the highest similarity to step 2, and use the power generation deviation vector of the typical meteorological day with the highest similarity to correct the predicted theoretical output power of the forecast day to obtain the photovoltaic output power of the forecast day.

[0018] Preferably, the step A3 calculates the theoretical output power of each typical meteorological day according to the three-dimensional spatial posture and physical parameters of each photovoltaic panel, and specifically includes the following steps:

[0019] A31: Divide each typical meteorological day into a period according to the same time to obtain several periods;

[0020] A32: Obtain the illuminance of a given period of time according to the three-dimensional spatial position and physical parameters of each photovoltaic panel and the position of the sun in that period of time;

[0021] A33: Obtain the theoretical output power for each period of a typical meteorological day based on the illuminance.

[0022] Preferably, the step A4 obtains the actual output power of each typical meteorological day, and obtains the power generation deviation vector corresponding to each typical meteorological day by comparing the actual output power of each typical meteorological day with the theoretical output power, which specifically includes the following steps:

[0023] A41: Get the actual output power in each period of a typical meteorological day;

[0024] A42: Compare the actual output power with the theoretical output power in each period of each typical meteorological day to obtain the power generation deviation value in each period of each typical meteorological day;

[0025] A43: Traverse the power generation deviation values ​​of all time periods to obtain the power generation deviation vector corresponding to each typical meteorological day.

[0026] Preferably, step B2: according to the meteorological forecast data of the forecast day, obtaining the typical meteorological day with the highest similarity with step 2, and using the power generation deviation vector of the typical meteorological day with the highest similarity to correct the predicted theoretical output power of the forecast day to obtain the photovoltaic output power of the forecast day, specifically includes the following steps:

[0027] B21: Based on the meteorological forecast data of the forecast day, calculate the similarity of several typical meteorological days formed with the historical meteorological data, and determine the typical meteorological day with the highest similarity;

[0028] B22: extract the power generation deviation vector of the typical meteorological day with the highest similarity, calculate the average value of the vector, and obtain the power generation deviation correction;

[0029] B23: The predicted theoretical output power is corrected by the power generation deviation correction to obtain the photovoltaic output power on the predicted day.

[0030] Preferably, it also includes a self-loss deviation detection step, wherein the self-loss deviation detection step calculates the power generation efficiency reduction rate of the corresponding photovoltaic panel and the average power generation efficiency reduction rate of the entire distributed photovoltaic system according to the image of the photovoltaic panel and a pre-set photovoltaic panel power generation efficiency reduction rate-surface similarity database;

[0031] The photovoltaic output power prediction step corrects the predicted theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity, and then corrects it again according to the average power generation efficiency reduction rate of the entire distributed photovoltaic system to obtain the photovoltaic output power of the predicted day.

[0032] Preferably, the self-loss deviation detection step comprises the following steps:

[0033] Step C1: obtaining a high-definition image of each photovoltaic panel of the distributed photovoltaic system taken orthogonally;

[0034] Step C2: performing similarity analysis based on the high-definition image taken orthogonally to each photovoltaic panel and the high-definition image taken orthogonally to the newly installed photovoltaic panel, and obtaining a similarity used to describe the surface of the photovoltaic panel and recording it as surface similarity;

[0035] Step C3: using the pre-set photovoltaic panel power generation efficiency reduction rate-surface similarity database, calculate the power generation efficiency reduction rate of the corresponding photovoltaic panel and the average power generation efficiency reduction rate of the entire distributed photovoltaic system.

[0036] A computer-readable storage medium stores program instructions, which are used to execute the above-mentioned method for accurately predicting distributed photovoltaic output power when the program instructions are executed.

[0037] A distributed photovoltaic output power accurate prediction system includes the computer-readable storage medium as described above.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention provides a method for accurately predicting distributed photovoltaic output power, comprising a meteorological environment impact deviation detection step and a photovoltaic output power prediction step; according to the impact of the meteorological environment on photovoltaic panels, several typical meteorological days are obtained, and the power generation deviation vector corresponding to the typical meteorological day is obtained according to the theoretical output power and the actual output power of the typical meteorological day; the predicted theoretical output power is calculated according to the meteorological environment of the predicted day and the typical meteorological day with the highest similarity is found according to the meteorological environment of the predicted day, the predicted theoretical output power is corrected according to the power generation deviation vector of the typical meteorological day with the highest similarity to obtain the photovoltaic output power of the predicted day, the impact of the meteorological environment on the photovoltaic panel power generation is comprehensively considered to obtain the power generation deviation vector, and then the predicted theoretical output power is corrected according to the power generation deviation vector to obtain the photovoltaic output power of the actual predicted day, so as to accurately reflect the actual power generation situation of the photovoltaic panel, predict the power generation more accurately, and make the power system operation more stable. DETAILED DESCRIPTION

[0040] The technical solutions and technical effects of the embodiments of the present invention are further elaborated in detail below.

[0041] A distributed photovoltaic output power accurate prediction method includes a meteorological environment impact deviation detection step and a photovoltaic output power prediction step.

[0042] The meteorological environment impact deviation detection step obtains several typical meteorological days according to the impact of the meteorological environment on the photovoltaic panel, and obtains the power generation deviation vector corresponding to the typical meteorological day according to the theoretical output power and the actual output power of the typical meteorological day;

[0043] The photovoltaic output power prediction step calculates the predicted theoretical output power according to the meteorological environment of the prediction day and finds the typical meteorological day with the highest similarity according to the meteorological environment of the prediction day, and corrects the predicted theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity to obtain the photovoltaic output power of the prediction day.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention provides a method for accurately predicting distributed photovoltaic output power, comprising a meteorological environment impact deviation detection step and a photovoltaic output power prediction step; according to the impact of the meteorological environment on photovoltaic panels, several typical meteorological days are obtained, and the power generation deviation vector corresponding to the typical meteorological day is obtained according to the theoretical output power and the actual output power of the typical meteorological day; the predicted theoretical output power is calculated according to the meteorological environment of the predicted day and the typical meteorological day with the highest similarity is found according to the meteorological environment of the predicted day, the predicted theoretical output power is corrected according to the power generation deviation vector of the typical meteorological day with the highest similarity to obtain the photovoltaic output power of the predicted day, the impact of the meteorological environment on the photovoltaic panel power generation is comprehensively considered to obtain the power generation deviation vector, and then the predicted theoretical output power is corrected according to the power generation deviation vector to obtain the photovoltaic output power of the actual predicted day, so as to accurately reflect the actual power generation situation of the photovoltaic panel, predict the power generation more accurately, and make the power system operation more stable.

[0046] Furthermore, the meteorological environment impact deviation detection step includes the following steps:

[0047] Step A1: obtaining multiple plane images of each photovoltaic station of the distributed photovoltaic system and obtaining historical meteorological data for one consecutive year, wherein the plane images must include all photovoltaic panels of each photovoltaic station; the historical meteorological data at least includes temperature, relative humidity, wind speed and direction, sunshine hours, etc., and fully considering meteorological factors, so as to take into account all meteorological factors that affect photovoltaic panel power generation as much as possible to reduce errors;

[0048] Specifically, a high-definition camera can be installed on a drone, and the drone can be controlled by programming to conduct aerial surveys of the photovoltaic station according to a predetermined route. The aerial survey track covers the entire range of the photovoltaic station, and the aerial survey height ensures that the resolution of the captured image is not less than 10cm. The aerial survey spacing is designed according to the spacing of the photovoltaic panels, and the captured images need to have at least 55%-65% overlap in the front-to-back direction and 25%-35% overlap in the lateral direction.

[0049] Specifically, historical meteorological data of the target PV station in the past year is collected. The data source can be actual measurements at the meteorological station or satellite remote sensing inversion.

[0050] Step A2: establishing a three-dimensional spatial position and posture of each photovoltaic panel according to the multiple plane images. Specifically, establishing a three-dimensional image of each photovoltaic station according to the multiple plane images obtained; and calculating the three-dimensional spatial position and posture of each photovoltaic panel according to the three-dimensional image of the photovoltaic station;

[0051] According to the historical meteorological data of one consecutive year, several typical meteorological days are formed. Specifically, a clustering algorithm is used to form several typical meteorological days, so as to classify multiple complex meteorological weathers, so as to facilitate the calculation of theoretical output power;

[0052] Specifically, the aerial survey images are processed, the longitude, latitude and elevation information in the drone route data is extracted, and the images are oriented. Then, feature matching algorithms such as SIFT are used to match features between adjacent images and calculate the relative orientation relationship between images. Iterative optimization of the Adjustment algorithm is used to obtain the absolute orientation elements of all images; a three-dimensional point cloud model of the photovoltaic station is generated through the PMVS or MVS grid reconstruction algorithm; the generated three-dimensional point cloud of the photovoltaic station is filtered to remove outliers and noise points; the filtered three-dimensional point cloud is plane segmented to extract the ground plane; the point cloud can be fitted using algorithms such as RANSAC to segment the point cloud into ground points and non-ground points; the extracted ground plane is used as the reference plane; the feature points of the photovoltaic panel are extracted, and generally the four vertices and several boundary points of the photovoltaic panel are selected as feature points; the vertices of the photovoltaic panel can be found through the topological analysis of the three-dimensional point cloud, or the feature points of the photovoltaic panel can be manually calibrated; the projection coordinates of each feature point of the photovoltaic panel on the reference ground are calculated to determine the two-dimensional boundary of the photovoltaic panel on the reference plane; the three-dimensional point fitting method is used to fit the photovoltaic panel plane with the feature points of the photovoltaic panel as data points, and the normal vector and plane equation are obtained to determine the spatial posture of the photovoltaic panel. The normal vector of the line connecting the feature points can also be directly calculated as the normal vector of the photovoltaic panel. Finally, the three-dimensional spatial position of the photovoltaic panel can be determined based on the spatial posture and two-dimensional boundary of each photovoltaic panel; based on the three-dimensional point cloud data, the position and posture of each photovoltaic panel in three-dimensional space can be accurately calculated.

[0053] Among them, commonly used filtering algorithms include statistical filtering, radial basis function filtering, etc.

[0054] Specifically, the historical meteorological data are tested and obvious outliers are corrected, missing data are supplemented, and the data are smoothed using methods such as time series analysis to ensure that the meteorological data are accurate and of high quality. Then, based on the time series clustering algorithm, the processed historical meteorological data are clustered and analyzed to cluster various meteorological conditions in a year into multiple typical meteorological days. According to the clustering results, the values ​​of various meteorological parameters of each cluster center, i.e., the typical meteorological day, are calculated. These typical meteorological days can represent various typical meteorological conditions that may occur in a year. Finally, the final number of typical meteorological day categories is determined based on the number of clusters and the percentage of explained variance. The number of typical meteorological day categories is generally between 10 and 30, which lays the foundation for subsequent theoretical power calculations.

[0055] Commonly used clustering algorithms include K-means, hierarchical clustering, DBSCAN, etc.

[0056] Step A3: Calculate the theoretical output power of each typical meteorological day according to the three-dimensional spatial position and physical parameters of each photovoltaic panel, so as to accurately calculate the theoretical output power of each typical meteorological day according to the location of the photovoltaic panel and weather reasons, so that the theoretical output power is calculated accurately;

[0057] Step A4: obtaining the actual output power of each typical meteorological day, and obtaining the power generation deviation vector corresponding to each typical meteorological day by comparing the actual output power of each typical meteorological day with the theoretical output power. The power generation deviation vector corresponding to each typical meteorological day obtained by the accurate calculation of the theoretical output power of each typical meteorological day is made more accurate.

[0058] Specifically, according to the complexity of the meteorological environment, different meteorological conditions are classified, and a power generation statistical model under different meteorological conditions is established. In order to classify and analyze the complex meteorological conditions, it is convenient to obtain the power generation deviation caused by meteorological reasons and the power generation deviation vector corresponding to each typical meteorological day.

[0059] Furthermore, the photovoltaic output power prediction step includes the following steps:

[0060] Step B1: Obtain the meteorological forecast data for the forecast day, and calculate the predicted theoretical output power according to the three-dimensional spatial position and physical parameters of each photovoltaic panel;

[0061] Specifically, the steps are the same as A3. The theoretical maximum output power of each photovoltaic panel in each time period can be calculated based on the light intensity. The theoretical maximum output power of all photovoltaic panels in each time period is summarized to obtain the theoretical predicted theoretical output power in each time period of the entire prediction day, forming a synthetic power prediction curve.

[0062] Step B2: According to the meteorological forecast data of the forecast day, obtain the typical meteorological day with the highest similarity with the one in step 2, and use the power generation deviation vector of the typical meteorological day with the highest similarity to correct the predicted theoretical output power of the forecast day to obtain the photovoltaic output power of the forecast day. By comparison, the typical meteorological day with the highest similarity with the meteorological data of the forecast day is obtained, and then the accurate power generation deviation vector is obtained, and then the predicted theoretical output power of the forecast day is corrected, so that the photovoltaic output power of the forecast day is more realistic and in line with reality.

[0063] Furthermore, the step A3 calculates the theoretical output power of each typical meteorological day according to the three-dimensional spatial posture and physical parameters of each photovoltaic panel, and specifically includes the following steps:

[0064] A31: Divide each typical meteorological day into a period according to the same time to obtain several periods;

[0065] Specifically, each day of the year is matched to a corresponding typical meteorological day, and for each typical meteorological day, the time in the day is divided into several time periods according to the time series with the same time as a time period.

[0066] Specifically, each typical meteorological day can be divided into 72 to 96 time periods with each period lasting 15 to 20 minutes.

[0067] A32: Obtain the illuminance of a given period of time according to the three-dimensional spatial position and physical parameters of each photovoltaic panel and the position of the sun in that period of time;

[0068] Specifically, according to the time, longitude and latitude and other parameters of each typical meteorological day, an astronomical algorithm is used to calculate the solar declination, right ascension and hour angle at the corresponding time, and then converted to the local solar altitude angle. According to the position and posture of each photovoltaic panel, physical parameters (area, efficiency, etc.) and the position of the sun in the time period, a physical model is used to calculate the illumination of the photovoltaic panel in the time period.

[0069] Specifically, typical physical models include HDKR, Perez, etc.

[0070] A33: Obtain the theoretical output power for each period of a typical meteorological day based on the illuminance.

[0071] Furthermore, the step A4 obtains the actual output power of each typical meteorological day, and obtains the power generation deviation vector corresponding to each typical meteorological day by comparing the actual output power of each typical meteorological day with the theoretical output power, which specifically includes the following steps:

[0072] A41: Get the actual output power in each period of a typical meteorological day;

[0073] Specifically, the actual power generation data of each typical meteorological day in the historical year is obtained. The data collection time is the same as the time of each time period. When there is missing data or obvious erroneous data, it must be corrected in time.

[0074] A42: Compare the actual output power with the theoretical output power in each period of each typical meteorological day to obtain the power generation deviation value in each period of each typical meteorological day;

[0075] A43: Traverse the power generation deviation values ​​of all time periods to obtain the power generation deviation vector corresponding to each typical meteorological day.

[0076] Specifically, according to the power generation deviation value of each time period in each typical meteorological day, a power generation deviation vector of several time period elements of each typical meteorological day is obtained, and each element represents the power generation deviation value of a time period.

[0077] Further, step B2: according to the meteorological forecast data of the forecast day, obtain the typical meteorological day with the highest similarity with step 2, and use the power generation deviation vector of the typical meteorological day with the highest similarity to correct the predicted theoretical output power of the forecast day to obtain the photovoltaic output power of the forecast day, which specifically includes the following steps:

[0078] B21: Based on the meteorological forecast data of the forecast day, calculate the similarity of several typical meteorological days formed with the historical meteorological data, and determine the typical meteorological day with the highest similarity;

[0079] Specifically, based on the meteorological parameters of the forecast day, its meteorological similarity with each typical meteorological day is calculated, and one or several typical meteorological days with the most similar meteorological conditions to the forecast day are determined.

[0080] Specifically, methods such as the Euclidean distance and correlation coefficient of meteorological elements may be used; similarity may be determined by registering and sorting and setting a threshold.

[0081] B22: extract the power generation deviation vector of the typical meteorological day with the highest similarity, calculate the average value of the vector, and obtain the power generation deviation correction;

[0082] B23: The predicted theoretical output power is corrected by the power generation deviation correction to obtain the photovoltaic output power on the predicted day.

[0083] Specifically, a power prediction curve of a time series of theoretical output power predicted on a prediction day is obtained, and an offset is performed according to a power generation deviation correction amount to obtain a photovoltaic output power curve of a time series of predicted power generation after preliminary correction.

[0084] Furthermore, it also includes a self-loss deviation detection step, wherein the self-loss deviation detection step calculates the power generation efficiency reduction rate of the corresponding photovoltaic panel and the average power generation efficiency reduction rate of the entire distributed photovoltaic system according to the image of the photovoltaic panel and a pre-set photovoltaic panel power generation efficiency reduction rate-surface similarity database;

[0085] The photovoltaic output power prediction step corrects the predicted theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity, and then corrects it again through the average power generation efficiency reduction rate of the entire distributed photovoltaic system to obtain the photovoltaic output power of the predicted day, so as to take into account the influence of meteorological reasons and the loss of the photovoltaic panel itself on the power generation of the photovoltaic panel, so that the photovoltaic output power of the predicted day is more accurate and true.

[0086] Specifically, the location information of each photovoltaic panel is obtained by three-dimensional acquisition of photovoltaic stations, and then the historical meteorological data of the area for one year is collected. Clustering algorithms such as K-means are used to extract several typical meteorological days representing different meteorological conditions in a year. On each typical meteorological day, the theoretical power generation of each period is calculated. The theoretical power generation is compared with the actual power generation, and a statistical model of power generation under various weather conditions is established. At the same time, the orthogonal images of photovoltaic panels are processed using convolutional neural networks and other methods to predict their service life. The power generation efficiency attenuation data of photovoltaic panels is further obtained from the age data. Finally, the meteorological model and the attenuation model are integrated to fully consider the factors affecting the power generation of photovoltaic panels, so as to achieve accurate prediction of the power generation of distributed photovoltaic systems.

[0087] Furthermore, the self-loss deviation detection step includes the following steps:

[0088] Step C1: obtaining a high-definition image of each photovoltaic panel of the distributed photovoltaic system taken orthogonally;

[0089] Specifically, the operator holds a structured light camera or LIDAR device to scan each photovoltaic panel in an orthogonal direction to ensure that the scanning range covers the entire photovoltaic panel and the scanning resolution is not less than 10 cm. Or the operator can use a drone for orthogonal shooting, control the drone to hover above each photovoltaic panel through programming, and control the camera to aim at the photovoltaic panel for orthogonal shooting, and perform three-dimensional reconstruction on the orthogonal image to obtain a three-dimensional model of the photovoltaic panel.

[0090] Among them, orthogonal shooting must ensure that the entire photovoltaic panel is within the framing range and the image is clear.

[0091] Step C2: performing similarity analysis based on the high-definition image taken orthogonally to each photovoltaic panel and the high-definition image taken orthogonally to the newly installed photovoltaic panel, and obtaining a similarity used to describe the surface of the photovoltaic panel and recording it as surface similarity;

[0092] Specifically, the orthogonal images of photovoltaic panels in historical years are preprocessed using image processing algorithms, including cropping, resizing, contrast enhancement, etc., to obtain images with enhanced effects, so that the images are clear and convenient for similarity comparison in the later stage. Orthogonal image samples of photovoltaic panels with different service years and different dust accumulation degrees are collected, and the service life and dust accumulation degrees of the images are manually annotated to form a training data set with multiple labels. The influence of the service life and dust accumulation degree of photovoltaic panels on the actual power generation of photovoltaic panels is considered from two aspects, so that the photovoltaic output power on the predicted day is more accurate and the operation of the power system is more stable.

[0093] Step C3: using the pre-set photovoltaic panel power generation efficiency reduction rate-surface similarity database, calculate the power generation efficiency reduction rate of the corresponding photovoltaic panel and the average power generation efficiency reduction rate of the entire distributed photovoltaic system.

[0094] Specifically, when training the convolutional neural network model, the photovoltaic panel image is input, and the predicted values ​​of the two dimensions of the service life and dust accumulation degree of the photovoltaic panel are output; the network structure and parameters are adjusted to complete the training; for each historical photovoltaic panel image, its service life and dust accumulation degree are predicted using the trained model; according to the predicted service life and dust accumulation degree, the power generation efficiency reduction rate database is queried to obtain the composite power generation efficiency reduction rate of each photovoltaic panel, and the arithmetic mean of the power generation efficiency reduction rate of all photovoltaic panels in the field is calculated as the average efficiency reduction rate of the whole field.

[0095] Among them, the decline rate in the database is determined based on two factors: the years of use and the degree of dust accumulation, which is more comprehensive and accurate.

[0096] Among them, the method for establishing the photovoltaic panel power generation efficiency reduction rate-surface similarity database is as follows:

[0097] 1. Collect image samples of photovoltaic panels of different service life. The service life range can be set from 0 to 25 years.

[0098] 2. For each image sample, measure the actual conversion efficiency of the photovoltaic panel using professional testing instruments.

[0099] 3. For each image sample, use image processing methods to remove the background and extract the surface area of ​​the photovoltaic panel.

[0100] 4. For all surface image samples, use image similarity algorithms (such as SSIM, SIFT, etc.) to calculate the similarity between images pairwise.

[0101] 5. Sort all samples by age from small to large, calculate the similarity of sample images between two adjacent years, and express it as the surface similarity of photovoltaic panels of that age.

[0102] 6. Based on the conversion efficiency data obtained from the test, calculate the efficiency decline rate for two consecutive years.

[0103] 7. The efficiency decline rate between years is matched with the corresponding surface similarity to form a database.

[0104] 8. Audit and verify samples in the database with similar years but large differences in efficiency decline rates to improve database quality.

[0105] 9. After the database is formed, the surface similarity of the newly acquired photovoltaic panel image can be predicted, and then the database can be queried to obtain the power generation efficiency reduction rate to complete the mapping with the image information.

[0106] A computer-readable storage medium stores program instructions, which are used to execute the above-mentioned method for accurately predicting distributed photovoltaic output power when the program instructions are executed.

[0107] A distributed photovoltaic output power accurate prediction system includes the computer-readable storage medium as described above.

[0108] The following is a specific embodiment of the method of the present invention, wherein the formulas or variables used in the embodiment are explained as follows:

[0109] p i =(x i ,y i , z i ) T : Represents the coordinates of the i-th point in the three-dimensional point cloud.

[0110] Calculate point p i The average distance to its K nearest neighbors.

[0111] Represents point p i The set of K nearest neighbor indices.

[0112] V={v 1 , v 2 , v 3 , v 4}: represents the coordinates of the four vertices of the photovoltaic panel.

[0113] n=(A * , B * , C * ) T : is the fitted ground normal vector.

[0114] Compute the projected coordinates of a point on a reference plane.

[0115] X i =(T i , H i , WS i , W.D. i , S.D. i ) T : represents the meteorological elements at the i-th moment;

[0116] is the explained variance ratio of the clusters.

[0117] S k , η k : are the area and conversion efficiency of the kth photovoltaic panel respectively.

[0118] E k，j =f(θj , β k , γ k , ...): Calculate the illumination of photovoltaic panel k in time period j.

[0119] P k，j =E k，j S k η k ; Calculate the output power of photovoltaic panels.

[0120] I' k =f(I k ): is the image enhancement function.

[0121] G(I';θ): is a convolutional neural network for predicting the life of photovoltaic panels.

[0122] Calculate the average efficiency drop rate.

[0123] X' j : It is the meteorological data of the jth period on the forecast day.

[0124] P'=[P' 1 , P' 2 , ..., P' 96 ]: It is the power prediction curve for the prediction day.

[0125] s i =Sim(X',c i ): Calculate meteorological similarity.

[0126] D i* =[ΔP 1 , ..., ΔP 96 ]: is the power generation deviation vector of a typical meteorological day.

[0127] Perform attenuation correction on power timing.

[0128] The following are the specific steps of the embodiment:

[0129] Step A1: Obtain multiple planar images of each photovoltaic station in the distributed photovoltaic system; collect historical meteorological data of the target photovoltaic station in the past year. The meteorological data needs to include temperature T, relative humidity H, wind speed WS, wind direction WD, sunshine hours SD, etc., with a time resolution of 10 minutes or higher.

[0130] Step A2: Generate a 3D point cloud P of the photovoltaic station based on multiple plane images = {p 1 , p 2 , ..., p N} Perform filtering to remove outliers and noise points; let p i =(xi ,y i , z i ) T represents the i-th 3D point, N is the total number of points in the point cloud. Traverse all points p i , calculate the average distance d to other points in the neighborhood i : in Represents point p i The K nearest neighbor index set, K is the specified number of neighbors; i is the threshold, when p i The distance to most of the neighboring points is significantly greater than d i When p i The outliers are removed; then, the plane is segmented on the filtered 3D point cloud to extract the ground plane. The RANSAC algorithm can be used to fit the point cloud and segment the point cloud into ground points and non-ground points. Let the plane model be: Ax+By+Cz+D=0; randomly select three points to determine a set of fitting plane coefficients (A, B, C, D), and calculate the distance from all points to the plane: Count the number of inliers n that are less than the threshold τ. When n exceeds the preset threshold N, the current plane model meets the ground assumption, and iterate to solve the ground plane model with the largest number of inliers (A * , B * , C * , D * ). The extracted ground plane point set is used as the reference surface. Next, the photovoltaic panel feature points are extracted. Generally, the four vertex coordinates of the photovoltaic panel are selected. 1 , v 2 , v 3 , v 4} and several boundary points as feature points. The vertices of the photovoltaic panel can be found through topological analysis of the three-dimensional point cloud, or the feature points of the photovoltaic panel can be manually calibrated. Then, the projection coordinates of each photovoltaic panel feature point on the reference ground are calculated to determine the two-dimensional boundary of the photovoltaic panel on the reference plane. Let the ground normal vector be n = (A * , B * , C * ) T , for each feature point v i Calculate its projection point u on the reference plane i : Where o = (0, 0, 0) T is the origin of the reference plane, u i =(x' i , y' i ,0) T That is v iProjection coordinates on the ground. The projection coordinates of all feature points determine the two-dimensional boundary of the photovoltaic panel on the ground. Then, the three-dimensional point fitting method is used to fit the photovoltaic panel plane with the feature points of the photovoltaic panel as data points to obtain the normal vector and plane equation to determine the spatial posture of the photovoltaic panel. Construct the feature point coordinate matrix V and use singular value decomposition (SVD) for plane fitting: V = UΣV T ; The last column of U and V is the normal vector n of the photovoltaic panel fitting plane; the last singular value of Σ corresponds to the normal vector n. Finally, the three-dimensional spatial position of the photovoltaic panel can be determined based on the spatial attitude normal vector n of each photovoltaic panel and the two-dimensional boundary vertex set U.

[0131] Conduct quality control on historical meteorological data, check and correct obvious outliers, and supplement missing data. Use time series analysis and other methods to smooth the data.

[0132] Assume that the total length of a year is N, and the meteorological data at the i-th moment in a year is: X i =(T i , H i , WS i , W.D. i , S.D. i ) T , i = 1, 2, ..., N; for abnormal data X i , find the adjacent m adjacent data and calculate the local mean Replace the abnormal points. For missing data, linear interpolation is used to supplement; then, based on the time series clustering algorithm, cluster analysis is performed on the processed historical meteorological data. Assume that the number of cluster categories is k, and after clustering, the class centers C of k categories are obtained. 1 , c 2 , ..., c k}. c j Represents the characteristic vector of the jth typical meteorological day; based on the clustering results, calculate the values ​​of various meteorological parameters of each cluster center, i.e., the typical meteorological day; these typical meteorological days can represent various typical meteorological conditions that may occur in a year; finally, determine the final number of typical meteorological day categories M based on the number of clusters k and the explained variance percentage EVR. The value of M is generally between 10 and 30. Repeat the above engineering adjustment parameters until the EVR result is reasonable.

[0133] Step A3: Match each day of the year to a corresponding typical meteorological day according to the determined number of typical meteorological day categories M. For example, by calculating the Euclidean distance between each day and each typical meteorological day, classify each day into the nearest typical meteorological day.

[0134] Suppose the weather data for day t of the year is X t, calculate its difference with the typical meteorological day c i Distance: X t Classification to d(X t , c i* ) is the smallest class, then X t The corresponding typical meteorological day category is i * ; Then, for each typical meteorological day, the time series within the day is divided into 15-minute periods. In this way, the day is evenly divided into periods. Let the jth period be t j =[(j-1)×15, j×15]. Next, calculate the solar altitude angle for each period. j , longitude and latitude, and other parameters, using astronomical algorithms to calculate the solar declination δ at the corresponding time j , right ascension α j Hour angle H j , and converted to the current solar altitude angle θ j :θ j =arcsin(sinδ j sinφ+cosδ j oeLh j cosφ); where φ is the geographic latitude. Then, according to the location and posture of each photovoltaic panel, physical parameters (area S k , Rated power generation efficiency η k etc.) and the time period t j The sun's position θ j , the physical model is used to calculate the illumination E of the photovoltaic panel during this period k，j . E k，j =f(θ j , β k , γ k , ...); where β k , γ k are the depression angle and azimuth angle of the kth photovoltaic panel respectively.

[0135] Then, the theoretical maximum output power P of the photovoltaic panel in this period can be calculated based on the illumination. k，j :P k，j =E k， j S k η k ; Calculate the theoretical maximum output power of all photovoltaic panels in all time periods in sequence.

[0136] Finally, the theoretical maximum output power of all photovoltaic panels in each period can be summarized to obtain the theoretical maximum power value P of the entire power plant in each period on the typical meteorological day. j : Where N is the total number of photovoltaic panels.

[0137] Step A4: Obtain the actual output power of each typical meteorological day, and compare the actual output power of each typical meteorological day with the theoretical output power to obtain the power generation deviation vector corresponding to each typical meteorological day.

[0138] Step C1: obtaining a high-definition image of each photovoltaic panel of the distributed photovoltaic system taken orthogonally;

[0139] Step C2: Use image processing algorithms to pre-process the orthogonal images of photovoltaic panels in historical years, including cropping, resizing, contrast enhancement, etc., to obtain enhanced images I k , where j represents the kth photovoltaic panel; commonly used image enhancement methods include histogram equalization, Gamma correction, median filtering, etc.: I' k =f(I k ); then, using deep learning models such as convolutional neural networks, a classifier is trained to take the new photovoltaic panel image as input and output similar categories of new and old photovoltaic panels.

[0140] Assume the number of photovoltaic panel status types is N c , collect the photovoltaic panel image sample set S = {I 1 , I 2 , ..., I Γ}, and its corresponding label is L = {l 1 , l 2 , ..., l Γ}, l i ∈[1,N c ], Γ is the number of sample sets.

[0141] Construct a convolutional neural network F(I;θ) and train it to minimize the classification loss function: Next, we collect orthogonal image samples of photovoltaic panels of different service life, and manually annotate the service life of the images to form a labeled training data set. y , the label of the year is Y={y 1 ,y 2 , ..., y M},y i ∈[1,N y ].

[0142] Step C3: The convolutional neural network G(I';θ) can also be trained to predict the life span: Then, for each historical photovoltaic panel image, the trained model G(I';θ * ) Predict its useful life According to the predicted years, the power generation efficiency reduction rate database is queried to obtain the power generation efficiency reduction rate r of each photovoltaic panel. k .

[0143] Let database be function but: Then, calculate the arithmetic mean of the power generation efficiency reduction rate of all photovoltaic panels in the field As the average efficiency drop rate for the whole game: Where: Z is the total number of photovoltaic panels. Finally, the average efficiency drop rate It can indicate the average aging degree of all photovoltaic panels in the field.

[0144] Step B1: First, obtain the weather forecast data for the forecast day, including the forecast values ​​of temperature T, relative humidity H, wind speed WS, wind direction WD, sunshine hours SD, etc. every 15 minutes. These data can be obtained from the output of the numerical forecast model.

[0145] Assume that there are a total of time periods in the forecast day, and the environmental parameter of the jth time period is X' j =(T j , H j , WS j , W.D. j , S.D. j ) T .

[0146] According to the time t of the prediction day j , longitude and latitude and other parameters, using astronomical algorithms to calculate the solar altitude angle θ for each period j :θ j =f(t j , lon, lat), then, according to the sun position θ in each period j , referring to the position and physical parameters of the photovoltaic panels, physical models such as HDKR are used to calculate the light intensity E' for each photovoltaic panel in each period k，j : E' k，j =f'(θ j , β k , γ k , ...; X' j ).

[0147] Then, the theoretical maximum output power P' of each photovoltaic panel in each period can be calculated based on the light intensity. k，j :P' k，j =E' k，j S k η k ; After that, the theoretical maximum output power of all photovoltaic panels in each period is summarized to obtain the theoretical maximum possible output power P' for each period of the entire forecast dayj : Where Z is the total number of photovoltaic panels; then, the power prediction curve P'=[P' 1 , P' 2 , ..., P' 96 ] reflects the maximum possible output power for each period of the predicted day without considering system losses in theory.

[0148] Step B2: Based on the meteorological parameter X' of the forecast day, calculate its difference with each typical meteorological day c in history. i The meteorological similarity s i :s i =Sim(X',c i ) ; Then, determine one or several typical meteorological days with the most similar meteorological conditions to the forecast day i* ,in: Similarity can be determined by ranking and setting a threshold to determine the most similar c i* ; Extract the power generation deviation vector D of the most similar typical meteorological day i* =[ΔP 1 , ..., ΔP 96 ], where ΔP i =P' i -P i ; Calculate the average deviation of the vector The overall power generation deviation correction amount. Obtain the predicted daily theoretical power generation time series P' and uniformly calculate it according to the correction amount Perform an offset to obtain a preliminary revised predicted power generation sequence P″; Afterwards, the average photovoltaic panel efficiency drop rate of the entire field was obtained. According to its value, the time series P″ after the initial correction is corrected for the second time: Finally, after the above two corrections, the final predicted power time series P″′ combining the actual meteorological deviation and system loss can be obtained.

[0149] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for accurately predicting distributed photovoltaic output power, characterized in that: It includes the meteorological environment impact deviation detection step and the photovoltaic output power prediction step. The meteorological environment impact deviation detection step obtains several typical meteorological days according to the impact of the meteorological environment on the photovoltaic panel, and obtains the power generation deviation vector corresponding to the typical meteorological day according to the theoretical output power and the actual output power of the typical meteorological day; The photovoltaic output power prediction step calculates the predicted theoretical output power according to the meteorological environment of the prediction day and finds the typical meteorological day with the highest similarity according to the meteorological environment of the prediction day, and corrects the predicted theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity to obtain the photovoltaic output power of the prediction day.

2. The distributed photovoltaic output power accurate prediction method according to claim 1, characterized in that: The meteorological environment impact deviation detection step comprises the following steps: Step A1: obtaining multiple planar images of each photovoltaic station of the distributed photovoltaic system and obtaining historical meteorological data for one consecutive year; Step A2: Establish the three-dimensional spatial position of each photovoltaic panel based on multiple planar images. Several typical meteorological days are formed based on the historical meteorological data of one consecutive year; Step A3: Calculate the theoretical output power of each typical meteorological day according to the three-dimensional spatial position and physical parameters of each photovoltaic panel; Step A4: Obtain the actual output power of each typical meteorological day, and compare the actual output power of each typical meteorological day with the theoretical output power to obtain the power generation deviation vector corresponding to each typical meteorological day.

3. The distributed photovoltaic output power accurate prediction method according to claim 2, characterized in that: The photovoltaic output power prediction step comprises the following steps: Step B1: Obtain the meteorological forecast data for the forecast day, and calculate the predicted theoretical output power according to the three-dimensional spatial position and physical parameters of each photovoltaic panel; Step B2: According to the meteorological forecast data of the forecast day, obtain the typical meteorological day with the highest similarity to step 2, and use the power generation deviation vector of the typical meteorological day with the highest similarity to correct the predicted theoretical output power of the forecast day to obtain the photovoltaic output power of the forecast day.

4. The distributed photovoltaic output power accurate prediction method according to claim 3, characterized in that: The step A3 calculates the theoretical output power of each typical meteorological day according to the three-dimensional spatial posture and physical parameters of each photovoltaic panel, and specifically includes the following steps: A31: Divide each typical meteorological day into a period according to the same time to obtain several periods; A32: Obtain the illuminance of a given period of time according to the three-dimensional spatial position and physical parameters of each photovoltaic panel and the position of the sun in that period of time; A33: Obtain the theoretical output power for each period of a typical meteorological day based on the illuminance.

5. The distributed photovoltaic output power accurate prediction method according to claim 4, characterized in that: The step A4 obtains the actual output power of each typical meteorological day, and obtains the power generation deviation vector corresponding to each typical meteorological day by comparing the actual output power of each typical meteorological day with the theoretical output power, which specifically includes the following steps: A41: Get the actual output power in each period of a typical meteorological day; A42: Compare the actual output power with the theoretical output power in each period of each typical meteorological day to obtain the power generation deviation value in each period of each typical meteorological day; A43: Traverse the power generation deviation values ​​of all time periods to obtain the power generation deviation vector corresponding to each typical meteorological day.

6. The distributed photovoltaic output power accurate prediction method according to claim 5, characterized in that: Step B2: According to the meteorological forecast data of the forecast day, obtain the typical meteorological day with the highest similarity with step 2, and use the power generation deviation vector of the typical meteorological day with the highest similarity to correct the predicted theoretical output power of the forecast day to obtain the photovoltaic output power of the forecast day, which specifically includes the following steps: B21: Based on the meteorological forecast data of the forecast day, calculate the similarity of several typical meteorological days formed with the historical meteorological data, and determine the typical meteorological day with the highest similarity; B22: extract the power generation deviation vector of the typical meteorological day with the highest similarity, calculate the average value of the vector, and obtain the power generation deviation correction; B23: The predicted theoretical output power is corrected by the power generation deviation correction to obtain the photovoltaic output power on the predicted day.

7. The distributed photovoltaic output power accurate prediction method according to any one of claims 1 to 6, characterized in that: The method further includes a self-loss deviation detection step, wherein the self-loss deviation detection step calculates the power generation efficiency reduction rate of the corresponding photovoltaic panel and the average power generation efficiency reduction rate of the entire distributed photovoltaic system according to the image of the photovoltaic panel and a pre-set photovoltaic panel power generation efficiency reduction rate-surface similarity database; The photovoltaic output power prediction step corrects the predicted theoretical output power according to the power generation deviation vector of the typical meteorological day with the highest similarity, and then corrects it again according to the average power generation efficiency reduction rate of the entire distributed photovoltaic system to obtain the photovoltaic output power of the predicted day.

8. The method for accurately predicting distributed photovoltaic output power according to claim 7, characterized in that: The self-loss deviation detection step comprises the following steps: Step C1: obtaining a high-definition image of each photovoltaic panel of the distributed photovoltaic system taken orthogonally; Step C2: performing similarity analysis based on the high-definition image taken orthogonally to each photovoltaic panel and the high-definition image taken orthogonally to the newly installed photovoltaic panel, and obtaining a similarity used to describe the surface of the photovoltaic panel and recording it as surface similarity; Step C3: using the pre-set photovoltaic panel power generation efficiency reduction rate-surface similarity database, calculate the power generation efficiency reduction rate of the corresponding photovoltaic panel and the average power generation efficiency reduction rate of the entire distributed photovoltaic system.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for accurately predicting the output power of distributed photovoltaic power as described in any one of claims 1 to 8.

10. A distributed photovoltaic output power accurate prediction system, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.