Ultra-short-term power prediction method under photovoltaic panel shading state based on drone

Abstract

The present invention relates to the field of new energy technology, and discloses an ultra-short-term power prediction method under a photovoltaic panel shielding state based on an unmanned aerial vehicle, by obtaining a measured image sequence and predicted meteorological data under a snowmelt event in a target photovoltaic power station; wherein the measured image sequence is collected by the unmanned aerial vehicle at a preset frequency, including measured image data for representing the snowmelt conditions of all photovoltaic panels arranged in the target photovoltaic field station; shielding area extraction is performed on the measured image sequence to obtain a measured shielding area sequence; time series dependency extraction is performed on the measured shielding area sequence to obtain measured shielding area change information; the measured image sequence and the measured shielding area change information are input into a first photovoltaic ultra-short-term power prediction model to perform power prediction, and an ultra-short-term power prediction result is obtained, thereby solving the problem of inaccurate power prediction caused by environmental influences on photovoltaic panels.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and in particular to an ultra-short-term power prediction method under a photovoltaic panel shielding state based on an unmanned aerial vehicle. Background Art

[0002] Photovoltaic panels, also known as solar panels, are devices that use solar energy to convert light energy into electrical energy. They are mainly composed of solar cells, which convert sunlight radiation into electrons through the photovoltaic effect, which in turn form electric current.

[0003] However, photovoltaic panels are easily affected by environmental factors during use; for example, in a low-temperature environment with snowfall, photovoltaic panels may be covered with snow, which will hinder sunlight from passing through the snow layer. The snow layer will increase scattering losses, affect power generation efficiency, and lead to inaccurate power prediction.

[0004] Therefore, there is an urgent need to propose an ultra-short-term power prediction method based on drones under photovoltaic panel shading conditions to solve the problem of inaccurate power prediction caused by the influence of the environment on photovoltaic panels in related technologies. Summary of the invention

[0005] In view of this, the present invention provides an ultra-short-term power prediction method under the shading state of photovoltaic panels based on drones, so as to solve the problem of inaccurate power prediction caused by the influence of the environment on photovoltaic panels in the related art.

[0006] In a first aspect, the present invention provides an ultra-short-term power prediction method under a photovoltaic panel obstruction state based on an unmanned aerial vehicle, and the ultra-short-term power prediction method under a photovoltaic panel obstruction state based on an unmanned aerial vehicle comprises: obtaining a measured image sequence and predicted meteorological data under a snowmelt event in a target photovoltaic power station; wherein the measured image sequence is collected by an unmanned aerial vehicle at a preset frequency, including measured image data for representing the snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station; performing shielding area extraction on the measured image sequence to obtain a measured shielding area sequence; performing temporal dependency extraction on the measured shielding area sequence to obtain measured shielding area change information; inputting the measured image sequence and the measured shielding area change information into a first photovoltaic ultra-short-term power prediction model for power prediction to obtain an ultra-short-term power prediction result; wherein the first photovoltaic ultra-short-term power prediction model constructs a training data set from historical shielding area change information and historical power change information in each ultra-short-term prediction period under a snowmelt event, and inputs the training data set into a pre-constructed preset power prediction model for model training, and during the training process, learns the historical shielding area change information and historical power change information.

[0007] As an exemplary embodiment, the photovoltaic ultra-short-term power prediction model includes at least two photovoltaic ultra-short-term power prediction sub-models, and the ultra-short-term power prediction method based on the unmanned aerial vehicle under the shading state of the photovoltaic panel also includes: clustering the measured shading area sequence according to the area change rate and each clustering cluster in the first historical data set to obtain a clustering result; based on the clustering result, selecting the photovoltaic ultra-short-term power prediction sub-model corresponding to the clustering cluster with a clustering degree greater than a preset degree; inputting the measured image sequence and the measured shading area change information into the photovoltaic ultra-short-term power prediction sub-model to obtain the ultra-short-term power prediction result.

[0008] As an exemplary embodiment, the photovoltaic ultra-short-term power prediction model also includes a correction sub-model, which is used to output correction parameters of the ultra-short-term power prediction results. The ultra-short-term power prediction method based on the photovoltaic panel shading state of the drone also includes: extracting each measured shading area value in the measured shading area sequence; inputting the measured shading area value into the correction sub-model to obtain the correction parameter.

[0009] As an exemplary embodiment, the ultra-short-term power prediction method for photovoltaic panels under shading conditions based on drones also includes: extracting shading thickness from measured image data of a measured image sequence to obtain a first measured shading thickness sequence; extracting time-series dependencies from the first measured shading thickness sequence to obtain first measured shading thickness change information; inputting the first measured shading thickness change information and the measured shading area change information into a second photovoltaic ultra-short-term power prediction model to obtain a photovoltaic ultra-short-term power prediction result; wherein the second photovoltaic ultra-short-term power prediction model is trained using a training set consisting of historical shading thickness change information, historical shading area change information, and historical power change information.

[0010] As an exemplary embodiment, the ultra-short-term power prediction method for photovoltaic panels under shading conditions based on drones also includes: extracting shading thickness from the measured image data of the measured image sequence to obtain a second measured shading thickness sequence; inputting the second measured shading thickness sequence into a third photovoltaic ultra-short-term power prediction model to obtain a photovoltaic ultra-short-term power prediction result; wherein the third photovoltaic ultra-short-term power prediction model is trained by a second historical data set corresponding to multiple historical snowfall events of the target photovoltaic power station, and the second historical data set includes multiple historical image sequences collected by drones at a preset frequency and their corresponding historical power sequences.

[0011] In the second aspect, the present invention provides a method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones. The method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones includes: obtaining a first historical data set corresponding to multiple historical snowmelt events of a target photovoltaic power station, wherein the first historical data set includes multiple historical image sequences and their corresponding historical power sequences collected by drones at a preset frequency in an ultra-short-term prediction period; the historical image sequence includes multiple historical image data for representing the snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station; the shading area of ​​the historical image sequence under each ultra-short-term prediction period is extracted to obtain a historical shading area sequence; the time series dependency relationship of the historical shading area sequence and the historical power sequence is extracted to obtain historical shading area change information and historical power change information, respectively; a training data set is constructed based on the historical shading area change information and the historical power change information under each ultra-short-term prediction period, and the training data set is input into a pre-constructed preset power prediction model for model training. During the training process, the time series correspondence between the historical shading area change information and the historical power change information is learned to obtain the first photovoltaic ultra-short-term power prediction model.

[0012] As an exemplary embodiment, the photovoltaic ultra-short-term power prediction model includes at least two photovoltaic ultra-short-term power prediction sub-models. The method for constructing an ultra-short-term power prediction model based on the shading state of the photovoltaic panels of the drone also includes: clustering the historical shading area sequences in each ultra-short-term prediction period according to the area change rate to obtain at least two cluster clusters; extracting the time series dependency relationship of each cluster cluster and the historical power sequence corresponding to each cluster cluster to obtain a training data set corresponding to each cluster cluster; wherein the training data set includes the historical shading area change information and the historical power change information corresponding to each cluster cluster; inputting the training data set into a pre-constructed preset power prediction model for model training to obtain a photovoltaic ultra-short-term power prediction sub-model corresponding to each cluster cluster.

[0013] As an exemplary embodiment, the photovoltaic ultra-short-term power prediction model also includes a correction sub-model, which is used to output correction parameters of the ultra-short-term power prediction results. The method for constructing an ultra-short-term power prediction model based on the shading state of the photovoltaic panels of the drone includes: extracting each historical shading area value in the historical shading area sequence and the historical power data corresponding to the historical shading area value in time series as the first training set; inputting the first training set into the correction sub-model for model training, and during the training process, learning the correspondence between the historical shading area value, the historical power data and the correction parameters.

[0014] As an exemplary embodiment, the method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones also includes: extracting the shading thickness of historical image sequences under each ultra-short-term prediction period to obtain a first historical shading thickness sequence; extracting the temporal dependency of the first historical shading thickness sequence to obtain first historical shading thickness change information; adding the first historical shading thickness change information to the training data set, and inputting the added new training data set into the power prediction model for model training. During the training process, the temporal correspondence between the historical shading area change information, the first historical shading thickness change information and the historical power change information is learned to obtain a second photovoltaic ultra-short-term power prediction model.

[0015] As an exemplary embodiment, a method for constructing an ultra-short-term power prediction model under shading conditions of photovoltaic panels based on drones also includes: obtaining a second historical data set corresponding to multiple historical snowfall events of the target photovoltaic power station; wherein the second historical data set includes multiple historical image sequences collected by drones at a preset frequency and their corresponding historical power sequences; the historical image sequence includes multiple historical image data for representing all photovoltaic panels installed in the target photovoltaic station; performing shading thickness extraction on the historical image sequence to obtain a second historical shading thickness sequence; performing temporal dependency extraction on the second historical shading thickness sequence and the historical power sequence to obtain second historical shading thickness change information and historical power change information, respectively; constructing a second training data set based on the second historical shading thickness change information and the historical power change information, and inputting the second training data set into a pre-constructed preset power prediction model for model training. During the training process, the temporal correspondence between the second historical shading thickness change information and the historical power change information is learned to obtain a third photovoltaic ultra-short-term power prediction model.

[0016] The present invention provides a method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones, and obtains a first historical data set corresponding to multiple historical snowmelt events of a target photovoltaic power station, wherein the first historical data set includes multiple historical image sequences and their corresponding historical power sequences collected by drones at a preset frequency in an ultra-short-term prediction period; the historical image sequence includes multiple historical image data for representing the snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station; the shading area of ​​the historical image sequence in each ultra-short-term prediction period is extracted to obtain a historical shading area sequence; the time series dependency relationship of the historical shading area sequence and the historical power sequence is extracted to obtain historical shading area change information and historical power change information, respectively; a training data set is constructed based on the historical shading area change information and the historical power change information in each ultra-short-term prediction period, and the training data set is input into a pre-constructed Model training is performed in a preset power prediction model, and during the training process, the time-series correspondence between historical shading area change information and historical power change information is learned to obtain a first photovoltaic ultra-short-term power prediction model; in the above-mentioned implementation mode of this embodiment, the historical shading area change information can reflect the time-series changes of the area of ​​the photovoltaic panel within the ultra-short-term prediction period, and the historical power change information can reflect the time-series changes of the power of the photovoltaic panel within the ultra-short-term prediction period. By adopting the above-mentioned model training method, the parameters of the preset power prediction model can be continuously adjusted during the training process, so that the obtained first photovoltaic ultra-short-term power prediction model can adapt to the accompanying relationship between area change and power change in different time intervals of the ultra-short-term prediction period, and can consider the power change caused by the blocking of sunlight by the snow layer and the increased scattering loss of the snow layer, that is, the photovoltaic ultra-short-term power prediction can be performed according to the actual snow coverage of the photovoltaic panel. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 is a flow chart of an ultra-short-term power prediction method based on a photovoltaic panel shielding state of an unmanned aerial vehicle according to an embodiment of the present invention;

[0019] Figure 2 The present invention is a flowchart of a method for constructing an ultra-short-term power prediction model under photovoltaic panel shading conditions based on a drone according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] Photovoltaic panels are easily affected by environmental factors during use; for example, in a low-temperature environment with snowfall, the photovoltaic panels may be covered with snow, which will hinder the sunlight from penetrating the snow layer. The snow layer will increase scattering losses, affect power generation efficiency, and thus lead to inaccurate power prediction; specifically, in the present invention, the period from the beginning of snowmelt to the end of snowmelt is recorded as a snowmelt event; in a snowmelt event, as the temperature rises, the snow surface begins to melt, and the snowmelt event begins at this time; the thickness of the snow covering the surface of the photovoltaic panel gradually decreases over time, and the power generation power of the photovoltaic panel is always zero. At the first preset moment, the snow covering the surface of the photovoltaic panel is of a preset thickness, but at this time the light can penetrate the snow layer When the radiation is directed to the surface of the photovoltaic panel, the photovoltaic panel begins to generate power. During the preset time interval corresponding to the first preset moment of the snowmelt event to the second preset moment when the snowmelt ends, the snow surface of the photovoltaic panel melts from a preset thickness to when the photovoltaic panel is completely exposed. The thickness of the snow covering the surface of the photovoltaic panel changes from a preset thickness to zero, and the area of ​​the snow changes from a preset area to zero. At this time, the power generation power of the photovoltaic panel gradually increases and returns to normal when the snow surface is completely melted. However, during this time interval, due to the blocking of sunlight by the snow layer and the increased scattering loss of the snow layer, the power prediction of the photovoltaic panel is inaccurate, and the output power of the photovoltaic panel is greatly affected by the area blocked by the snow.

[0022] Therefore, within the preset time interval of the snowmelt event, when the photovoltaic panels are in power, the direct irradiance and ultra-short-term power prediction are inaccurate due to the blocking of sunlight by the snow layer and the increased scattering loss of the snow layer. In addition, it usually takes several days from the beginning to the end of snowmelt, and the snowmelt rate is different in different time periods. For example, on the same day, the snowmelt rates in the morning, noon and afternoon are not uniform, and in a continuous period, the snowmelt rates in the first 15 minutes and the adjacent last 15 minutes are also uneven. The time dimension of photovoltaic ultra-short-term power prediction is usually 15 minutes. If this time dimension is used for power prediction, the amount of snow changes unevenly during the prediction period, and the power prediction is inaccurate.

[0023] Based on this, according to an embodiment of the present invention, a method for ultra-short-term power prediction under photovoltaic panel shading state based on a drone is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that here.

[0024] Figure 1 It is shown that this embodiment provides an ultra-short-term power prediction method based on a photovoltaic panel under a shading state based on a drone. The prediction method is based on Figure 2 The ultra-short-term power prediction model constructed by the ultra-short-term power prediction model based on the photovoltaic panel shading state of the drone is used for ultra-short-term power prediction.

[0025] In such Figure 1 As shown, the process includes the following steps:

[0026] Step S101, obtaining a measured image sequence and predicted meteorological data under a snowmelt event of a target photovoltaic power station; wherein the measured image sequence is collected by a drone at a preset frequency, including measured image data for representing the snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station.

[0027] In this embodiment, the measured image sequence is obtained by collecting the drone according to a preset collection frequency; specifically, the drone is controlled to conduct an inspection every 5 minutes according to a preset inspection path, and each inspection lasts 2-3 minutes, so that multiple groups of image data obtained by multiple inspections are used as the measured image sequence; the predicted meteorological data can be determined through the weather forecast.

[0028] Step S102: extract the occlusion area of ​​the measured image sequence to obtain a measured occlusion area sequence.

[0029] In this embodiment, the occlusion area of ​​the measured image sequence can be extracted by image recognition to obtain a measured occlusion area sequence; the specific implementation method can refer to the content recorded in the above embodiment, which will not be repeated here.

[0030] Step S103: extracting the time-series dependency of the measured occlusion area sequence to obtain the measured occlusion area change information.

[0031] In order to determine the temporal change of the shading area of ​​the photovoltaic panel corresponding to the current measured image sequence, in this embodiment, the temporal dependency extraction of the measured shading area sequence is performed; as a possible implementation method, the measured shading area sequence can be input into GRU or LSTM, so as to obtain the measured shading area change information by performing temporal dependency extraction on the measured shading area sequence.

[0032] Step S104, input the measured image sequence and the measured shading area change information into the first photovoltaic ultra-short-term power prediction model for power prediction, and obtain the ultra-short-term power prediction result; wherein, the first photovoltaic ultra-short-term power prediction model constructs a training data set from the historical shading area change information and the historical power change information in each ultra-short-term prediction period under the snowmelt event, and inputs the training data set into the pre-constructed preset power prediction model for model training, and in the process of training, learns the historical shading area change information and the historical power change information.

[0033] In this embodiment, a photovoltaic ultra-short-term power prediction model is constructed by constructing a training data set based on historical shading area change information and historical power change information under each ultra-short-term prediction period, and the training data set is input into a pre-constructed preset power prediction model for model training, and the temporal correspondence between historical image sequences, historical shading area change information and historical power change information is learned during the training process; the photovoltaic ultra-short-term power prediction model obtained by this training method can adapt to the accompanying relationship between area change and power change under snowmelt events, therefore, by inputting the measured shading area change information and predicted meteorological data into the photovoltaic ultra-short-term power prediction model, the accompanying relationship between the current measured shading area change information and power change can be considered to perform photovoltaic power prediction, and then the photovoltaic power prediction result suitable for the time interval of the current ultra-short-term prediction period can be obtained, which can solve the problem of inaccurate prediction when using traditional photovoltaic ultra-short-term power prediction models for power prediction under snowfall events.

[0034] Different snowmelt rates during snowmelt events will result in different shading areas of photovoltaic panels per unit time, which in turn manifests as different output powers of photovoltaic panels. Based on this, in order to distinguish the impact of different snow accumulation rates on power, as an exemplary embodiment, an ultra-short-term power prediction method for photovoltaic panels under shading conditions based on drones also includes: clustering the measured shading area sequence according to the area change rate and each clustering cluster in the first historical data set to obtain a clustering result; based on the clustering result, selecting a photovoltaic ultra-short-term power prediction sub-model corresponding to a clustering cluster with a clustering degree greater than a preset degree; inputting the measured image sequence and the measured shading area change information into the photovoltaic ultra-short-term power prediction sub-model to obtain an ultra-short-term power prediction result.

[0035] Specifically, the K clustering method can be used to select the photovoltaic ultra-short-term power prediction sub-model corresponding to the cluster cluster whose similarity with the current measured shading area is greater than the preset degree for power prediction, so that the influence of different snow accumulation rates on the shading thickness of the photovoltaic panel per unit time can be considered in the prediction.

[0036] In order to consider the influence of snow thickness on the ultra-short-term power prediction results from the static feature dimension, so that the ultra-short-term power prediction results can be corrected according to the static feature during prediction, as an exemplary embodiment, the photovoltaic ultra-short-term power prediction model also includes a correction sub-model, which is used to output correction parameters of the ultra-short-term power prediction results. The method for constructing an ultra-short-term power prediction model based on the shading state of the photovoltaic panels of the drone includes: extracting each measured shading area value in the measured shading area sequence; inputting the measured shading area value into the correction sub-model to obtain the correction parameter.

[0037] In this embodiment, the correction sub-model is used to output the correction parameter of the ultra-short-term power prediction result, and the correction parameter can be a correction coefficient, a correction value, or a combination of a correction coefficient and a correction value. Exemplarily, when the correction parameter is a correction coefficient, the correction coefficient is multiplied by the ultra-short-term power prediction result; when the correction parameter is a correction parameter, the correction coefficient is added to the ultra-short-term power prediction result; when the correction parameter is a combination of a correction coefficient and a correction value, the correction coefficient is multiplied by the correction value and then added to the ultra-short-term power prediction result.

[0038] During a snowmelt event, the snow covering the surface of the photovoltaic panels still has a certain thickness, and the power generation is also affected by the covering thickness of the photovoltaic panels. In order to take the above-mentioned influence into consideration, as an exemplary embodiment, the ultra-short-term power prediction method under the shading state of the photovoltaic panels based on the drone also includes: extracting the shading thickness of the measured image data of the measured image sequence to obtain a first measured shading thickness sequence; extracting the time-series dependency relationship of the first measured shading thickness sequence to obtain the first measured shading thickness change information; inputting the first measured shading thickness change information and the measured shading area change information into the second photovoltaic ultra-short-term power prediction model to obtain the photovoltaic ultra-short-term power prediction result; wherein the second photovoltaic ultra-short-term power prediction model is trained by a training set consisting of historical shading thickness change information, historical shading area change information, and historical power change information.

[0039] In this embodiment, deep learning, machine learning and other methods can be used to extract image features representing occlusion thickness information from each group of historical image sequences; specifically, please refer to the contents in the above embodiments, which will not be repeated in this embodiment.

[0040] In order to determine the impact of the temporal change in the shading thickness of the photovoltaic panels during a snowmelt event on the predicted power data, in this embodiment, the time series dependency extraction is performed on the first measured shading thickness sequence to obtain the first measured shading thickness change information; as a possible implementation method, the historical shading thickness sequence can be input into the GRU or LSTM to obtain the first measured shading thickness change information.

[0041] After obtaining the measured shielding area change information, the measured shielding area change information and the first measured shielding thickness change information are input into the second photovoltaic ultra-short-term power prediction model to obtain the photovoltaic ultra-short-term power prediction result; wherein the second photovoltaic ultra-short-term power prediction model is trained with a training set consisting of the first historical shielding thickness change information, the historical shielding area change information, and the historical power change information.

[0042] In the above-mentioned implementation mode of the present application, the measured shading area change information can reflect the time-series change of the shading area, and the first measured shading thickness change information can reflect the time-series change of the thickness of the photovoltaic panel. Therefore, by adopting the above-mentioned power prediction method, the accompanying relationship between thickness change and area change and power change can be considered in the power prediction process to solve the problem of inaccurate prediction when using the traditional photovoltaic ultra-short-term power prediction model for power prediction under snowmelt events.

[0043] As an exemplary embodiment, the ultra-short-term power prediction method under the shading state of photovoltaic panels based on drones also includes: extracting the shading thickness of the measured image data of the measured image sequence to obtain a second measured shading thickness sequence; inputting the second measured shading thickness sequence into a third photovoltaic ultra-short-term power prediction model to obtain a photovoltaic ultra-short-term power prediction result; wherein the third photovoltaic ultra-short-term power prediction model is trained by a second historical data set corresponding to multiple historical snowfall events of the target photovoltaic power station, and the second historical data set includes multiple historical image sequences collected by drones at a preset frequency and their corresponding historical power sequences.

[0044] Exemplarily, after obtaining the corresponding photovoltaic ultra-short term power prediction results using the first, second and third photovoltaic ultra-short term power prediction models, the photovoltaic ultra-short term power prediction results obtained by the first, second and third photovoltaic ultra-short term power prediction models can be weighted and summed using a weighting method to obtain the final photovoltaic ultra-short term power prediction result; wherein the weight corresponding to each result can be determined by the degree of similarity between the second measured shading thickness sequence and the data set for training each model, and the weight is positively correlated with the degree of similarity.

[0045] See also Figure 2 As shown in the figure, the method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones includes the following steps:

[0046] Step S201, obtaining a first historical data set corresponding to multiple historical snowmelt events of the target photovoltaic power station, wherein the first historical data set includes multiple historical image sequences and their corresponding historical power sequences collected by drones at a preset frequency in an ultra-short-term prediction period; the historical image sequences include multiple historical image data for representing the snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station.

[0047] In the present invention, the drone inspects the target photovoltaic power station according to a preset frequency and a preset inspection path; wherein the preset frequency is determined according to the area of ​​the target photovoltaic power station and the preset inspection path, and the area and the preset inspection path can be used to perform three-dimensional modeling and image recognition of the target photovoltaic power station in advance based on GIS technology and remote sensing technology, and the preset frequency is less than the ultra-short-term prediction period; specifically, the drone is controlled to inspect once every 5 minutes according to the preset inspection path, and each inspection lasts 2-3 minutes.

[0048] The amount of data collected by the drone is huge, and there are many noise points or other interferences caused by other factors in the image, and the single image data collected by the drone cannot reflect the snow-covered conditions of all photovoltaic panels in the target photovoltaic station; therefore, after the drone collects the image data, the image data is preprocessed and image recognized to obtain image data that can represent the snow-covered conditions of all photovoltaic panels set up in the target photovoltaic station; specifically. For noise, the non-local means filtering algorithm (NL-Means) can be used to denoise the entire image, and the weighted average of the corresponding areas of the detected image is performed by using the similarity of the image blocks and the neighborhood pixel values, and the Gaussian noise in the image will be effectively removed; and for problems such as image blur, the image needs to be enhanced, and the multi-scale image pyramid enhancement technology can be used to suppress noise while amplifying the image residual details at different scales, and then accumulating and superimposing layer by layer, so as to achieve the purpose of image enhancement; third, since the image shooting area not only contains photovoltaic equipment, but also some irrelevant areas around it. Therefore, it is necessary to segment the image and extract the relevant areas from it. For intelligent segmentation, the regionality of infrared thermal imaging and straight line detection can be combined through the Line Segment Detector (LSD) algorithm and the Otsu algorithm to segment the real photovoltaic module area and filter out irrelevant targets.

[0049] Furthermore, based on the image data obtained after the above image preprocessing, deep learning, machine learning and other algorithms are used for image recognition to obtain the snow coverage of all photovoltaic panels in the target photovoltaic station; among them, the machine learning model may include support vector machine, feature model, etc. The deep learning model can use the convolutional neural network model (CNN) to realize image recognition. CNN is not only faster, but they can also detect multiple instances of an object from an image, even if the image is slightly distorted, stretched or changed in some other form. In deep image recognition, convolutional neural networks are even better than humans in tasks such as classifying objects into fine-grained categories. In addition, the most popular deep learning models such as YOLO, SSD and RCNN also use convolutional layers to parse images or photos to obtain the snow coverage of all photovoltaic panels in the target photovoltaic station.

[0050] Furthermore, the start image data and the end image data of the snowmelt event are determined based on the snow cover condition of the photovoltaic panels, and the image data of the snowmelt event are intercepted with the start image data as the starting point and the ultra-short-term prediction period as the sequence time length of the historical image sequence to obtain the historical image sequence; wherein, the interception method can adopt a rolling window or a sliding window, and the sliding step size can be preset according to needs.

[0051] After obtaining the historical image sequence, the historical power data corresponding to each historical image data time series in the historical image sequence is obtained as the historical power sequence, and a first historical data set corresponding to multiple historical snowmelt events of the target photovoltaic power station is constructed based on the historical image sequence and the historical power sequence.

[0052] Step S202 , extracting the occlusion area of ​​the historical image sequence in each ultra-short-term prediction period to obtain a historical occlusion area sequence.

[0053] In this embodiment, after obtaining the first historical data set, the occlusion area of ​​the historical image sequence in each ultra-short-term prediction period can be extracted by image recognition to obtain a historical occlusion area sequence.

[0054] Exemplarily, deep learning, machine learning and other methods can be used to extract image features representing thickness information from each historical image data. Before extracting image features representing thickness information from historical image data, image preprocessing including image denoising, image enhancement, and removal of peripheral irrelevant areas is performed on the historical image data.

[0055] In this embodiment, a convolutional neural network is used to extract features from the preprocessed image. Specifically, the preprocessed image data is input into a convolutional neural network (CNN) to extract feature representation of the image. In the CNN, the image is converted into a vector form. The convolution layer slides on the image through a filter (also called a convolution kernel) to extract features from the image. Different filters can capture different features such as edges, textures, and colors, and sublimate the image from the pixel level into a feature map with semantic information.

[0056] Specifically, CNN decomposes the color image input into three channels: R (red), G (green), and B (blue), each of which has a value between 0 and 255.

[0057] Among them, because ordinary neural networks use a fully connected method for input and hidden layer to extract features, when processing images, slightly larger images will result in huge calculations and become very slow. Therefore, convolution operations are used to extract features.

[0058] Among them, the CNN convolutional neural network is activated using the linear rectification function (Rectified Linear Unit, ReLU). Unilateral inhibition is used to convert all negative values ​​to 0, while positive values ​​remain unchanged. This makes the neurons in the neural network also have sparse activation. The model after sparseness achieved by ReLU can better mine relevant features and fit the training data.

[0059] Furthermore, the target input is pooled to extract features and reduce the amount of data passed to the next stage.

[0060] Furthermore, full connection is performed based on the obtained results, which plays the role of a "classifier" in the convolutional neural network to obtain the occlusion area corresponding to each image, and then obtain the historical occlusion area sequence.

[0061] Step S203, extracting the time series dependency relationship of the historical shielding area sequence and the historical power sequence to obtain historical shielding area change information and historical power change information respectively.

[0062] In order to determine the impact of the temporal change in the shading area of ​​the photovoltaic panels caused by snowfall in each snowfall event on the historical power data, in this embodiment, the time series dependency of the historical shading area sequence and the historical power sequence is extracted to obtain the historical shading area change information and the historical power change information, respectively.

[0063] As a possible implementation method, the historical occlusion area sequence and the historical power sequence can be input into a gated recurrent unit (GRU) or a long short-term memory network (LSTM) to extract the temporal dependency of the historical occlusion area sequence and the historical power sequence through GRU or LSTM to obtain the historical occlusion area change information and the historical power change information, respectively.

[0064] Step S204, constructing a training data set based on the historical shading area change information and the historical power change information under each ultra-short-term prediction period, and inputting the training data set into a pre-built preset power prediction model for model training. During the training process, the time series correspondence between the historical shading area change information and the historical power change information is learned to obtain the first photovoltaic ultra-short-term power prediction model.

[0065] In this embodiment, the input of the pre-constructed preset power prediction model may include multi-dimensional data; for example, the input of the preset power prediction model may include meteorological data, shading area data, etc., wherein the meteorological data may include temperature, humidity, wind speed, pressure, and irradiance, etc.; therefore, a training data set is constructed based on the historical shading area change information and the historical power change information under each ultra-short-term prediction period, and the training data set may include historical shading area change information, historical power change information and its corresponding meteorological data; after constructing the training data set, the training data set is input into the pre-constructed preset power prediction model for model training, and during the training process, the time series correspondence between the historical shading area change information and the historical power change information is learned to obtain the first photovoltaic ultra-short-term power prediction model.

[0066] It should be understood that in the above-mentioned implementation mode of this embodiment, the historical blocked area change information can reflect the time series change of the area of ​​the photovoltaic panel within the ultra-short-term prediction period, and the historical power change information can reflect the time series change of the power of the photovoltaic panel within the ultra-short-term prediction period. By adopting the above-mentioned model training method, the parameters of the preset power prediction model can be continuously adjusted during the training process, so that the first photovoltaic ultra-short-term power prediction model obtained can adapt to the accompanying relationship between area change and power change in different time intervals within the snowfall event, and can consider the power changes caused by the blocking of sunlight by the snow layer and the increased scattering loss of the snow layer, that is, it can perform photovoltaic ultra-short-term power prediction according to the actual snow coverage of the photovoltaic panel.

[0067] As an exemplary embodiment, the photovoltaic ultra-short-term power prediction model includes at least two photovoltaic ultra-short-term power prediction sub-models. The method for constructing an ultra-short-term power prediction model based on the shading state of the photovoltaic panels of the drone also includes: clustering the historical shading area sequences in each ultra-short-term prediction period according to the area change rate to obtain at least two cluster clusters; extracting the time dependency relationship of each cluster cluster and the historical power sequence corresponding to each cluster cluster to obtain a training data set corresponding to each cluster cluster; wherein the training data set includes the historical shading area change information and the historical power change information corresponding to each cluster cluster; inputting the training data set into a pre-constructed preset power prediction model for model training to obtain a photovoltaic ultra-short-term power prediction sub-model corresponding to each cluster cluster.

[0068] Different snowmelt rates during snowmelt events will result in different shading areas of photovoltaic panels per unit time, which in turn causes different output powers of photovoltaic panels. Based on this, in order to distinguish the impact of different snow accumulation rates on power, in the present invention, clusters are divided according to different snow accumulation rates to train different photovoltaic ultra-short-term power prediction sub-models for different cluster clusters.

[0069] In this embodiment, K clustering can be used to cluster the historical shading area sequences within each ultra-short-term prediction period according to the area change rate; wherein, the historical shading area sequences within each ultra-short-term prediction period can be divided and then clustered, and the historical shading area sequences between each ultra-short-term prediction period can also be divided and then clustered to obtain at least two cluster clusters; wherein each cluster cluster can reflect the historical shading area sequence and historical power sequence corresponding to different snow accumulation rates; further, GRU and LSTM can be used to extract the time series dependency relationship of each cluster cluster and the historical surface subsequence and historical power sequence corresponding to each cluster cluster, respectively, to obtain a training data set corresponding to each cluster cluster; wherein, the training data set includes the historical shading area change information and the historical power change information corresponding to each cluster cluster; further, the training data set is input into a pre-constructed preset power prediction model for model training to obtain a photovoltaic ultra-short-term power prediction sub-model corresponding to each cluster cluster, so that the influence of different snow accumulation rates on the shading area of ​​​​the photovoltaic panel per unit time can be considered during model training.

[0070] In order to consider the impact of snow thickness on ultra-short-term power prediction results from the static feature dimension, so that the trained model can correct the ultra-short-term power prediction results according to the static feature, as an exemplary embodiment, the photovoltaic ultra-short-term power prediction model also includes a correction sub-model, which is used to output correction parameters of the ultra-short-term power prediction results. The method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels of unmanned aerial vehicles includes: extracting each historical shading area value in the historical shading area sequence and the historical power data corresponding to the historical shading area value in time series as the first training set; inputting the first training set into the correction sub-model for model training, and during the training process, learning the correspondence between the historical shading area value, the historical power data and the correction parameters.

[0071] In this embodiment, the correction sub-model is used to output a correction parameter of the ultra-short-term power prediction result. The correction parameter may be a correction coefficient, a correction value, or a combination of a correction coefficient and a correction value.

[0072] Exemplarily, after obtaining the historical shading area sequence, each historical shading area value in the historical shading area sequence and the historical power data corresponding to the historical shading area value in time series are extracted as the first training set, and the first training set is input into the correction sub-model for model training. The correction sub-model may be a neural network model, which is used to map the mathematical relationship between the historical shading area value, the historical power data corresponding to the time series, and the correction parameter; specifically, the training process may be to input the first training set into the correction sub-model for model training, obtain the correction parameter output by the correction sub-model, perform mathematical operations on the correction parameter and the photovoltaic ultra-short-term power prediction result output by the trained first photovoltaic ultra-short-term power prediction model, and obtain the first corrected prediction result; in each round of iteration, the parameters of the correction sub-model are continuously adjusted to make the error between the first corrected prediction result and the historical power data corresponding to the time series of the first corrected prediction result smaller, until the error meets the preset conditions.

[0073] During a snowmelt event, the snow covering the surface of the photovoltaic panels still has a certain thickness, and the power generation is also affected by the covering thickness of the photovoltaic panels. In order to take the above-mentioned influence into consideration, as an exemplary embodiment, a method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on a drone also includes: extracting the shading thickness of historical image sequences under each ultra-short-term prediction period to obtain a first historical shading thickness sequence; extracting the temporal dependency relationship of the first historical shading thickness sequence to obtain first historical shading thickness change information; adding the first historical shading thickness change information to the training data set, and inputting the added new training data set into the power prediction model for model training. During the training process, the temporal correspondence between the historical shading area change information, the first historical shading thickness change information and the historical power change information is learned to obtain a second photovoltaic ultra-short-term power prediction model.

[0074] In this embodiment, deep learning, machine learning and other methods can be used to extract image features representing occlusion thickness information from each group of historical image sequences; CNN can be used to implement image recognition; for machines. In deep image recognition, convolutional neural networks are even better than humans in tasks such as classifying objects into fine-grained categories. In addition, deep learning models such as YOLO, SSD and RCNN also use convolutional layers to parse images or photos.

[0075] Exemplarily, a support vector machine, a feature model, etc. may be used to extract image features representing occlusion thickness information from each group of historical image sequences.

[0076] In order to determine the impact of the temporal change in the shading thickness of the photovoltaic panels during a snowmelt event on the historical power data, in this embodiment, the time dependency of the first historical shading thickness sequence is extracted to obtain the first historical shading thickness change information; as a possible implementation method, the first historical shading thickness sequence can be input into a GRU or LSTM to obtain the first historical shading thickness change information.

[0077] After obtaining the first historical occlusion thickness change information, the first historical occlusion thickness change information is added to the data set containing the first historical occlusion thickness change information or historical power change information corresponding to its time series to form a new training data set, and the added new training data set is input into the power prediction model for model training.

[0078] It should be understood that in the above-mentioned implementation mode of this embodiment, the first historical shading thickness change information can reflect the time-series change of the shading thickness of the photovoltaic panel during the snowmelt event, the first historical shading thickness change information can reflect the time-series change of the thickness of the photovoltaic panel, and the historical power change information can reflect the time-series change of the power of the photovoltaic panel. Therefore, by adopting the above-mentioned model training method, the parameters of the preset power prediction model can be continuously adjusted during the training process, so that the obtained second photovoltaic ultra-short-term power prediction model can adapt to the accompanying relationship between the change of snow layer thickness, the change of snow layer area and the change of power during the snowmelt event, so as to solve the problem of inaccurate prediction when using the traditional photovoltaic ultra-short-term power prediction model for power prediction during snowfall events.

[0079] In order to consider the impact of the snow thickness in the entire snowfall event on power prediction, as an exemplary embodiment, the method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones also includes: obtaining a second historical data set corresponding to multiple historical snowfall events of the target photovoltaic power station; wherein the second historical data set includes multiple historical image sequences collected by drones at a preset frequency and their corresponding historical power sequences; the historical image sequence includes multiple historical image data for representing all photovoltaic panels installed in the target photovoltaic station; extracting the shading thickness of the historical image sequence to obtain a second historical shading thickness sequence; extracting the temporal dependency relationship of the second historical shading thickness sequence and the historical power sequence to obtain the second historical shading thickness change information and the historical power change information, respectively; constructing a second training data set based on the second historical shading thickness change information and the historical power change information, and inputting the second training data set into a pre-constructed preset power prediction model for model training. During the training process, the temporal correspondence between the second historical shading thickness change information and the historical power change information is learned to obtain a third photovoltaic ultra-short-term power prediction model.

[0080] In this embodiment, the snowfall event is an event corresponding to the target photovoltaic power station from the start of snowfall to the completion of snow melting.

[0081] Exemplarily, the second historical data set includes multiple groups of historical image sequences and historical power sequences; wherein the historical image sequences are collected by a drone according to a preset collection frequency; exemplarily, the preset inspection route can pre-perform three-dimensional modeling and image recognition of the target photovoltaic power station based on GIS technology and remote sensing technology to determine the distribution of photovoltaic panels in the target photovoltaic power station, and further determine based on the distribution.

[0082] Exemplarily, the historical power sequence can be measured by the power generation power of the photovoltaic power station; wherein the time axis of the historical power sequence is the same as the historical image sequence.

[0083] After obtaining the historical image data, image recognition based on occlusion thickness can be performed on each historical image data in each group of historical image sequences by image recognition, so as to extract the occlusion thickness of the historical image sequence of each group of historical image sequences and obtain a second historical occlusion thickness sequence; the specific implementation method can refer to the contents recorded in the above embodiments and will not be repeated here.

[0084] In order to determine the impact of the temporal change in the shading thickness of the photovoltaic panels caused by snowfall in each snowfall event on the historical power data, in this embodiment, the second historical shading thickness sequence and the historical power sequence are extracted for timing dependencies to obtain the second historical shading thickness change information and the historical power change information, respectively; the specific implementation method can refer to the contents recorded in the above embodiments and will not be repeated here.

[0085] In this embodiment, the input of the pre-constructed preset power prediction model may include multi-dimensional data; for example, the input of the preset power prediction model may include meteorological data, shading thickness data, etc., wherein the meteorological data may include temperature, humidity, wind speed, pressure, and irradiance, etc.; therefore, a training data set is constructed based on the second historical shading thickness change information and the historical power change information, and the training data set may include the second historical shading thickness change information, the historical power change information and its corresponding meteorological data; after constructing the training data set, the training data set is input into the pre-constructed preset power prediction model for model training, and during the training process, the time series correspondence between the second historical shading thickness change information and the historical power change information is learned to obtain a third photovoltaic short-term power prediction model.

[0086] It should be understood that in the above-mentioned implementation mode of this embodiment, the second historical shading thickness change information can reflect the temporal changes in the thickness of the snow layer of the photovoltaic panel during the entire snowfall cycle, and the historical power change information can reflect the temporal changes in the output power of the photovoltaic panel during the entire snowfall cycle. By adopting the above-mentioned model training method, during the training process, the parameters of the preset power prediction model can be continuously adjusted so that the third photovoltaic short-term power prediction model can adapt to the accompanying relationship between thickness changes and power changes in different time intervals during the entire snowfall cycle, so as to solve the problem of inaccurate prediction when using traditional photovoltaic short-term power prediction models for power prediction during snowfall events.

[0087] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0088] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling one or more computer devices (which may be personal computers, servers or network devices, etc.) to execute all or part of the steps of the method in the above embodiments.

[0089] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0090] In the several embodiments provided in the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0091] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution provided in this embodiment.

[0092] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0093] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

[0094] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0095] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for ultra-short-term power prediction under photovoltaic panel shading state based on drone, characterized in that: include: Acquire measured image sequences and predicted meteorological data under snowmelt events of a target photovoltaic power station; wherein the measured image sequences are collected by a drone at a preset frequency, and include measured image data for representing snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station; Extracting the occlusion area of ​​the measured image sequence to obtain a measured occlusion area sequence; Extracting the time-series dependency of the measured occlusion area sequence to obtain the measured occlusion area change information; The measured image sequence and the measured shading area change information are input into the first photovoltaic ultra-short-term power prediction model for power prediction to obtain an ultra-short-term power prediction result; wherein, the first photovoltaic ultra-short-term power prediction model constructs a training data set from the historical shading area change information and the historical power change information in each ultra-short-term prediction period under a snowmelt event, and the training data set is input into a pre-constructed preset power prediction model for model training, and during the training process, the historical shading area change information and the historical power change information are learned.

2. The ultra-short-term power prediction method based on a UAV under a photovoltaic panel shading state as claimed in claim 1 is characterized in that: The photovoltaic ultra-short-term power prediction model includes at least two photovoltaic ultra-short-term power prediction sub-models, and the ultra-short-term power prediction method based on the photovoltaic panel shielding state of the drone also includes: Clustering the measured shielding area sequence with each clustering cluster in the first historical data set according to the area change rate to obtain a clustering result; Selecting, based on the clustering result, a photovoltaic ultra-short-term power prediction sub-model corresponding to the clustering cluster whose clustering degree is greater than a preset degree; The measured image sequence and the measured shielding area change information are input into the photovoltaic ultra-short-term power prediction sub-model to obtain an ultra-short-term power prediction result.

3. The ultra-short-term power prediction method based on a UAV under a photovoltaic panel shading state as claimed in claim 2 is characterized in that: The photovoltaic ultra-short-term power prediction model also includes a correction sub-model, and the correction sub-model is used to output correction parameters of the ultra-short-term power prediction result. The ultra-short-term power prediction method based on the photovoltaic panel shielding state of the drone also includes: Extracting each measured occlusion area value in the measured occlusion area sequence; The measured occlusion area value is input into the correction sub-model to obtain the correction parameter.

4. The ultra-short-term power prediction method based on a UAV under a photovoltaic panel shielding state as described in any one of claims 1 to 3, characterized in that: The ultra-short-term power prediction method under the photovoltaic panel shading state based on the drone also includes: Extracting occlusion thickness from measured image data of the measured image sequence to obtain a first measured occlusion thickness sequence; Extracting a time-series dependency relationship of the first measured occlusion thickness sequence to obtain first measured occlusion thickness change information; The first measured shading thickness change information and the measured shading area change information are input into the second photovoltaic ultra-short-term power prediction model to obtain the photovoltaic ultra-short-term power prediction result; wherein, the second photovoltaic ultra-short-term power prediction model is trained by a training set consisting of the first historical shading thickness change information, the historical shading area change information, and the historical power change information.

5. The ultra-short-term power prediction method based on a UAV under a photovoltaic panel shielding state according to any one of claims 1 to 3, characterized in that: The ultra-short-term power prediction method under the photovoltaic panel shading state based on the drone also includes: Extracting occlusion thickness from the measured image data of the measured image sequence to obtain a second measured occlusion thickness sequence; The second measured shading thickness sequence is input into the third photovoltaic ultra-short-term power prediction model to obtain the photovoltaic ultra-short-term power prediction result; wherein, the third photovoltaic ultra-short-term power prediction model is trained by a second historical data set corresponding to multiple historical snowfall events of the target photovoltaic power station, and the second historical data set includes multiple historical image sequences collected by drones at a preset frequency and their corresponding historical power sequences.

6. A method for constructing an ultra-short-term power prediction model under photovoltaic panel shading state based on drones, characterized in that: The method for constructing an ultra-short-term power prediction model under the photovoltaic panel shading state based on the drone includes: Acquire a first historical data set corresponding to multiple historical snowmelt events of a target photovoltaic power station, wherein the first historical data set includes a plurality of historical image sequences collected by a drone at a preset frequency in an ultra-short-term prediction period and their corresponding historical power sequences; the historical image sequence includes a plurality of historical image data for representing the snowmelt conditions of all photovoltaic panels installed in the target photovoltaic station; Extracting the occlusion area of ​​the historical image sequence in each of the ultra-short-term prediction periods to obtain a historical occlusion area sequence; Extracting the temporal dependency relationship of the historical blocking area sequence and the historical power sequence to obtain historical blocking area change information and historical power change information respectively; A training data set is constructed based on the historical shading area change information and the historical power change information under each ultra-short-term prediction period, and the training data set is input into a pre-constructed preset power prediction model for model training. During the training process, the time series correspondence between the historical shading area change information and the historical power change information is learned to obtain the first photovoltaic ultra-short-term power prediction model.

7. The method for constructing an ultra-short-term power prediction model under photovoltaic panel shading state based on an unmanned aerial vehicle according to claim 6, characterized in that: The photovoltaic ultra-short-term power prediction model includes at least two photovoltaic ultra-short-term power prediction sub-models, and the method for constructing an ultra-short-term power prediction model under the photovoltaic panel shading state based on the drone also includes: Clustering the historical shielding area sequences in each of the ultra-short-term prediction periods according to the area change rate to obtain at least two clustering clusters; Extracting the temporal dependency relationship between each cluster and the corresponding historical power sequence respectively, to obtain a training data set corresponding to each cluster; wherein the training data set includes the historical occlusion area change information and the historical power change information corresponding to each cluster; The training data set is input into a pre-built preset power prediction model for model training to obtain a photovoltaic ultra-short-term power prediction sub-model corresponding to each of the clusters.

8. The method for constructing an ultra-short-term power prediction model under the shading state of photovoltaic panels based on drones as claimed in claim 7, characterized in that: The photovoltaic ultra-short-term power prediction model also includes a correction sub-model, which is used to output correction parameters of the ultra-short-term power prediction result. The method for constructing an ultra-short-term power prediction model based on the photovoltaic panel shading state of the drone includes: Extracting each historical blocking area value in the historical blocking area sequence and historical power data corresponding to the historical blocking area value in time series as a first training set; The first training set is input into the correction sub-model for model training. During the training process, the corresponding relationship between the historical occlusion area value, the historical power data and the correction parameter is learned.

9. The method for constructing an ultra-short-term power prediction model based on a photovoltaic panel under shading state based on an unmanned aerial vehicle according to any one of claims 6 to 8, characterized in that: The method for constructing an ultra-short-term power prediction model under the photovoltaic panel shading state based on the drone also includes: Extracting the occlusion thickness of the historical image sequence in each ultra-short-term prediction period to obtain a first historical occlusion thickness sequence; Extracting a temporal dependency relationship of the first historical occlusion thickness sequence to obtain first historical occlusion thickness change information; The first historical shading thickness change information is added to the training data set, and the added new training data set is input into the power prediction model for model training. During the training process, the time series correspondence between the historical shading area change information, the first historical shading thickness change information and the historical power change information is learned to obtain a second photovoltaic ultra-short-term power prediction model.

10. The method for constructing an ultra-short-term power prediction model under photovoltaic panel shading state based on an unmanned aerial vehicle according to any one of claims 6 to 8, characterized in that: The method for constructing an ultra-short-term power prediction model under the photovoltaic panel shading state based on the drone also includes: Acquire a second historical data set corresponding to multiple historical snowfall events of the target photovoltaic power station; wherein the second historical data set includes multiple historical image sequences collected by drones at a preset frequency and their corresponding historical power sequences; the historical image sequences include multiple historical image data for representing all photovoltaic panels installed in the target photovoltaic station; Extracting occlusion thickness from the historical image sequence to obtain a second historical occlusion thickness sequence; Extracting the time series dependency of the historical shielding thickness sequence and the historical power sequence to obtain second historical shielding thickness change information and historical power change information respectively; A second training data set is constructed based on the second historical shading thickness change information and the historical power change information, and the second training data set is input into a pre-constructed preset power prediction model for model training. During the training process, the time series correspondence between the second historical shading thickness change information and the historical power change information is learned to obtain a third photovoltaic ultra-short-term power prediction model.

Images