Photovoltaic power generation power prediction system and method based on satellite cloud picture

Abstract

The invention discloses a photovoltaic power generation power prediction system and method based on a satellite cloud picture, and relates to the technical field of data processing, and the system comprises a satellite cloud picture data obtaining module which is used for obtaining the satellite cloud picture data of a target area; the cloud layer feature extraction module is used for performing cloud layer feature extraction on the satellite cloud picture data; the data pairing module is used for extracting historical photovoltaic power generation data at the time corresponding to the cloud layer feature data and pairing the historical photovoltaic power generation data; a power prediction model training module; and the power generation power prediction module is used for performing photovoltaic power generation power prediction according to the satellite cloud picture data. The technical problems that an existing photovoltaic power generation power prediction method is insufficient in prediction precision under the large-range and complex weather conditions, and influences of cloud layer changes on the power generation power are difficult to deal with in real time are solved, and the purpose of predicting the power generation power through a deep learning model based on satellite cloud picture data is achieved. And the accuracy and real-time performance of photovoltaic power generation power prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a photovoltaic power prediction system and method based on satellite cloud images. Background Art

[0002] As an important part of renewable energy, the power generation capacity of photovoltaic power generation system is significantly affected by the intensity of solar radiation and meteorological conditions. Due to weather changes, especially the movement of clouds, it will directly affect the light intensity and power generation efficiency of photovoltaic power stations. Therefore, accurately predicting photovoltaic power generation is of great practical significance for the operation management, power dispatch and grid load balance of photovoltaic power stations.

[0003] Traditional photovoltaic power generation prediction methods are mainly based on historical meteorological data or solar radiation intensity for short-term or medium-term predictions. These methods usually rely on local weather stations or meteorological models, but have problems such as sparse data, limited spatial coverage, and insufficient prediction accuracy, especially when facing large-scale photovoltaic power plants or vast geographical areas. The prediction accuracy of traditional methods is difficult to meet the high requirements of modern photovoltaic power plants for real-time performance and accuracy. Summary of the invention

[0004] The present application provides a photovoltaic power generation prediction system and method based on satellite cloud images, which is used to solve the technical problems that the existing photovoltaic power generation prediction methods have insufficient prediction accuracy under large-scale and complex weather conditions and are difficult to respond to the impact of cloud changes on power generation in real time.

[0005] In a first aspect of the present application, a photovoltaic power prediction system based on satellite cloud images is provided, the system comprising: a satellite cloud image data acquisition module, used to acquire satellite cloud image data of a target area and construct a satellite cloud image data tracking library; a cloud feature extraction module, used to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library using a pre-trained image processing model, the cloud feature data including cloud distribution, cloud height, and cloud speed; a data pairing module, used to extract historical photovoltaic power generation data corresponding to the cloud feature data for pairing, and construct a photovoltaic power generation data set based on cloud features; a power prediction model training module, used to use the photovoltaic power generation data set based on cloud features to perform training convergence through a deep learning model to obtain a photovoltaic power generation prediction model; a power generation prediction module, used to integrate and connect the image processing model with the photovoltaic power generation prediction model to construct a power generation prediction model, and deploy the power generation prediction model in a photovoltaic power generation control center for photovoltaic power generation power prediction based on satellite cloud image data.

[0006] The second aspect of the present application provides a photovoltaic power prediction method based on satellite cloud images, the method comprising: obtaining satellite cloud image data of a target area and constructing a satellite cloud image data tracking library; using a pre-trained image processing model to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library, the cloud feature data including cloud distribution, cloud height, and cloud speed; extracting historical photovoltaic power generation data corresponding to the cloud feature data for pairing, and constructing a photovoltaic power generation data set based on cloud features; using the photovoltaic power generation data set based on cloud features to perform training convergence through a deep learning model to obtain a photovoltaic power generation prediction model; integrating and connecting the image processing model and the photovoltaic power generation prediction model to construct a power generation prediction model, and deploying the power generation prediction model in a photovoltaic power generation control center for photovoltaic power generation power prediction based on satellite cloud image data.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The photovoltaic power generation prediction system and method based on satellite cloud images provided in this application relates to the field of data processing technology. The satellite cloud image data acquisition module is used to obtain cloud image data and build a tracking library; the cloud layer features (cloud layer distribution, cloud height, cloud speed) are extracted through the cloud layer feature extraction module; the historical photovoltaic power generation data are paired through the data pairing module to build a data set; the power prediction model training module uses deep learning to train the data set to obtain a prediction model; the power generation prediction module integrates image processing and prediction models, and is deployed in the control center to predict photovoltaic power generation. The technical problem that the existing photovoltaic power generation prediction method has insufficient prediction accuracy under large-scale and complex weather conditions and is difficult to respond to the impact of cloud layer changes on power generation in real time is solved, and the technical effect of improving the accuracy and real-time performance of photovoltaic power generation prediction is achieved through a deep learning model based on satellite cloud image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of the structure of a photovoltaic power generation prediction system based on satellite cloud images provided in an embodiment of the present application;

[0011] Figure 2 A schematic diagram of the flow of a photovoltaic power generation power prediction method based on satellite cloud images provided in an embodiment of the present application.

[0012] Explanation of the reference numerals: satellite cloud image data acquisition module 11, cloud feature extraction module 12, data pairing module 13, power prediction model training module 14, power generation prediction module 15. DETAILED DESCRIPTION

[0013] The present application provides a photovoltaic power generation prediction system and method based on satellite cloud images, which is used to solve the technical problems that the existing photovoltaic power generation prediction methods have insufficient prediction accuracy under large-scale and complex weather conditions and are difficult to respond to the impact of cloud changes on power generation in real time.

[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices.

[0016] Embodiment 1, as Figure 1 As shown, the present application provides a photovoltaic power generation prediction system based on satellite cloud images, the system comprising:

[0017] The satellite cloud image data acquisition module 11 is used to acquire the satellite cloud image data of the target area and build a satellite cloud image data tracking library.

[0018] Specifically, the main function of the satellite cloud image data acquisition module 11 of the present application is to obtain cloud image data of the target area in real time through satellite remote sensing technology, and organize and store it in the satellite cloud image data tracking library to support the subsequent analysis and modeling work of photovoltaic power generation prediction. The workflow of this module can be divided into three core steps: data acquisition, data processing and data storage.

[0019] First, the satellite cloud image data acquisition module regularly obtains satellite cloud image data of the target area by connecting to multiple meteorological satellite platforms, such as NOAA, EUMETSAT, GOES, etc. These satellites have high-frequency and high-resolution remote sensing imaging capabilities, and can monitor cloud changes over a wide range in real time. For example, every hour or every half hour, the satellite will obtain cloud image data for the area, which not only contains spatial distribution information of the clouds, but also includes key information such as cloud height and cloud speed. Cloud height refers to the vertical height of the clouds, which directly affects the degree of shielding of solar radiation, and thus affects the efficiency of photovoltaic power generation; cloud speed refers to the horizontal movement speed of the clouds, which affects the time that the clouds block solar radiation. Cloud distribution describes the coverage area of ​​the clouds on the ground and determines the changes in radiation intensity in the area.

[0020] After data collection is completed, the module will pre-process the satellite cloud image data to ensure its quality and timeliness. This includes image denoising, geometric correction, and data synchronization to ensure that the data meets the requirements of subsequent analysis. Next, the module uses image processing algorithms to extract features from each cloud image, identifying and extracting key feature information such as cloud distribution, cloud height, and cloud speed. These extracted features will be combined with photovoltaic power generation data to form a complete cloud image data set to provide support for subsequent deep learning model training.

[0021] Finally, the satellite cloud image data after preprocessing and feature extraction will be stored in the satellite cloud image data tracking library. This library not only includes the original satellite cloud image data, but also records the timestamp, geographic location and other information corresponding to each image to ensure the temporal and spatial consistency of the data. By building this data tracking library, the system can achieve long-term tracking and monitoring of cloud changes in the target area, providing accurate input data for subsequent photovoltaic power generation prediction.

[0022] The cloud feature extraction module 12 is used to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library using a pre-trained image processing model. The cloud feature data includes cloud distribution, cloud height, and cloud speed.

[0023] Furthermore, the cloud feature extraction module 12 is further configured to perform the following steps:

[0024] P21: Extract historical satellite cloud image data from the satellite cloud image data tracking library, including multi-weather type and multi-season data samples, and preprocess the data samples, including denoising, contrast enhancement, and cropping; P22: Identify cloud borders and cloud height features of the data samples, and construct training sample sets and test sample sets; P23: Train and test the convolutional network model through the training sample sets and test sample sets to achieve the convergence target and obtain an image recognition layer; P24: Calculate the cloud movement speed according to the time difference of the data samples to obtain a cloud speed processing layer; P25: Connect and fuse the image recognition layer with the cloud speed processing layer to construct the image processing model.

[0025] It should be understood that the cloud feature extraction module 12 of the present application is responsible for using the pre-trained image processing model to perform in-depth analysis on the satellite cloud image data and extract key information therefrom, such as cloud distribution, cloud height, cloud speed, etc. These cloud features are crucial for photovoltaic power generation prediction.

[0026] First, historical satellite cloud image data is extracted from the satellite cloud image data tracking library, which covers a variety of weather types and different seasons. By extracting representative sample data, the system can obtain rich scene information to ensure that the model can adapt to different meteorological changes. After the data is extracted, the module performs a series of preprocessing operations on these data samples for subsequent image processing and feature extraction. The preprocessing steps include denoising, contrast enhancement, and cropping. Denoising removes noise and interference information in the image to improve data quality; by enhancing the contrast of the image, the cloud features are more obvious and easy to identify; the image is cropped to the target area, unnecessary background areas are removed, and the focus is on the cloud information in the satellite image.

[0027] After completing data preprocessing, the module identifies cloud features for the processed data samples, mainly including the feature identification of cloud borders and cloud heights. Cloud borders are used to identify the spatial distribution of clouds, while cloud heights are used to describe the vertical position of clouds and their impact on solar radiation. By marking these features, the system can clearly identify the cloud area in the image and its related features. In addition, the identification results will be used to construct training sample sets and test sample sets, which will be used for subsequent training and testing of convolutional neural network (CNN) models. The quality of the training sample sets and test sample sets directly affects the accuracy and robustness of the model.

[0028] After the cloud features are identified, the module will use the constructed training sample set and test sample set to train and test the convolutional neural network (CNN) model. Convolutional neural network is a powerful deep learning model that is particularly suitable for image recognition tasks. During the training phase, the CNN model continuously adjusts the network weights so that the image recognition layer (i.e. the primary layer of the network) can accurately identify the cloud features in the image. This process is achieved through the back propagation algorithm and gradient descent method until the model reaches a convergence state and can accurately identify the characteristics of the cloud layer on new data. The testing phase evaluates the performance of the model on unknown data to ensure its generalization ability.

[0029] After completing the cloud feature identification, the next step is to calculate the movement speed of the clouds (cloud speed). Cloud speed refers to the speed at which clouds move on the time axis, which affects the duration of cloud blocking of solar radiation. In order to calculate cloud speed, the system analyzes continuous satellite image data based on the time difference of different data samples. This process is achieved through image comparison and feature tracking technology to determine the rate of change of clouds from one location to another. The calculation of cloud speed is crucial for the prediction of photovoltaic power generation, because the movement of clouds directly affects the changes in solar radiation in a certain area.

[0030] Finally, the cloud feature extraction module integrates the image recognition layer with the cloud speed processing layer to construct the final image processing model. The image recognition layer extracts the spatial distribution and cloud height features of the cloud layer from the image through a convolutional neural network, while the cloud speed processing layer calculates the moving speed of the cloud layer based on the time difference. The combination of these two parts of information can provide more accurate input data for photovoltaic power generation prediction. The final image processing model can simultaneously consider the spatial characteristics and dynamic changes of the cloud layer, thereby achieving more accurate photovoltaic power generation prediction.

[0031] Furthermore, the cloud layer movement speed calculation is performed according to the time difference of the data samples to obtain the cloud speed processing layer. Step P24 of the embodiment of the present application also includes:

[0032] P24-1: Obtain two or more consecutive cloud image frames for pixel comparison, and calculate the intensity change of each pixel at different time points; P24-2: Obtain the change difference based on the intensity change of each pixel at different time points, and obtain the cloud speed and direction based on the change difference.

[0033] Alternatively, cloud speed directly affects the change in solar radiation, which in turn affects the prediction of photovoltaic power generation. Therefore, accurately calculating the movement speed and direction of clouds is crucial to improving the accuracy of prediction.

[0034] First, the cloud feature extraction module extracts two or more consecutive frames of cloud image data from the satellite cloud image data tracking library. Each frame of cloud image represents the cloud distribution information at a certain point in time. By comparing the pixels of consecutive frames of cloud images, the module calculates the intensity change of each pixel at different time points. In this process, the system analyzes the distribution and changes of the cloud layer in the image and records the intensity value of each pixel as a representation of the cloud layer.

[0035] Next, the module calculates the difference in intensity based on the intensity change of each pixel at different time points. The difference in intensity reflects the degree of movement of the clouds at different time points, which can be used to infer the speed and direction of movement of the clouds. By comparing the changes in pixel intensity of consecutive frame images, the displacement of the clouds can be determined. Based on these displacement data, the system can accurately calculate the movement speed of each pixel point and infer the direction of movement of the clouds based on the direction of displacement.

[0036] Through the above operations, the cloud speed processing layer is constructed, which provides time-varying information about cloud movement. This information is crucial for photovoltaic power generation prediction, because the speed and direction of cloud movement determine the rate of change of solar radiation, which directly affects the change of power generation. Finally, the cloud speed processing layer is integrated with the image recognition layer (cloud distribution and cloud height information extracted through the convolutional network model) to form a complete cloud feature extraction model, providing accurate input data for subsequent photovoltaic power generation prediction.

[0037] Furthermore, step P25 of the embodiment of the present application also includes:

[0038] P25-1: Concatenate the cloud distribution and cloud height feature vectors output by the image recognition layer with the cloud speed features output by the cloud speed processing layer; P25-2: Input the concatenated features into the fully connected layer, and output the image processing results, including cloud distribution, cloud height, and cloud speed, through the output layer task setting.

[0039] Specifically, the outputs of the image recognition layer and the cloud speed processing layer are combined and further processed through a fully connected layer to finally generate the required cloud feature output.

[0040] First, the module splices the output features of the image recognition layer (i.e., the feature vectors of cloud distribution and cloud height) with the output features of the cloud speed processing layer (i.e., cloud speed and direction). The output of the image recognition layer mainly includes the spatial distribution information and cloud height features of the clouds, while the output of the cloud speed processing layer provides the moving speed and direction of the clouds. These two outputs complement each other in information, and through splicing, a more comprehensive and detailed cloud feature vector can be formed, which contains multi-dimensional information such as cloud distribution, height, speed, and moving direction. This splicing step provides richer feature input for the subsequent fully connected layer.

[0041] Next, the concatenated feature vectors will be input into the fully connected layer for further nonlinear transformation and feature fusion. The fully connected layer is responsible for fusing multiple input features into a unified output and performing information processing according to task requirements. Through the processing of the fully connected layer, the system will output the final results such as cloud distribution, cloud height and cloud speed. These output results will be used as the final output of the image processing model for subsequent photovoltaic power generation prediction to provide accurate cloud feature information.

[0042] Through this step, the system can combine the spatial distribution, height changes and movement speed of clouds to comprehensively reflect the impact of clouds on solar radiation and provide strong support for photovoltaic power generation prediction.

[0043] Furthermore, step P25-1 of the embodiment of the present application also includes:

[0044] P25-11: Temporally and spatially align the input cloud map samples of the image recognition layer with the input cloud map samples of the cloud speed processing layer; P25-12: Based on the temporal and spatial alignment, perform dimension-upgrading and projection of the output results of the image recognition layer to the cloud speed feature variable dimension for feature splicing.

[0045] Optionally, further refine the splicing process of the image recognition layer and the cloud speed processing layer outputs to ensure that the input data is aligned in time and space, and to project the feature dimensions up.

[0046] First, the input cloud map samples of the image recognition layer and the input cloud map samples of the cloud speed processing layer need to be aligned in time and space. Since the calculation of cloud speed involves multiple frames of continuous cloud maps, and the image recognition layer processes single or multiple frames of cloud map data, it is necessary to ensure that the two correspond in time and space coordinates before stitching. Time alignment means that the two must come from the same time window to ensure that the state of the cloud layer is synchronized in time; spatial alignment ensures that the positions of clouds in different frames of cloud maps are consistent in geographic space. This step is crucial to the accuracy of subsequent feature stitching, avoiding errors caused by time and space mismatches.

[0047] After completing the time and space alignment, the output of the image recognition layer needs to be projected to match the feature dimension of the cloud speed processing layer output. The output of the image recognition layer usually contains cloud distribution and cloud height information, while the output of the cloud speed processing layer contains cloud speed (moving speed) and direction. Due to the different feature dimensions of the two, in order to effectively splice the two, the output of the image recognition layer must first be upgraded, that is, through a specific projection operation, the cloud distribution and cloud height information is expanded to be consistent with the cloud speed feature dimension. After completing the dimensionality upgrade, the module splices the two feature sets together to form a comprehensive cloud feature vector, which contains multi-dimensional information such as the spatial distribution of clouds, cloud height, moving speed and direction.

[0048] This process ensures that feature data from different sources can be effectively combined in the same spatial and temporal dimensions, providing accurate and comprehensive data support for subsequent feature fusion and fully connected layer processing.

[0049] The data pairing module 13 is used to extract the historical photovoltaic power generation data corresponding to the cloud layer characteristic data for pairing, and construct a photovoltaic power generation data set based on cloud layer characteristics.

[0050] Specifically, the main task of the data pairing module 13 of the present application is to match and pair the cloud features in the satellite cloud image data with the historical photovoltaic power generation data in time, and then construct a photovoltaic power generation data set based on cloud features. This data set is the basis for training the photovoltaic power generation power prediction model, so its construction process must ensure the accuracy and timeliness of the data.

[0051] First, it is necessary to ensure that the cloud features in the satellite cloud image data are accurately matched with the historical photovoltaic power generation data in time. Since the satellite cloud image data provides features such as cloud distribution, cloud height, and cloud speed, and these features are obtained in time series, it is necessary to align these timestamps with the timestamps in the historical photovoltaic power generation data. Photovoltaic power generation data is usually recorded in hours, days, or months. The module needs to extract the power generation data of the corresponding time period from the historical photovoltaic power generation data according to the time granularity of the cloud image data.

[0052] Once the time alignment is completed, the module will pair the cloud features (such as cloud distribution, cloud height, and cloud speed) with the photovoltaic power generation data of the corresponding time period to construct a photovoltaic power generation data set based on cloud features. This data set contains multiple samples, each of which includes the cloud features at a time point and the photovoltaic power generation at that time point. By pairing these data, the system can capture the relationship between cloud features and photovoltaic power generation, providing training data for subsequent deep learning models.

[0053] During the data pairing process, the module will also perform certain data cleaning and processing, such as removing missing values, outlier processing, data standardization, etc. These processing steps ensure that the constructed data set meets the training requirements in terms of quality and consistency, and avoid data problems affecting the model training effect.

[0054] Through this process, the data pairing module provides a high-quality photovoltaic power generation dataset based on cloud characteristics for subsequent deep learning model training, ensuring that the prediction model can accurately reflect the relationship between photovoltaic power generation and cloud changes.

[0055] The power prediction model training module 14 is used to use the photovoltaic power generation data set based on cloud layer characteristics to perform training convergence through a deep learning model to obtain a photovoltaic power generation power prediction model.

[0056] Furthermore, the power prediction model training module 14 is also used to perform the following steps:

[0057] P41: Setting cloud feature training screening conditions, wherein the cloud feature training screening conditions are that the photovoltaic control parameters remain unchanged; P42: Based on the cloud feature training screening conditions, sample screening is performed on the photovoltaic power generation data set based on cloud features, and a cloud feature prediction sample set is constructed; P43: The cloud feature prediction sample set is segmented according to a preset segmentation ratio, and a cloud prediction training set and a cloud prediction test set are constructed; P44: The deep learning model is trained through the cloud prediction training set, and the trained deep learning model is tested for convergence through the cloud prediction test set until the convergence requirements are met, thereby obtaining the photovoltaic power generation power prediction model.

[0058] It should be understood that the main task of the power prediction model training module 14 of the present application is to use the photovoltaic power generation data set based on cloud characteristics, train and optimize through a deep learning model, and finally obtain a photovoltaic power generation power prediction model.

[0059] During the training process, a set of training screening conditions must first be set to ensure the stability and validity of the data. Specifically, the cloud feature training screening conditions require that the photovoltaic control parameters remain unchanged during the training process. Photovoltaic control parameters include the installation angle, direction, and module efficiency of photovoltaic modules. They will not change significantly in the short term, so they should be considered constants during model training to avoid interference with the prediction results.

[0060] According to the set cloud feature training screening conditions, the module performs sample screening on the photovoltaic power generation data set based on cloud features. In this process, the system screens samples that meet the training conditions from the original data set, and these samples will be used to build a cloud feature prediction sample set. Through screening, it is ensured that the samples in the data set have a strong correlation between cloud features (such as cloud distribution, cloud height, cloud speed) and photovoltaic power generation data, reducing the impact of noise data and ensuring the quality of training data.

[0061] Once the cloud feature prediction sample set is built, the module will then split the sample set according to the preset ratio. Usually, the data set is divided into a training set and a test set to ensure that the training and evaluation processes of the model are independent. Specifically, the module separates the cloud prediction training set and the cloud prediction test set according to the preset split ratio (for example, 80% of the data is used for training and 20% of the data is used for testing). The training set is used to train the model, and the test set is used to verify the generalization ability of the model and evaluate its performance.

[0062] After completing the data set segmentation, the module uses the cloud prediction training set to train the deep learning model. The deep learning model will perform multiple rounds of iterative training based on the input cloud feature data (such as cloud distribution, cloud height, cloud speed) and the corresponding photovoltaic power generation data, adjust the model parameters, and optimize the prediction effect. During the training process, the model will continuously learn the relationship between the input features and the output results, and gradually improve the prediction accuracy.

[0063] During the training process, the module will also use the cloud layer prediction test set to evaluate the trained model. The test set is used to verify the performance of the model on unseen data. Through continuous training and testing, the model will gradually converge until it reaches the preset convergence standard (such as the loss function value is lower than a certain threshold, or the prediction accuracy of the model reaches a certain standard).

[0064] Finally, after the deep learning model is fully trained and tested for convergence, an accurate photovoltaic power generation prediction model is obtained. This model can accurately predict the power generation of photovoltaic power generation systems under different meteorological conditions based on the cloud characteristics in real-time satellite cloud image data, thereby providing support for the operation optimization, load scheduling and maintenance management of photovoltaic power stations.

[0065] Furthermore, step P44 of the embodiment of the present application also includes:

[0066] P44-1: Obtain an initial generation power generation prediction model, which is obtained through deep learning based on meteorological data and photovoltaic control parameter samples; P44-2: Take the initial generation power generation prediction model as the basic model, and perform incremental learning on the basic model based on the cloud prediction training set and the cloud prediction test set to obtain the photovoltaic power generation prediction model.

[0067] In a possible embodiment of the present application, the incremental learning process of the model may be further refined to improve the accuracy and robustness of the model.

[0068] At the beginning of the training process, the module first obtains a first-generation power generation prediction model through deep learning training based on meteorological data and photovoltaic control parameter samples. The training of this model usually includes historical meteorological data of photovoltaic power plants (such as temperature, humidity, radiation intensity, etc.) and photovoltaic control parameters (such as component efficiency, tilt angle, installation direction, etc.). The first-generation model obtained in this step is built based on traditional meteorological and control parameter data, which can provide a preliminary estimate for the prediction of photovoltaic power generation.

[0069] After obtaining the initial generation power prediction model, the module uses it as the basic model and further optimizes the model through incremental learning. The incremental learning process updates the basic model based on the cloud prediction training set and the cloud prediction test set. By introducing cloud characteristics (such as cloud distribution, cloud height, cloud speed, etc.) into the model training process, the system can better capture the relationship between photovoltaic power generation and cloud changes, thereby improving the prediction accuracy.

[0070] During the incremental learning process, the basic model will continuously adjust its parameters according to the new cloud feature data to optimize its ability to predict photovoltaic power generation. Through this incremental learning, the model can gradually adapt to different meteorological and cloud change conditions and improve its prediction ability in different weather scenarios. Ultimately, after sufficient training and convergence, the module will obtain an efficient and accurate photovoltaic power prediction model that can predict future photovoltaic power generation based on real-time cloud map data and historical power generation data.

[0071] The power generation prediction module 15 is used to integrate and connect the image processing model and the photovoltaic power generation prediction model to construct a power generation prediction model, and deploy the power generation prediction model in the photovoltaic power generation control center for photovoltaic power generation prediction based on satellite cloud image data.

[0072] Optionally, the main task of the power generation prediction module 15 of the present application is to integrate and connect the image processing model with the photovoltaic power generation prediction model, build a complete power generation prediction model, and deploy it in the photovoltaic power generation control center to perform photovoltaic power generation prediction in real time based on satellite cloud map data.

[0073] First, the power generation prediction module 15 integrates the previously trained image processing model and the photovoltaic power generation prediction model. The image processing model extracts the characteristics of cloud distribution, cloud height, cloud speed, etc. by analyzing the satellite cloud image data, while the photovoltaic power generation prediction model predicts the photovoltaic power generation based on these cloud characteristics. The combination of the two forms an end-to-end prediction system. Through this integration, the system can realize a complete prediction chain from satellite cloud image data to photovoltaic power generation.

[0074] The integrated model can use cloud characteristics (such as cloud distribution, cloud height, cloud speed, etc.) as input, and process them through a deep learning model, and finally output the power generation of the photovoltaic power station at a certain point in the future. This prediction model can comprehensively consider multiple factors, including cloud dynamics, solar radiation changes, etc., to improve the accuracy and real-time nature of power generation prediction. Through continuous optimization and updating, the system can gradually improve the accuracy of predictions and adapt to the needs of power generation predictions under different meteorological conditions.

[0075] After completing model integration and optimization, the power generation prediction model will be deployed to the photovoltaic power generation control center. The control center receives satellite cloud image data in real time and uses the deployed power generation prediction model to quickly predict photovoltaic power generation. This process can help operators understand the power generation status of photovoltaic power stations in real time and perform reasonable scheduling and optimization. For example, the control center can predict the impact of cloud changes on solar radiation in advance, and then adjust the operation strategy of the photovoltaic power station to maximize power generation benefits.

[0076] Through the application of power generation prediction models, the photovoltaic power generation control center can obtain the power generation forecast for a period of time in real time, perform power prediction, load adjustment and other operations, thereby improving the power generation efficiency and stability of the power station. The system's real-time prediction capability also provides effective decision-making support for responding to sudden meteorological changes (such as sudden changes in clouds, heavy rain, etc.), helping operation and maintenance personnel to respond promptly under abnormal weather conditions.

[0077] In summary, the embodiments of the present application have at least the following technical effects:

[0078] This application obtains satellite cloud image data and extracts cloud features, combines it with historical photovoltaic power generation data of the corresponding time, constructs a power generation data set based on cloud features, trains the data set with a deep learning model, obtains a photovoltaic power generation power prediction model, and integrates the image processing model with the power prediction model, which is deployed in the photovoltaic power generation control center to achieve real-time power generation prediction based on satellite cloud image data.

[0079] The technical effect of improving the accuracy and real-time performance of photovoltaic power prediction has been achieved through a deep learning model based on satellite cloud image data.

[0080] Embodiment 2, based on the same inventive concept as the photovoltaic power generation power prediction method based on satellite cloud images in the above embodiment, Figure 2 As shown, the present application provides a photovoltaic power generation prediction method based on satellite cloud images. The method in the embodiment of the present application and the system embodiment are based on the same inventive concept. The method includes:

[0081] Acquire satellite cloud image data of the target area and build a satellite cloud image data tracking library; use a pre-trained image processing model to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library, where the cloud feature data includes cloud distribution, cloud height, and cloud speed; extract and pair historical photovoltaic power generation data corresponding to the cloud feature data to build a photovoltaic power generation data set based on cloud features; use the photovoltaic power generation data set based on cloud features to perform training and convergence through a deep learning model to obtain a photovoltaic power generation prediction model; integrate and connect the image processing model and the photovoltaic power generation prediction model to build a power generation prediction model, and deploy the power generation prediction model in a photovoltaic power generation control center for photovoltaic power generation power prediction based on satellite cloud image data.

[0082] Furthermore, the use of the pre-trained image processing model to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library includes:

[0083] Historical satellite cloud image data are extracted from the satellite cloud image data tracking library, including data samples of multiple weather types and multiple seasons, and the data samples are preprocessed, and the preprocessing includes denoising, contrast enhancement, and cropping; cloud borders and cloud height features are marked for the data samples, and a training sample set and a test sample set are constructed; convolutional network model training and testing are performed using the training sample set and the test sample set to achieve a convergence target and obtain an image recognition layer; cloud movement speed calculation is performed according to the time difference of the data samples to obtain a cloud speed processing layer; the image recognition layer is connected and fused with the cloud speed processing layer to construct the image processing model.

[0084] Furthermore, the cloud movement speed calculation is performed according to the time difference of the data samples to obtain the cloud speed processing layer, including: obtaining two or more consecutive frames of cloud images for pixel comparison, and calculating the intensity change of each pixel at different time points; according to the intensity change of each pixel at different time points, the change difference is obtained, and the cloud speed and direction are obtained based on the change difference.

[0085] Furthermore, the image recognition layer is connected and fused with the cloud speed processing layer to construct the image processing model, including:

[0086] The output cloud layer distribution and cloud height feature vectors of the image recognition layer are spliced ​​with the output cloud speed features of the cloud speed processing layer; the spliced ​​features are input into the fully connected layer, and the image processing results, including cloud layer distribution, cloud height and cloud speed, are output through the output layer task setting.

[0087] Furthermore, the output cloud layer distribution and cloud height feature vector of the image recognition layer are spliced ​​with the output cloud speed feature of the cloud speed processing layer, including:

[0088] The input cloud map samples of the image recognition layer are aligned in time and space with the input cloud map samples of the cloud speed processing layer; based on the time and space alignment, the output results of the image recognition layer are projected to the cloud speed feature variable dimension for feature splicing.

[0089] Furthermore, the photovoltaic power generation data set based on cloud layer features is used to perform training convergence through a deep learning model to obtain a photovoltaic power generation prediction model, including: setting cloud layer feature training screening conditions, wherein the cloud layer feature training screening conditions are that the photovoltaic control parameters remain unchanged; performing sample screening on the photovoltaic power generation data set based on cloud layer features based on the cloud layer feature training screening conditions to construct a cloud layer feature prediction sample set; segmenting the cloud layer feature prediction sample set according to a preset segmentation ratio to construct a cloud layer prediction training set and a cloud layer prediction test set; training the deep learning model through the cloud layer prediction training set, and testing and converging the trained deep learning model through the cloud layer prediction test set until the convergence requirements are met to obtain the photovoltaic power generation prediction model.

[0090] Furthermore, obtaining the photovoltaic power prediction model also includes: obtaining an initial generation power prediction model, wherein the initial generation power prediction model is obtained through deep learning based on meteorological data and photovoltaic control parameter samples; using the initial generation power prediction model as a basic model, and performing incremental learning on the basic model based on the cloud prediction training set and the cloud prediction test set to obtain the photovoltaic power prediction model.

[0091] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0092] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0093] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. Photovoltaic power generation prediction system based on satellite cloud images, characterized in that: include: Satellite cloud image data acquisition module, used to obtain satellite cloud image data of the target area and build a satellite cloud image data tracking library; A cloud feature extraction module is used to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library using a pre-trained image processing model, where the cloud feature data includes cloud distribution, cloud height, and cloud speed; The data pairing module is used to extract the historical photovoltaic power generation data corresponding to the cloud layer characteristic data and pair them to construct a photovoltaic power generation data set based on cloud layer characteristics; A power prediction model training module is used to use the photovoltaic power generation data set based on cloud features to perform training convergence through a deep learning model to obtain a photovoltaic power generation power prediction model; The power generation prediction module is used to integrate and connect the image processing model with the photovoltaic power generation prediction model to construct a power generation prediction model, and deploy the power generation prediction model in the photovoltaic power generation control center for photovoltaic power generation prediction based on satellite cloud image data.

2. The photovoltaic power generation prediction system based on satellite cloud images according to claim 1, characterized in that: The cloud feature extraction module is also used for: Extracting historical satellite cloud image data from the satellite cloud image data tracking library, including multi-weather type and multi-season data samples, and preprocessing the data samples, the preprocessing including denoising, contrast enhancement, and cropping; Marking cloud borders and cloud height features of the data samples, and constructing a training sample set and a test sample set; The convolutional network model is trained and tested by using the training sample set and the test sample set to achieve a convergence target and obtain an image recognition layer; The cloud movement speed is calculated according to the time difference of the data samples to obtain the cloud speed processing layer; The image recognition layer is connected and fused with the cloud speed processing layer to construct the image processing model.

3. The photovoltaic power generation prediction system based on satellite cloud images as claimed in claim 2, characterized in that: The cloud feature extraction module is also used for: Obtain two or more consecutive cloud images for pixel comparison and calculate the intensity change of each pixel at different time points; According to the intensity change of each pixel at different time points, the change difference is obtained, and the cloud speed and direction are obtained based on the change difference.

4. The photovoltaic power generation prediction system based on satellite cloud images as claimed in claim 2, characterized in that: The cloud feature extraction module is also used for: splicing the output cloud layer distribution and cloud height feature vector of the image recognition layer with the output cloud speed feature of the cloud speed processing layer; The concatenated features are input into the fully connected layer, and the image processing results, including cloud distribution, cloud height, and cloud speed, are output through the output layer task setting.

5. The photovoltaic power generation prediction system based on satellite cloud images as claimed in claim 4, characterized in that: The cloud feature extraction module is also used for: Aligning the input cloud map samples of the image recognition layer with the input cloud map samples of the cloud speed processing layer in time and space; Based on the time and space alignment, the output result of the image recognition layer is projected to the cloud speed feature variable dimension for feature splicing.

6. The photovoltaic power generation prediction system based on satellite cloud images according to claim 1, characterized in that: The power prediction model training module is also used for: Setting cloud layer feature training screening conditions, wherein the cloud layer feature training screening conditions are that photovoltaic control parameters remain unchanged; Based on the cloud feature training screening condition, sample screening is performed on the photovoltaic power generation data set based on cloud features to construct a cloud feature prediction sample set; Segment the cloud feature prediction sample set according to a preset segmentation ratio to construct a cloud prediction training set and a cloud prediction test set; The deep learning model is trained using the cloud layer prediction training set, and the trained deep learning model is tested for convergence using the cloud layer prediction test set until convergence requirements are met, thereby obtaining the photovoltaic power generation prediction model.

7. The photovoltaic power generation prediction system based on satellite cloud images as claimed in claim 6, characterized in that: The power prediction model training module is also used for: Obtaining a primary generation power prediction model, wherein the primary generation power prediction model is obtained through deep learning based on meteorological data and photovoltaic control parameter samples; The initial generation power generation prediction model is used as a basic model, and incremental learning is performed on the basic model based on the cloud layer prediction training set and the cloud layer prediction test set to obtain the photovoltaic power generation prediction model.

8. A photovoltaic power generation prediction method based on satellite cloud images, characterized in that: The method comprises: Obtain satellite cloud image data of the target area and build a satellite cloud image data tracking library; Using a pre-trained image processing model to extract cloud features from the satellite cloud image data in the satellite cloud image data tracking library, the cloud feature data including cloud distribution, cloud height, and cloud speed; The historical photovoltaic power generation data corresponding to the cloud feature data are extracted and paired to construct a photovoltaic power generation data set based on cloud features; The photovoltaic power generation data set based on cloud features is used to train and converge a deep learning model to obtain a photovoltaic power generation power prediction model; The image processing model is integrated and connected with the photovoltaic power generation prediction model to construct a power generation prediction model, and the power generation prediction model is deployed in a photovoltaic power generation control center for photovoltaic power generation prediction based on satellite cloud image data.

Images