Unmanned aerial vehicle-based short-term power prediction model construction method and prediction method

Abstract

The application relates to the field of new energy technology, and discloses a short-term power prediction model construction method and prediction method based on a UAV, historical data sets corresponding to historical multiple snowfall events of a target photovoltaic power station are acquired; the historical image data of each group of historical image sequences is subjected to occlusion thickness extraction to obtain historical occlusion thickness sequences; time sequence dependency relationship extraction is performed on the historical occlusion thickness sequences and historical power sequences to respectively obtain historical occlusion thickness change information and historical power change information; a training data set is constructed based on the historical occlusion thickness change information and the historical power change information, the training data set is input into a preset power prediction model which is constructed in advance to perform model training, in the training process, the time sequence corresponding relationship of the historical occlusion thickness change information and the historical power change information is learned to obtain a first photovoltaic short-term power prediction model, and the power prediction model can adapt to the accompanying relationship between different thickness changes and power changes.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, specifically to a method for constructing and predicting short-term power prediction models based on unmanned aerial vehicles (UAVs). Background Technology

[0002] A photovoltaic (PV) panel, also known as a solar panel, is a device that converts solar energy into electrical energy. It mainly consists of solar cells, which use the photovoltaic effect to convert sunlight into electrons, thereby generating an electric current.

[0003] However, photovoltaic panels are easily affected by environmental factors during use. For example, in low-temperature environments with snowfall, photovoltaic panels may be covered with snow, which obstructs sunlight from passing through the snow layer. Furthermore, the snow layer increases scattering losses. Therefore, the shading and scattering of sunlight by the snow layer will affect the power generation efficiency, leading to inaccurate power prediction.

[0004] Therefore, there is an urgent need to propose a short-term power prediction method based on drones to solve the problem of inaccurate power prediction caused by environmental influences on photovoltaic panels in related technologies. Summary of the Invention

[0005] In view of this, the present invention provides a method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs) to solve the problem of inaccurate power prediction caused by environmental influences on photovoltaic panels in related technologies.

[0006] In a first aspect, the present invention provides a method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs). The method includes: acquiring historical datasets corresponding to multiple historical snowfall events at a target photovoltaic power station; wherein the historical datasets include multiple sets of historical image sequences and historical power sequences, the historical image sequences being acquired by the UAV at a preset acquisition frequency; extracting occlusion thickness from the historical image data of each set of historical image sequences to obtain a historical occlusion thickness sequence; extracting temporal dependencies from the historical occlusion thickness sequence and the historical power sequence to obtain historical occlusion thickness change information and historical power change information, respectively; constructing a training dataset based on the historical occlusion thickness change information, the historical power change information, and the corresponding meteorological data; inputting the training dataset into a pre-constructed preset power prediction model for model training; during the training process, learning the temporal correspondence between the historical occlusion thickness change information and the historical power change information to obtain a first photovoltaic short-term power prediction model.

[0007] As an exemplary embodiment, the photovoltaic short-term power prediction model includes at least two photovoltaic short-term power prediction sub-models. The method for constructing a drone-based short-term power prediction model further includes: truncating a historical shading thickness sequence according to a preset duration to obtain a historical shading thickness sub-sequence; clustering the historical shading thickness sub-sequences according to the thickness change rate to obtain at least two clusters; extracting the temporal dependency relationship for each cluster and the historical power sequence corresponding to each cluster to obtain a training dataset corresponding to each cluster; wherein, the training dataset includes historical shading thickness change information and historical power change information corresponding to each cluster; and inputting the training dataset into a pre-constructed preset power prediction model for model training to obtain photovoltaic short-term power prediction sub-models corresponding to each cluster.

[0008] As an exemplary embodiment, the photovoltaic short-term power prediction model further includes a first correction sub-model. The first correction sub-model is used to output the first correction parameter of the short-term power prediction result. The method for constructing the short-term power prediction model based on UAV includes: extracting each historical shading thickness value and the historical power data corresponding to the time sequence of the historical shading thickness values ​​from the historical shading thickness sequence as a first training set; inputting the first training set into the first correction sub-model for model training, and learning the correspondence between the historical shading thickness value, the historical power data and the first correction parameter during the training process.

[0009] As an exemplary embodiment, the method for constructing a short-term power prediction model based on UAVs further includes: extracting the occlusion area from historical image sequences to obtain a historical occlusion area sequence; extracting the temporal dependency relationship from the historical occlusion area sequence to obtain historical occlusion area change information; adding the historical occlusion area change information to the training dataset; inputting the newly added training dataset into the power prediction model for model training; and during the training process, learning the correspondence between historical occlusion thickness change information, historical occlusion area change information, and historical power change information to obtain a second photovoltaic short-term power prediction model.

[0010] As an exemplary embodiment, the first photovoltaic short-term power prediction model further includes a second correction sub-model. The method for constructing a short-term power prediction model based on UAVs further includes: extracting the occlusion area from historical image sequences to obtain a historical occlusion area sequence; extracting the temporal dependency relationship from the historical occlusion area sequence to obtain historical occlusion area change information; inputting the historical occlusion area change information, historical occlusion thickness change information, and historical power change information into the second correction sub-model; and during the training process, learning the correspondence between historical occlusion thickness change information and historical power change information, as well as the correspondence between historical occlusion thickness change information, historical power change information, and the second correction parameter, to obtain the second correction sub-model.

[0011] Secondly, this invention provides a short-term power prediction method based on unmanned aerial vehicles (UAVs). The method includes: acquiring a sequence of measured images of a target photovoltaic (PV) power plant and predicted meteorological data, wherein the measured image sequence is acquired by the UAV at a preset acquisition frequency; extracting the shading thickness from the measured image sequence to obtain a measured shading thickness sequence; extracting the temporal dependency of the measured shading thickness sequence to obtain measured shading thickness variation information; and inputting the measured shading thickness variation information and predicted meteorological data into a PV short-term power prediction model to obtain a PV short-term power prediction result. The PV short-term power prediction model is constructed based on historical shading thickness variation information and historical power variation information to build a training dataset. The training dataset is input into a pre-constructed preset power prediction model for model training, and the model learns the temporal correspondence between historical shading thickness variation information and historical power variation information during the training process.

[0012] As an exemplary embodiment, the short-term power prediction method based on UAVs further includes: truncating the measured shading thickness sequence according to a preset duration to obtain a measured shading thickness subsequence; clustering the measured shading thickness subsequence with clusters, and selecting the photovoltaic short-term power prediction sub-model corresponding to the cluster with a cluster degree greater than a preset degree; wherein, the clusters are obtained by clustering the historical shading thickness subsequences according to the thickness change rate; and inputting the measured shading thickness subsequence into the photovoltaic short-term power prediction sub-model to obtain the photovoltaic short-term power prediction result.

[0013] As an exemplary embodiment, the photovoltaic short-term power prediction model further includes a first correction sub-model, which is used to output a first correction parameter for the short-term power prediction result. The UAV-based short-term power prediction method includes: extracting actual shading data from the measured shading thickness sequence; inputting the actual shading data into the first correction sub-model to obtain the first correction parameter; and correcting the photovoltaic short-term power prediction result based on the first correction parameter.

[0014] As an exemplary embodiment, the UAV-based short-term power prediction method further includes: extracting the occlusion area from the measured image sequence to obtain measured occlusion area data; extracting the temporal dependency relationship from the measured occlusion area data to obtain measured occlusion area change information; and inputting the measured occlusion area change information and the measured occlusion thickness change information into the second photovoltaic short-term power prediction model to obtain photovoltaic short-term power prediction results.

[0015] As an exemplary embodiment, the photovoltaic short-term power prediction model also includes a second correction sub-model. The method for constructing the short-term power prediction model based on UAVs further includes: extracting the occlusion area from the measured image sequence to obtain measured occlusion area data; extracting the temporal dependency relationship from the measured occlusion area data to obtain measured occlusion area change information; inputting the measured occlusion area change information and the measured occlusion thickness change information into the second correction sub-model to obtain second correction parameters; and correcting the photovoltaic short-term power prediction results based on the second correction parameters.

[0016] The present invention provides a method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs). The method includes: acquiring historical datasets corresponding to multiple snowfall events at a target photovoltaic power station; wherein the historical datasets include multiple sets of historical image sequences and historical power sequences, and the historical image sequences are acquired by the UAV according to a preset acquisition frequency; extracting occlusion thickness from the historical image data of each set of historical image sequences to obtain a historical occlusion thickness sequence; extracting the temporal dependency relationship between the historical occlusion thickness sequence and the historical power sequence to obtain historical occlusion thickness change information and historical power change information, respectively; constructing a training dataset based on the historical occlusion thickness change information and historical power change information; and inputting the training dataset into a pre-constructed preset power prediction model for modeling. During training, the temporal correspondence between historical shading thickness change information and historical power change information is learned to obtain a first photovoltaic short-term power prediction model. In the above implementation method of this embodiment, the historical shading thickness change information can reflect the temporal change of the photovoltaic panel thickness during the entire snowfall cycle, and the historical power change information can reflect the temporal change of the photovoltaic panel power during the entire snowfall cycle. By using the above 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 short-term power prediction model can adapt to the co-occurrence relationship between thickness change and power change in different time intervals during the entire snowfall cycle, thereby solving the problem of inaccurate prediction when using traditional photovoltaic short-term power prediction models for power prediction under snowfall events. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the method for constructing a short-term power prediction model based on an unmanned aerial vehicle (UAV) according to an embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating a short-term power prediction method based on an unmanned aerial vehicle (UAV) according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Photovoltaic panels are easily affected by environmental factors during use. For example, in low-temperature environments with snowfall, photovoltaic panels may be covered in snow, which obstructs sunlight from passing through the snow layer. The snow layer increases scattering losses, affecting power generation efficiency and leading to inaccurate power prediction. Specifically, in this invention, the period from the start of a snowfall event to the complete melting of the snow is called the entire snowfall cycle, denoted as a snowfall event. In this snowfall event:

[0022] If the photovoltaic panel is completely covered by snow at the first preset moment after the start of the snowfall event, the photovoltaic panel will still generate power during the first time interval from the start of the snowfall to the first preset moment, even though it is covered by snow. However, the photovoltaic panel will gradually be covered by snow during this interval, and the power prediction will be inaccurate due to the snow blocking sunlight and the increased scattering loss.

[0023] If the snowfall event continues until the second preset time, within the second time interval corresponding to the first preset time and the second preset time, the thickness of the snow covering the photovoltaic panel surface gradually increases over time until the second preset time when the snowfall event ends, at which point the thickness of the snow covering the photovoltaic panel surface reaches its maximum value. When the photovoltaic panel is completely covered by snow, the snow layer still has some light transmission, and the photovoltaic panel still generates power until the snow layer reaches a preset thickness that is completely opaque. At this point, the photovoltaic panel generates zero power, and this continues until the snowfall ends. Therefore, within this interval, the photovoltaic panel generates power gradually decreases until it reaches zero.

[0024] After the snowfall event ends, as the temperature rises, the snowfall cycle changes to a snow melting event where the snow surface begins to melt. For example, at the third preset time, the snow on the photovoltaic panel melts to a preset thickness. In the third time interval between the second and third preset times, the thickness of the snow covering the photovoltaic panel gradually decreases over time, and the photovoltaic panel's power generation is always zero.

[0025] During the fourth time interval from the second preset time when the snow melting event begins to the fourth preset time when the snow melting ends, the snow coverage area of ​​the photovoltaic panels will continuously and dynamically change. For example, due to issues such as the installation angle of the photovoltaic panels, for the same photovoltaic panel, some snow will melt first, while other snow will melt later; or, after snowfall, for photovoltaic power stations, due to snow removal efficiency, some photovoltaic panels will be cleared of snow, but other photovoltaic panels will still be covered with snow. In this case, if the dynamic changes in area are not considered, there will be problems with inaccurate power prediction.

[0026] Therefore, during the first, second, third, and fourth time intervals of the entire snowfall cycle, when the photovoltaic panels have power, the power prediction is inaccurate due to the snow layer blocking sunlight and the increased scattering loss. The period from the start of snowfall to the end of snow melting usually takes about a week, while the time dimension of short-term photovoltaic power prediction is usually 1 day, 2 days, ..., N days. Under snowfall events, the traditional short-term photovoltaic power prediction model has the problem of inaccurate prediction.

[0027] Based on this, according to an embodiment of the present invention, an embodiment of a method for constructing a short-term power prediction model based on a UAV is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although the flowchart shows the order in which the logical historical occlusion thickness subsequences are clustered according to the thickness change rate, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] This embodiment provides a method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs). Figure 1 This is a flowchart of a method for constructing a short-term power prediction model based on an unmanned aerial vehicle (UAV) according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0029] Step S101: Obtain historical datasets corresponding to multiple snowfall events in the history of the target photovoltaic power station; wherein, the historical datasets include multiple sets of historical image sequences and historical power sequences, and the historical image sequences are collected by UAVs according to a preset acquisition frequency.

[0030] For example, a snowfall event is the event corresponding to the start of snowfall and the completion of snow melting for the target photovoltaic power station.

[0031] For example, the historical dataset includes multiple sets of historical image sequences and historical power sequences; wherein, the historical image sequences are acquired by drones according to a preset acquisition frequency; specifically, the drones can be controlled to perform an inspection every 5 minutes, 10 minutes, 15 minutes, etc., according to a preset inspection path, and capture image data during each inspection, and the multiple sets of image data captured by multiple inspections are used as the historical image sequences corresponding to the current snowfall event.

[0032] For example, the preset inspection route can be pre-modeled and image-recognized based on GIS and remote sensing technologies to determine the distribution of photovoltaic panels in the target photovoltaic power station, and further determined based on the distribution.

[0033] For example, historical power sequences can be obtained by measuring the power generation of a photovoltaic power plant; wherein the time axis of the historical power sequence is the same as that of the historical image sequence.

[0034] Step S102: Extract the occlusion thickness from the historical image data of each set of historical image sequences to obtain a historical occlusion thickness sequence.

[0035] In this embodiment, after obtaining the historical image data, image recognition can be used to perform image recognition based on occlusion thickness on each historical image data in each set of historical image sequences, so as to extract the occlusion thickness of the historical image data in each set of historical image sequences and obtain a historical occlusion thickness sequence.

[0036] For example, 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 the historical image data, image preprocessing, including image denoising, image enhancement, and removal of irrelevant surrounding regions, is performed on the historical image data.

[0037] 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 the feature representation of the image. In the CNN, the image is converted into a vector form. The convolutional layer slides across the image through filters (also called convolutional kernels) to extract features from the image. Different filters can capture different features such as edges, textures, and colors, transforming the image from the pixel level into a feature map with semantic information.

[0038] Specifically, CNN decomposes the color image input into three channels: R (red), G (green), and B (blue), with each value ranging from 0 to 255.

[0039] Because conventional neural networks use fully connected layers for feature extraction between the input and hidden layers, processing even slightly large images can lead to enormous computational costs and become very slow. Therefore, convolutional operations are used to extract features.

[0040] Specifically, the Rectified Linear Unit (ReLU) function is used to activate the CNN convolutional neural network. One-sided inhibition is applied to reduce all negative values ​​to zero, while leaving positive values ​​unchanged. This results in sparse activation of the neurons in the neural network. The sparse model achieved through ReLU can better extract relevant features and fit the training data.

[0041] Furthermore, the target input is pooled to extract features, reducing the amount of data passed to the next stage.

[0042] Furthermore, based on the obtained results, a fully connected layer is formed, which acts as a "classifier" in the convolutional neural network to obtain the occlusion thickness corresponding to each image, thereby obtaining the historical occlusion thickness sequence.

[0043] Step S103: Extract the temporal dependency of the historical occlusion thickness sequence and the historical power sequence to obtain the historical occlusion thickness change information and the historical power change information, respectively.

[0044] In order to determine the impact of the temporal variation of the photovoltaic panel shading thickness caused by snowfall on historical power data in each snowfall event, in this embodiment, the temporal dependency relationship is extracted from the historical shading thickness sequence and the historical power sequence to obtain historical shading thickness variation information and historical power variation information, respectively.

[0045] As a possible implementation, the historical occlusion thickness sequence and historical power sequence can be input into a gated recurrent unit (GRU) or a long short-term memory (LSTM) network to extract the temporal dependencies of the historical occlusion thickness sequence and historical power sequence through GRU or LSTM, so as to obtain the historical occlusion thickness change information and historical power change information respectively.

[0046] Step S104: Construct a training dataset based on historical shading thickness change information, historical power change information, and corresponding meteorological data. Input the training dataset into a pre-built preset power prediction model for model training. During the training process, learn the temporal correspondence between historical shading thickness change information and historical power change information to obtain the first photovoltaic short-term power prediction model.

[0047] In this embodiment, the input to the pre-built preset power prediction model may include multi-dimensional data; for example, the input to 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 dataset is constructed based on historical shading thickness change information and historical power change information, which may include historical shading thickness change information, historical power change information, and their corresponding meteorological data; after constructing the training dataset, the training dataset is input into the pre-built preset power prediction model for model training. During the training process, the temporal correspondence between historical shading thickness change information and historical power change information is learned to obtain the first photovoltaic short-term power prediction model.

[0048] It should be understood that, in the above-described implementation of this embodiment, the historical shading thickness change information can reflect the temporal change of snow thickness of the photovoltaic panel throughout the entire snowfall cycle, and the historical power change information can reflect the temporal change of power of the photovoltaic panel throughout the entire snowfall cycle. By using the above-described 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 short-term power prediction model can adapt to the co-occurrence of thickness change and power change in different time intervals throughout the entire snowfall cycle, thereby solving the problem of inaccurate prediction when using the traditional photovoltaic short-term power prediction model for power prediction under snowfall events.

[0049] As an exemplary embodiment, the photovoltaic short-term power prediction model includes at least two photovoltaic short-term power prediction sub-models. The method for constructing a drone-based short-term power prediction model further includes: truncating a historical shading thickness sequence according to a preset duration to obtain a historical shading thickness sub-sequence; clustering the historical shading thickness sub-sequences according to the thickness change rate to obtain at least two clusters; extracting the temporal dependency relationship for each cluster and the historical power sequence corresponding to each cluster to obtain a training dataset corresponding to each cluster; wherein, the training dataset includes historical shading thickness change information and historical power change information corresponding to each cluster; and inputting the training dataset into a pre-constructed preset power prediction model for model training to obtain photovoltaic short-term power prediction sub-models corresponding to each cluster.

[0050] Different snowfall rates during snowfall weather lead to changes in the snow accumulation rate on photovoltaic panels. These different snow accumulation rates result in varying shading thicknesses on the photovoltaic panels per unit time, which in turn manifests as different power variations. Therefore, to differentiate the impact of different snow accumulation rates on power variations, this invention divides the system into clusters based on different snow accumulation rates, and trains different short-term photovoltaic power prediction sub-models for each cluster.

[0051] In this embodiment, the historical shading thickness sequence is truncated according to a preset duration to obtain a historical shading thickness subsequence. After obtaining the historical shading thickness subsequence, K-clustering can be used to cluster the historical shading thickness subsequence according to the thickness change rate to obtain at least two clusters. Each cluster can reflect the historical shading thickness subsequence and historical power subsequence corresponding to different snow accumulation rates. Further, GRU and LSTM can be used to extract the temporal dependencies of each cluster and the corresponding historical power subsequence and historical shading thickness subsequence to obtain the training dataset corresponding to each cluster. The training dataset includes the historical shading thickness change information and historical power change information corresponding to each cluster. Further, the training dataset is input into a pre-built preset power prediction model for model training to obtain the photovoltaic short-term power prediction sub-model corresponding to each cluster, so that the influence of different snow accumulation rates on the shading thickness of photovoltaic panels per unit time can be considered during model training.

[0052] For example, the preset duration is greater than or equal to the duration corresponding to the preset acquisition frequency of the drone.

[0053] To consider the impact of snow thickness on short-term power prediction results from a static feature perspective, and thus enable the trained model to correct the short-term power prediction results based on this static feature, as an exemplary embodiment, the photovoltaic short-term power prediction model further includes a first correction sub-model. The first correction sub-model is used to output the first correction parameter of the short-term power prediction result. The method for constructing a UAV-based short-term power prediction model includes: extracting each historical shading thickness value and the historical power data corresponding to the time sequence of the historical shading thickness values ​​from the historical shading thickness sequence as a first training set; inputting the first training set into the first correction sub-model for model training, and learning the correspondence between historical shading thickness values, historical power data and the first correction parameter during the training process.

[0054] In this embodiment, the first correction sub-model is used to output the first correction parameter of the short-term power prediction result. The first correction parameter can be a correction coefficient, a correction value, or a combination of a correction coefficient and a correction value.

[0055] For example, after obtaining the historical shading thickness sequence, each historical shading thickness value and the historical power data corresponding to the time series of the historical shading thickness values ​​are extracted from the historical shading thickness sequence as a first training set. The first training set is then input into a first correction sub-model for model training. The first correction sub-model can be a neural network model, which maps the mathematical relationship between the historical shading thickness values, the time series-corresponding historical power data, and the first correction parameters. Specifically, the training process can be as follows: the first training set is input into the first correction sub-model to obtain the first correction parameters output by the first correction sub-model; the first correction parameters are then mathematically calculated with the photovoltaic short-term power prediction result output by the trained first photovoltaic short-term power prediction model to obtain the first corrected prediction result; in each iteration, the parameters of the first correction sub-model are continuously adjusted to minimize the error between the first corrected prediction result and the historical power data corresponding to the time series of the first corrected prediction result, until the error meets a preset condition.

[0056] During the first and fourth time intervals, although the photovoltaic panels are covered with snow, the snow cover is relatively small, and the photovoltaic panels still generate power. Furthermore, when the photovoltaic panels are not completely covered by snow, the power generation is also affected by the coverage area of ​​the photovoltaic panels. To account for these effects, as an exemplary embodiment, the method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs) further includes: extracting the occlusion area from historical image sequences to obtain a historical occlusion area sequence; extracting the temporal dependency relationship from the historical occlusion area sequence to obtain historical occlusion area change information; adding the historical occlusion area change information to the training dataset; inputting the newly added training dataset into the power prediction model for model training; and during the training process, learning the correspondence between historical occlusion thickness change information, historical occlusion area change information, and historical power change information to obtain a second photovoltaic short-term power prediction model.

[0057] In this embodiment, deep learning, machine learning, and other methods can be used to extract image features representing occlusion area information from the historical image sequences of each group of historical image sequences; for example, CNN can be used to implement image recognition; for machine learning... In deep image recognition, convolutional neural networks even outperform humans in tasks such as classifying objects into fine-grained categories. Furthermore, the most popular deep learning models such as YOLO, SSD, and RCNN also use convolutional layers to parse images or photographs.

[0058] For example, support vector machines, feature models, etc., can also be used to extract image features representing occlusion area information from the historical image sequences of each set of historical image sequences.

[0059] To determine the impact of the temporal changes in the shading area of ​​photovoltaic panels caused by snowfall in the first and fourth time intervals of each snowfall event on historical power data, in this embodiment, the temporal dependency of the historical shading area sequence is extracted to obtain historical shading area change information. As a possible implementation, the historical shading area sequence can be input into GRU or LSTM to obtain historical shading area change information.

[0060] After obtaining the historical occlusion area change information, the historical occlusion area change information is added to the dataset containing the historical occlusion thickness change information or historical power change information corresponding to its time series, forming a new training dataset. The new training dataset is then input into the power prediction model for model training.

[0061] It should be understood that, in the above-described implementation of this embodiment, the historical shading area change information can reflect the temporal changes in the shading area of ​​the photovoltaic panel within the first and fourth time intervals, the historical shading thickness change information can reflect the temporal changes in the thickness of the photovoltaic panel, and the historical power change information can reflect the temporal changes in the power of the photovoltaic panel. Therefore, by adopting the above-described 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 short-term power prediction model can adapt to the accompanying relationship between thickness changes, area changes, and power changes in different time intervals throughout the entire snowfall cycle. Moreover, it adapts to the accompanying relationship between thickness changes, area changes, and power changes in the first and fourth time intervals, thereby solving the problem of inaccurate prediction when using the traditional photovoltaic short-term power prediction model for power prediction under snowfall events.

[0062] As an exemplary embodiment, the first photovoltaic short-term power prediction model further includes a second correction sub-model. The method for constructing a short-term power prediction model based on UAVs further includes: extracting the occlusion area from historical image sequences to obtain a historical occlusion area sequence; extracting the temporal dependency from the historical occlusion area sequence to obtain historical occlusion area change information; obtaining historical power change information corresponding to the temporal sequence of historical occlusion area change information as a second training dataset; inputting the second training dataset into the second correction sub-model, and during the training process, learning the correspondence between historical occlusion area change information, historical power change information, and the second correction parameter to obtain the second correction sub-model.

[0063] In this embodiment, the second correction sub-model is used to output the second correction parameter of the short-term power prediction result. The second correction parameter can be a correction coefficient, a correction value, or a combination of a correction coefficient and a correction value.

[0064] For example, after obtaining the historical occlusion area sequence, the occlusion area is extracted from the historical image sequence to obtain a historical occlusion area sequence. Further, temporal dependency extraction is performed on the historical occlusion area sequence to obtain historical occlusion area change information. For example, GRU or LSTM can be used to implement temporal dependency extraction. After obtaining the historical occlusion area change information, historical power change information corresponding to the temporal sequence of the historical occlusion area change information is obtained as a second training dataset. The second training dataset is input into a second correction sub-model for model training. The second correction sub-model can be a neural network model, which is used to map historical occlusion area changes. The mathematical relationship between the information, the historical power change information corresponding to the time series, and the second correction parameter is established. Specifically, the training process can be as follows: inputting the second training set into the second correction sub-model to obtain the second correction parameter output by the second correction sub-model; performing mathematical operations on the second correction parameter with the photovoltaic short-term power prediction result output by the trained first photovoltaic short-term power prediction model or the second photovoltaic short-term power prediction model to obtain the second correction prediction result; in each iteration, continuously adjusting the parameters of the second correction sub-model to make the error between the second correction prediction result and the historical power data corresponding to the time series of the second correction prediction result smaller, until the error meets the preset conditions.

[0065] Secondly, this embodiment provides a short-term power prediction method based on unmanned aerial vehicles (UAVs). Figure 2 This is a flowchart of a short-term power prediction method based on an unmanned aerial vehicle (UAV) according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0066] Step S201: Obtain the measured image sequence and predicted meteorological data of the target photovoltaic power station; wherein, the measured image sequence is obtained by the UAV according to the preset acquisition frequency.

[0067] In this embodiment, the measured image sequence is obtained by the UAV according to a preset acquisition frequency; specifically, the UAV can be controlled to perform an inspection every 5 minutes, 10 minutes, 15 minutes, etc., according to a preset inspection path, and image data is captured during each inspection. The multiple sets of image data captured by multiple inspections are used as the measured image sequence; the predicted meteorological data can be determined by weather forecast.

[0068] Step S202: Extract the occlusion thickness from the measured image sequence to obtain the measured occlusion thickness sequence.

[0069] In this embodiment, the occlusion thickness can be extracted from the measured image sequence using image recognition to obtain the measured occlusion thickness sequence; the specific implementation method can be found in the content described in the above embodiments, and will not be repeated here.

[0070] Step S203: Extract the temporal dependency of the measured occlusion thickness sequence to obtain the measured occlusion thickness change information.

[0071] In order to determine the temporal variation of the shading thickness of the photovoltaic panel corresponding to the current measured image sequence, in this embodiment, the temporal dependency of the measured shading thickness sequence is extracted. As a possible implementation, the measured shading thickness sequence can be input into GRU or LSTM to extract the temporal dependency of the measured shading thickness sequence through GRU or LSTM, so as to obtain the measured shading thickness variation information.

[0072] Step S204: Input the measured shading thickness change information and the predicted meteorological data into the photovoltaic short-term power prediction model to obtain the photovoltaic short-term power prediction result; wherein, the photovoltaic short-term power prediction model is based on the historical shading thickness change information and historical power change information to construct a training dataset, and the training dataset is input into the pre-constructed preset power prediction model for model training, and the time-series correspondence between the historical shading thickness change information and the historical power change information is learned during the training process.

[0073] In this embodiment, the photovoltaic short-term power prediction model is based on a training dataset constructed from historical shading thickness variation information and historical power variation information. The training dataset is input into a pre-built preset power prediction model for model training. During the training process, the model learns the temporal correspondence between historical shading thickness variation information and historical power variation information. The photovoltaic short-term power prediction model obtained by this training method can adapt to the co-occurrence of thickness variation and power variation in different time intervals throughout the entire snowfall cycle. Therefore, by inputting the measured shading thickness variation information and the predicted meteorological data into the photovoltaic short-term power prediction model, the co-occurrence of the current measured shading thickness variation information and power variation can be considered to predict photovoltaic power, thereby obtaining photovoltaic power prediction results adapted to the time interval under the current snowfall event. This can solve the problem of inaccurate prediction when using traditional photovoltaic short-term power prediction models for power prediction under snowfall events.

[0074] As an exemplary embodiment, the short-term power prediction method based on UAVs further includes: truncating the measured shading thickness sequence according to a preset duration to obtain a measured shading thickness subsequence; clustering the measured shading thickness subsequence with clusters, and selecting the photovoltaic short-term power prediction sub-model corresponding to the cluster with a cluster degree greater than a preset degree; wherein, the clusters are obtained by clustering the historical shading thickness subsequences according to the thickness change rate; and inputting the measured shading thickness subsequence into the photovoltaic short-term power prediction sub-model to obtain the photovoltaic short-term power prediction result.

[0075] Different snowfall rates during snowfall weather will cause changes in the snow accumulation rate of photovoltaic panels, and different snow accumulation rates will result in different shading thicknesses of photovoltaic panels per unit time, which in turn will manifest as different power changes of photovoltaic panels. Therefore, in order to select the corresponding photovoltaic short-term power prediction sub-model based on the current measured shading thickness, clustering is performed based on the measured shading thickness subsequence and clusters, and the photovoltaic short-term power prediction sub-model corresponding to the cluster with a cluster degree greater than a preset degree can be selected. Specifically, the K-clustering method can be used to select the photovoltaic short-term power prediction sub-model corresponding to the cluster with a similarity to the current measured shading thickness greater than a preset degree for power prediction, so that the influence of different snow accumulation rates on the shading thickness of photovoltaic panels per unit time can be considered during prediction.

[0076] For example, the preset duration is greater than or equal to the duration corresponding to the preset acquisition frequency of the drone.

[0077] To consider the impact of snow thickness on short-term power prediction results from a static feature perspective, and to enable the short-term power prediction results to be corrected based on this static feature during prediction, as an exemplary embodiment, the photovoltaic short-term power prediction model further includes a first correction sub-model. The first correction sub-model is used to output a first correction parameter for the short-term power prediction results. The method for constructing a drone-based short-term power prediction model includes: extracting actual shading data from the measured shading thickness sequence; inputting the actual shading data into the first correction sub-model to obtain the first correction parameter; and correcting the photovoltaic short-term power prediction results based on the first correction parameter.

[0078] In this embodiment, the first correction sub-model is used to output a first correction parameter for the short-term power prediction result. This first correction parameter can be a correction coefficient, a correction value, or a combination of a correction coefficient and a correction value. For example, when the first correction parameter is a correction coefficient, the correction coefficient is multiplied by the short-term power prediction result; when the first correction parameter is a correction parameter, the correction coefficient is added to the short-term power prediction result; when the first correction parameter is a combination of a correction coefficient and a correction value, the correction coefficient and the correction value are multiplied and then added to the short-term power prediction result.

[0079] During the first and fourth time intervals, although the photovoltaic panels are covered with snow, the snow cover is relatively small, and the photovoltaic panels still generate power. When the photovoltaic panels are not completely covered by snow, the power generation is also affected by the coverage area of ​​the photovoltaic panels. To account for these effects, as an exemplary embodiment, the UAV-based short-term power prediction method further includes: extracting the occlusion area from the measured image sequence to obtain measured occlusion area data; extracting the temporal dependency relationship from the measured occlusion area data to obtain measured occlusion area change information; and inputting the measured occlusion area change information and the measured occlusion thickness change information into the second photovoltaic short-term power prediction model to obtain the photovoltaic short-term power prediction result.

[0080] In this embodiment, deep learning, machine learning and other methods can be used to extract image features representing occlusion area information from the historical image sequences of each group of historical image sequences; specifically, please refer to the content in the above embodiments, which will not be repeated in this embodiment.

[0081] To determine the impact of the current change in the shading area of ​​the photovoltaic panel over time on historical power data, in this embodiment, the temporal dependency of the measured shading area sequence is extracted to obtain the measured shading area change information. As a possible implementation, the historical shading area sequence can be input into GRU or LSTM to obtain the measured shading area change information.

[0082] After obtaining the measured shading area change information, the measured shading area change information and the measured shading thickness change information are input into the second photovoltaic short-term power prediction model to obtain the photovoltaic short-term power prediction result; wherein, the second photovoltaic short-term power prediction model is obtained by training the model using a training dataset consisting of historical shading thickness change information, historical shading area change information and historical power change information.

[0083] In the above-described embodiments of this application, the measured shading area change information can reflect the temporal change of the shading area, and the measured shading thickness change information can reflect the temporal change of the photovoltaic panel thickness. Therefore, by adopting the above-described power prediction method, the accompanying relationship between thickness change, area change and power change can be considered during the power prediction process, so as to solve the problem of inaccurate prediction when using the traditional photovoltaic short-term power prediction model for power prediction under snowfall events.

[0084] As an exemplary embodiment, the photovoltaic short-term power prediction model also includes a second correction sub-model. The method for constructing the short-term power prediction model based on UAVs further includes: extracting the occlusion area from the measured image sequence to obtain measured occlusion area data; extracting the temporal dependency relationship from the measured occlusion area data to obtain measured occlusion area change information; inputting the measured occlusion area change information and the measured occlusion thickness change information into the second correction sub-model to obtain second correction parameters; and correcting the photovoltaic short-term power prediction results based on the second correction parameters.

[0085] In this embodiment, the second correction sub-model is used to output the second correction parameter of the short-term power prediction result. The second correction parameter can be a correction coefficient, a correction value, or a combination of a correction coefficient and a correction value.

[0086] In the above-described embodiments of this application, the measured shading area change information can reflect the temporal change of the shading area, and the measured shading thickness change information can reflect the temporal change of the photovoltaic panel thickness. Therefore, by adopting the above-described power prediction method, the accompanying relationship between thickness change, area change and power change can be considered during the power prediction process, so as to solve the problem of inaccurate prediction when using the traditional photovoltaic short-term power prediction model for power prediction under snowfall events.

[0087] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0088] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause 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 methods in the above embodiments.

[0089] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or the indirect coupling or communication connection of units or modules may be electrical or other forms.

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

[0092] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0093] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0094] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0095] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs), characterized in that, include: Acquire historical datasets corresponding to multiple snowfall events at the target photovoltaic power station; wherein, the historical datasets include multiple sets of historical image sequences and historical power sequences, and the historical image sequences are acquired by UAVs according to a preset acquisition frequency; Occlusion thickness is extracted from the historical image data of each set of historical image sequences to obtain a historical occlusion thickness sequence. Temporal dependencies are extracted from the historical occlusion thickness sequence and the historical power sequence to obtain historical occlusion thickness change information and historical power change information, respectively. A training dataset is constructed based on the historical shading thickness change information, the historical power change information, and the corresponding meteorological data. The training dataset is then input into a pre-constructed preset power prediction model for model training. During the training process, the time-series correspondence between the historical shading thickness change information and the historical power change information is learned to obtain the first photovoltaic short-term power prediction model.

2. The method for constructing a short-term power prediction model based on unmanned aerial vehicles as described in claim 1, characterized in that, The photovoltaic short-term power prediction model includes at least two photovoltaic short-term power prediction sub-models, and the method for constructing the UAV-based short-term power prediction model further includes: The historical occlusion thickness sequence is truncated according to a preset duration to obtain a historical occlusion thickness subsequence; The historical occlusion thickness subsequences are clustered according to the thickness change rate to obtain at least two clusters; Temporal dependencies are extracted from each cluster and the corresponding historical power sequence to obtain a training dataset for each cluster; wherein, the training dataset includes the historical occlusion thickness change information and the historical power change information corresponding to each cluster. The training dataset is input into a pre-built preset power prediction model for model training to obtain photovoltaic short-term power prediction sub-models corresponding to each cluster.

3. The method for constructing a short-term power prediction model based on unmanned aerial vehicles as described in claim 2, characterized in that, The photovoltaic short-term power prediction model further includes a first correction sub-model, which is used to output the first correction parameter of the short-term power prediction result. The method for constructing the UAV-based short-term power prediction model includes: Extract each historical occlusion thickness value from the historical occlusion thickness sequence and the historical power data corresponding to the time sequence of the historical occlusion thickness values ​​as the first training set; The first training set is input into the first correction sub-model for model training. During the training process, the correspondence between the historical occlusion thickness value, the historical power data and the first correction parameter is learned.

4. The method for constructing a short-term power prediction model based on unmanned aerial vehicles as described in any one of claims 1-3, characterized in that, The method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs) also includes: The occlusion area is extracted from the historical image sequence to obtain the historical occlusion area sequence; Temporal dependencies are extracted from the historical occlusion area sequence to obtain information on changes in historical occlusion area; The historical shading area change information is added to the training dataset, and the new training dataset is input into the power prediction model for model training. During the training process, the correspondence between the historical shading thickness change information, the historical shading area change information, and the historical power change information is learned to obtain the second photovoltaic short-term power prediction model.

5. The method for constructing a short-term power prediction model based on unmanned aerial vehicles (UAVs) as described in any one of claims 1-3, characterized in that, The first photovoltaic short-term power prediction model also includes a second modified sub-model, and the method for constructing the UAV-based short-term power prediction model further includes: The occlusion area is extracted from the historical image sequence to obtain the historical occlusion area sequence; Temporal dependencies are extracted from the historical occlusion area sequence to obtain information on changes in historical occlusion area; The historical power change information corresponding to the time series of the historical occlusion area change information is obtained as the second training dataset; The second training dataset is input into the second correction sub-model. During the training process, the correspondence between the historical occlusion area change information, the historical power change information, and the second correction parameter is learned to obtain the second correction sub-model.

6. A short-term power prediction method based on unmanned aerial vehicles (UAVs), characterized in that, The UAV-based short-term power prediction method includes: Acquire a sequence of measured images and predicted meteorological data of the target photovoltaic power station, wherein the sequence of measured images is acquired by a UAV at a preset acquisition frequency; The occlusion thickness is extracted from the measured image sequence to obtain the measured occlusion thickness sequence; Temporal dependency extraction is performed on the measured occlusion thickness sequence to obtain the measured occlusion thickness variation information; The measured shading thickness change information and the predicted meteorological data are input into the photovoltaic short-term power prediction model to obtain the photovoltaic short-term power prediction result. The photovoltaic short-term power prediction model is constructed based on historical shading thickness change information and historical power change information to build a training dataset. The training dataset is input into the pre-constructed preset power prediction model for model training. During the training process, the time-series correspondence between the historical shading thickness change information and the historical power change information is learned.

7. The short-term power prediction method based on UAVs as described in claim 6, characterized in that, The UAV-based short-term power prediction method also includes: The measured occlusion thickness sequence is truncated according to a preset duration to obtain a measured occlusion thickness subsequence; Clustering is performed based on the measured shading thickness subsequence and clusters, and the photovoltaic short-term power prediction sub-model corresponding to the clusters with a clustering degree greater than a preset degree is selected; wherein, the clusters are obtained by clustering the historical shading thickness subsequences according to the thickness change rate; The measured shading thickness subsequence is input into the photovoltaic short-term power prediction sub-model to obtain the photovoltaic short-term power prediction result.

8. The short-term power prediction method based on UAVs as described in claim 7, characterized in that, The photovoltaic short-term power prediction model further includes a first correction sub-model, which is used to output a first correction parameter for the short-term power prediction result. The UAV-based short-term power prediction method further includes: Extract the actual occlusion data from the measured occlusion thickness sequence; The actual occlusion data is input into the first correction sub-model to obtain the first correction parameter; The short-term power prediction results of photovoltaics are corrected based on the first correction parameter.

9. The short-term power prediction method based on unmanned aerial vehicles as described in any one of claims 6-8, characterized in that, The UAV-based short-term power prediction method also includes: The occlusion area is extracted from the measured image sequence to obtain the measured occlusion area data; Temporal dependency extraction is performed on the measured shading area data to obtain the measured shading area change information; The measured shading area change information and the measured shading thickness change information are input into the second photovoltaic short-term power prediction model to obtain the photovoltaic short-term power prediction result.

10. The short-term power prediction method based on unmanned aerial vehicles as described in any one of claims 6-8, characterized in that, The photovoltaic short-term power prediction model also includes a second-modifier model, and the UAV-based short-term power prediction method also includes: The occlusion area is extracted from the measured image sequence to obtain the measured occlusion area data; Temporal dependency extraction is performed on the measured shading area data to obtain the measured shading area change information; The measured occlusion area change information and the measured occlusion thickness change information are input into the second correction sub-model to obtain the second correction parameters; The short-term photovoltaic power prediction results are corrected based on the second correction parameter.

Images