Power generation prediction method and system based on ground cloud data

Abstract

This invention discloses a method and system for predicting power generation based on ground-based cloud image data, belonging to the field of power generation prediction technology. The method includes: collecting ground-based cloud image data streams from power plants, and simultaneously deploying irradiance measurement devices to collect solar irradiance data from the power plants; extracting features from the ground-based cloud image data streams to obtain a key feature set of the cloud images; performing correlation fitting based on the key feature set of the cloud images and the solar irradiance data from the power plants to establish a cloud image feature-irradiance correlation model; integrating the cloud image feature-irradiance correlation model into a power generation prediction model to build a cloud image feature-irradiance-power generation prediction model; performing power generation prediction, and outputting the target power generation prediction value. This solves the technical problem in existing technologies where power generation prediction based on traditional meteorological data does not respond promptly to rapid changes in cloud cover, resulting in insufficient accuracy in power generation prediction, and achieves the technical effect of improving the accuracy of power generation prediction.

Description

Technical Field

[0001] This invention relates to the field of power generation prediction technology, and specifically to a method and system for power generation prediction based on ground-based cloud map data. Background Technology

[0002] With the continuous expansion of photovoltaic (PV) power generation capacity in the power system, the accuracy of power generation prediction has become a crucial factor affecting grid security dispatch, power balance, and the capacity for renewable energy absorption. Existing PV power generation prediction methods typically rely on numerical weather prediction data, historical irradiance data, or statistical regression models to predict future power generation by analyzing the mapping relationship between meteorological elements and power generation. However, these methods primarily focus on large-scale, low-temporal-resolution meteorological information, failing to adequately respond to short-term solar irradiance fluctuations caused by the rapid movement, formation, or dissipation of local cloud clusters, and struggling to accurately depict the transient impact of cloud cover changes on the incident irradiance of PV arrays. In scenarios with frequent cloud cover changes or complex weather conditions, this can easily lead to increased prediction deviations, limiting the accuracy and stability of the prediction results and making it difficult to meet the grid's application requirements for short-term, high-precision power generation prediction. Summary of the Invention

[0003] This application provides a method and system for predicting power generation based on ground-based cloud image data, which solves the technical problem that existing power generation prediction based on traditional meteorological data does not respond promptly to rapid changes in cloud cover, resulting in insufficient accuracy of power generation prediction.

[0004] A first aspect of this application provides a method for predicting power generation based on ground-based cloud map data, the method comprising: An all-sky imager is deployed at the target power station to collect ground-based cloud image data streams. Simultaneously, an irradiance measurement device is deployed to collect solar irradiance data. Feature extraction is performed on the ground-based cloud image data streams to obtain a key feature set. Based on the key feature set and the solar irradiance data, a cloud image feature-irradiance correlation model is established. A power generation prediction model is pre-established, and the cloud image feature-irradiance correlation model is integrated into the power generation prediction model for fusion, thus constructing a cloud image feature-irradiance-power generation prediction model. Power generation is predicted based on this model, and the target power generation prediction value is output.

[0005] A second aspect of this application provides a power generation prediction system based on ground-based cloud map data, the system comprising: Data Acquisition Module: Deploys an all-sky imager at the target site to acquire ground-based cloud image data streams, and simultaneously deploys an irradiance measurement device to acquire solar irradiance data at the site; Fitting Module: Extracts features from the ground-based cloud image data streams to obtain a key feature set of the cloud image, and performs correlation fitting based on the key feature set of the cloud image and the solar irradiance data at the site to establish a cloud image feature-irradiance correlation model; Fusion Module: Pre-establishes a power generation prediction model, and introduces the cloud image feature-irradiance correlation model into the power generation prediction model for fusion to build a cloud image feature-irradiance-power generation prediction model; Prediction Module: Predicts power generation based on the cloud image feature-irradiance-power generation prediction model and outputs the target power generation prediction value.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, an all-sky imager is deployed at the target power station to collect ground-based cloud image data streams. Simultaneously, an irradiance measurement device is deployed to collect solar irradiance data. Then, features are extracted from the ground-based cloud image data streams to obtain a key feature set. Based on this key feature set and the station's solar irradiance data, a correlation fitting is performed to establish a cloud image feature-irradiance correlation model. Further, a power generation prediction model is pre-established, and the cloud image feature-irradiance correlation model is integrated into the power generation prediction model for fusion, constructing a cloud image feature-irradiance-power generation prediction model. Finally, power generation is predicted based on the cloud image feature-irradiance-power generation prediction model, outputting the target power generation prediction value. This solves the technical problem in existing technologies where power generation prediction based on traditional meteorological data does not respond promptly to rapid changes in cloud cover, leading to insufficient prediction accuracy. This achieves the technical effect of improving the accuracy of power generation prediction. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A schematic diagram of the power generation prediction method based on ground-based cloud map data provided in this application embodiment; Figure 2 A schematic diagram of the power generation prediction system based on ground-based cloud map data provided in this application embodiment.

[0009] Figure labeling: Data acquisition module 11, Fitting module 12, Fusion module 13, Prediction module 14. Detailed Implementation

[0010] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0011] Example 1, as Figure 1 As shown, this application provides a method for predicting power generation based on ground-based cloud map data, wherein the method includes: An all-sky imager is deployed at the target site to collect ground-based cloud image data streams, while an irradiance measurement device is deployed to collect solar irradiance data at the site.

[0012] An all-sky imager is deployed within the target power station. Installed in an unobstructed, open area, the imager's field of view covers the entire sky above the station and continuously captures cloud conditions using a fisheye lens. The imager collects cloud images over the station according to a preset sampling period, forming a ground-based cloud image data stream with timestamps. Simultaneously, an irradiance measurement device is deployed within the target power station, positioned to match the solar irradiance conditions of the photovoltaic modules. This device synchronously collects the station's total solar irradiance data and performs time stamping and standardization on the collected irradiance data. By synchronizing and storing the ground-based cloud image data stream and the solar irradiance data in a correlated manner, fundamental data support is provided for subsequent cloud image feature extraction, cloud image feature-irradiance correlation modeling, and power generation prediction.

[0013] Feature extraction is performed on the ground-based cloud map data stream of the site to obtain a key feature set of the cloud map. Based on the key feature set of the cloud map and the solar irradiance data of the site, a correlation fitting is performed to establish a cloud map feature-irradiance correlation model.

[0014] First, the acquired ground-based cloud image data stream is preprocessed. This preprocessing includes format standardization, geometric correction, and noise suppression to eliminate the influence of imaging distortion and environmental interference. After preprocessing, cloud regions are identified and segmented in the ground-based cloud image, and foreground region information representing cloud morphology is extracted. Based on this foreground region information, a pre-trained convolutional neural network is used to jointly extract cloud texture features, edge features, and spatial distribution features, forming a key feature set for the cloud image. On this basis, the key feature set of the cloud image is matched and integrated with the solar irradiance data collected concurrently from the field station according to timestamps, constructing a sample data set of cloud image features and irradiance. Using the key features of the cloud image as input variables and the solar irradiance data as output variables, a correlation fitting analysis is performed on the sample data set. Through regression modeling and parameter optimization, a cloud image feature-irradiance correlation model is established to characterize the mapping relationship between changes in cloud image features and changes in solar irradiance, providing a basis for irradiance correction for subsequent power generation prediction.

[0015] Furthermore, the key feature set of the cloud map is obtained, including: A cloud image data preprocessing step is constructed, which includes format conversion, geometric correction, filtering and denoising, and image enhancement. The cloud image data stream of the site foundation is filtered according to the cloud image data preprocessing step to obtain a usable site foundation cloud image data stream. Edge detection and cloud image segmentation are performed on the usable site foundation cloud image data stream to obtain foreground region information of the foundation cloud layer. A convolutional neural network is used to extract cloud layer correlation features from the foreground region information of the foundation cloud layer to obtain a key feature set of the cloud image.

[0016] Preferably, a cloud image data preprocessing step is constructed, which includes format conversion, geometric correction, filtering and denoising, and image enhancement processing of the original ground-based cloud image data. Specifically, the format conversion is used to unify the data format of cloud image data from different acquisition batches; the geometric correction is used to correct imaging distortion introduced by the fisheye lens of the all-sky imager; the filtering and denoising is used to suppress environmental noise and random interference generated during the imaging process; and the image enhancement is used to improve the contrast between cloud areas and background areas.

[0017] The ground-based cloud image data stream is processed sequentially according to the cloud image data preprocessing steps to obtain a usable ground-based cloud image data stream that meets the feature extraction requirements. Based on this, edge detection and cloud image segmentation are performed on the usable ground-based cloud image data stream to identify cloud contour boundaries and distinguish cloud areas from the sky background area, obtaining information about the foreground cloud area. Finally, the foreground cloud area information is input into a pre-constructed and trained convolutional neural network, which automatically extracts the morphological structure, texture distribution, and spatial correlation features of the clouds, outputting a multi-dimensional feature vector to characterize changes in cloud state. This multi-dimensional feature vector constitutes the key feature set of the cloud image, providing feature input for subsequent modeling of the correlation between cloud image features and solar irradiance.

[0018] A convolutional neural network (CNN) is a deep learning model for automatic image feature extraction. It comprises an input layer, at least one convolutional layer, a pooling layer, and a feature output layer, connected sequentially. The input layer receives image data corresponding to the foreground region information of the cloud layer. The convolutional layer performs local perception and weight-sharing convolution operations on the input image using multiple convolutional kernels to extract texture features, edge features, and spatial distribution features in the cloud region. The pooling layer performs dimensionality reduction on the convolutional features, enhancing their translation invariance and reducing computational complexity. The feature output layer converges the features obtained from multiple convolutions and pooling, outputting a multidimensional feature vector characterizing the cloud morphology and its changing properties.

[0019] During the training phase, the convolutional neural network learns parameters based on historical ground-based cloud image samples and iteratively updates the network weights through a backpropagation algorithm, enabling the network to automatically learn the discriminative features between cloud regions and background regions. During the application phase, the information of the ground-based cloud foreground region is input into the trained convolutional neural network, which outputs the corresponding key feature set of cloud images for subsequent correlation modeling of cloud image features and solar irradiance.

[0020] Furthermore, a cloud image feature-irradiance correlation model is established, including: The key feature set of the cloud map and the solar irradiance data of the field station are paired and integrated according to the collection time to obtain cloud map irradiance data sample pairs; each key feature of the cloud map in the cloud map irradiance data sample pair is used as the independent variable and the solar irradiance of the field station is used as the dependent variable to perform multivariate regression fitting to obtain an initial cloud map irradiance fitting model; the performance of the initial cloud map irradiance fitting model is verified and the parameters are optimized to establish a cloud map feature-irradiance correlation model.

[0021] First, the key feature set of the cloud image and the solar irradiance data from the field station are aligned and matched according to their respective acquisition timestamps. Key features of the cloud image and their corresponding solar irradiance data within the same acquisition time or time window are paired and integrated to form cloud image irradiance data sample pairs. These cloud image irradiance data sample pairs reflect the correspondence between changes in cloud state and changes in solar irradiance. Based on this, each key feature of the cloud image in the cloud image irradiance data sample pair is used as an independent variable, and the corresponding solar irradiance from the field station is used as the dependent variable. A multiple regression fitting analysis is performed to jointly model the influence of different key cloud image features on solar irradiance, resulting in an initial cloud image irradiance fitting model that describes the mapping relationship between cloud image features and solar irradiance. Finally, the initial cloud map irradiance fitting model is subjected to performance verification and parameter optimization. By performing fitting error analysis, prediction deviation evaluation and stability test on historical sample data, the model parameters are adjusted and corrected. When the model prediction results meet the preset accuracy requirements, the cloud map feature-irradiance correlation model is determined and established for subsequent irradiance prediction correction and power generation prediction fusion.

[0022] A power generation prediction model is pre-established, and the cloud map feature-irradiance correlation model is introduced into the power generation prediction model for fusion to build a cloud map feature-irradiance-power generation prediction model.

[0023] Furthermore, a pre-established power generation prediction model includes: Collect historical power generation data of the target power station, which includes solar irradiance data, equipment operation data, and corresponding actual power generation data; extract and screen the power impact features of the historical power generation data to obtain a power generation impact feature sample set; use an LSTM network to train and validate the power generation impact feature sample set to pre-establish a power generation prediction model.

[0024] When establishing a power generation prediction model, the first step is to collect a power generation dataset from the target power station within its historical operating cycle. This historical power generation dataset includes at least solar irradiance data, equipment operation data, and actual power generation data at the corresponding time points, all related to the power generation process. The collected historical power generation dataset undergoes data cleaning, outlier removal, and time alignment to form standardized historical sample data for model training. Based on this, power impact features are extracted and screened from the historical power generation dataset. Feature parameters characterizing the trend and dynamic characteristics of power generation changes are extracted from the solar irradiance data and equipment operation data. The degree of influence of different features on power generation is analyzed, and a power generation impact feature sample set is obtained. Finally, the power generation impact feature sample set is input into a pre-constructed long short-term memory network for power generation prediction training. The training results are then evaluated for error and parameters are fine-tuned using validation data. When the model's prediction accuracy meets preset requirements, the power generation prediction model is pre-established and used for subsequent fusion modeling incorporating cloud map features and irradiance correlation information.

[0025] Furthermore, a cloud image feature-irradiance-power generation prediction model is constructed, including: Irradiance prediction is corrected by combining the cloud map feature dataset with the cloud map feature-irradiance correlation model to obtain an irradiance correction dataset; the cloud map feature dataset and the irradiance correction dataset are then introduced into the power generation prediction model for data splicing and fusion to obtain a cloud map feature-corrected irradiance-power generation fusion dataset; prediction training and optimization are performed based on the cloud map feature-corrected irradiance-power generation fusion dataset to build a cloud map feature-irradiance-power generation prediction model.

[0026] In constructing the cloud image feature-irradiance-power generation prediction model, the following steps are first taken: First, combining the cloud image feature dataset, the established cloud image feature-irradiance correlation model is used to predict and correct the solar irradiance at the corresponding time. The original irradiance data is compensated based on changes in cloud image features, resulting in an irradiance correction dataset reflecting the impact of cloud cover. On this basis, the cloud image feature dataset and the irradiance correction dataset are introduced into the pre-established power generation prediction model. Multi-source feature data are aligned and fused according to a unified time window to construct a cloud image feature-corrected irradiance-power generation fusion dataset containing cloud image features, corrected irradiance information, and corresponding power generation data. Finally, the power generation prediction model is trained and optimized based on the cloud image feature-corrected irradiance-power generation fusion dataset. By evaluating the error between the model's prediction results and the actual power generation, the model parameters are iteratively adjusted, ultimately completing the cloud image feature-irradiance-power generation prediction model for outputting the target power generation prediction result.

[0027] Furthermore, the obtained cloud map feature-corrected irradiance-power generation fusion dataset includes: The irradiance correction dataset is introduced into the power generation prediction model to predict power output, resulting in a power generation prediction dataset. The cloud map feature dataset, the irradiance correction dataset, and the power generation prediction dataset are then time-aligned and fused according to a preset time window to obtain a cloud map feature-corrected irradiance-power generation fusion dataset.

[0028] When obtaining the cloud map feature-corrected irradiance-power generation fusion dataset, the irradiance correction dataset is first used as input to the power generation prediction model. Based on the power generation prediction model, the power generation at the corresponding time is predicted to obtain the power generation prediction dataset, which reflects the power generation change trend after considering the influence of cloud map features. On this basis, the cloud map feature dataset, the irradiance correction dataset, and the power generation prediction dataset are time-aligned according to a preset time window. Cloud map feature data, irradiance correction data, and power generation prediction data within the same time window are paired and synchronously integrated, and then spliced ​​and fused according to a unified data organization format to form a cloud map feature-corrected irradiance-power generation fusion dataset containing multi-source feature information, which is used for subsequent training and optimization of the power generation prediction model.

[0029] Furthermore, based on the cloud map feature-corrected irradiance-power generation fusion dataset, prediction training and optimization are performed to build a cloud map feature-irradiance-power generation prediction model, including: Based on the cloud map feature-corrected irradiance-power generation fusion dataset, model prediction training and actual testing and evaluation are performed to obtain the power generation prediction model performance parameters; based on the power generation prediction model performance parameters, a power model optimization strategy is constructed; based on the power model optimization strategy, the model parameters are iteratively optimized to build the cloud map feature-irradiance-power generation prediction model.

[0030] When performing prediction training and optimization based on the cloud map feature-corrected irradiance-power generation fusion dataset, the fusion dataset is first divided into a training dataset and a test dataset. The training dataset is then input into the power generation prediction model for prediction training, enabling the model to learn the joint mapping relationship between cloud map features, corrected irradiance, and power generation. Subsequently, the trained power generation prediction model is tested and evaluated using the test dataset. By comparing and analyzing the power generation prediction results output by the model with the corresponding actual power generation data, the power generation prediction model performance parameters characterizing the model's prediction performance are obtained. Based on this, according to the power generation prediction model performance parameters, the prediction error, stability, and generalization ability of the model under different operating conditions are comprehensively analyzed, and a power model optimization strategy for the power generation prediction model is constructed. This power model optimization strategy is used to determine the direction and method of adjusting the model parameters. Finally, based on the power model optimization strategy, the model parameters of the power generation prediction model are iteratively optimized. Through multiple rounds of training and evaluation, the model parameters are continuously corrected. When the model's prediction performance meets the preset requirements, the cloud map feature-irradiance-power generation prediction model is completed and used to output the target power generation prediction result.

[0031] Based on the cloud map feature-irradiance-power generation prediction model, power generation is predicted, and the target power generation prediction value is output.

[0032] When predicting power generation based on the cloud map feature-irradiance-power generation prediction model, the key feature data of the cloud map, which is collected or updated in real time, along with the corresponding irradiance correction data, are input into the model. The model then calculates and extrapolates the power generation within a preset time window and outputs the predicted power generation result for that time. This predicted power generation result serves as the target power generation prediction value and can be used in applications such as power plant operation monitoring, grid dispatching, and power generation planning.

[0033] In summary, the embodiments of this application have at least the following technical effects: First, an all-sky imager is deployed at the target power station to collect ground-based cloud image data streams. Simultaneously, an irradiance measurement device is deployed to collect solar irradiance data. Then, features are extracted from the ground-based cloud image data streams to obtain a key feature set. Based on this key feature set and the station's solar irradiance data, a correlation fitting is performed to establish a cloud image feature-irradiance correlation model. Further, a power generation prediction model is pre-established, and the cloud image feature-irradiance correlation model is integrated into the power generation prediction model for fusion, constructing a cloud image feature-irradiance-power generation prediction model. Finally, power generation is predicted based on the cloud image feature-irradiance-power generation prediction model, outputting the target power generation prediction value. This solves the technical problem in existing technologies where power generation prediction based on traditional meteorological data does not respond promptly to rapid changes in cloud cover, leading to insufficient prediction accuracy. This achieves the technical effect of improving the accuracy of power generation prediction.

[0034] Example 2, based on the same inventive concept as the power generation prediction method based on ground-based cloud map data in the aforementioned examples, such as... Figure 2 As shown, this application provides a power generation prediction system based on ground-based cloud map data, wherein the system includes: Data acquisition module 11: Deploys an all-sky imager at the target power station to collect ground-based cloud image data streams, and simultaneously deploys an irradiance measurement device to collect solar irradiance data at the power station; Fitting module 12: Extracts features from the ground-based cloud image data streams to obtain a key feature set of the cloud image, and performs correlation fitting based on the key feature set of the cloud image and the solar irradiance data at the power station to establish a cloud image feature-irradiance correlation model; Fusion module 13: Pre-establishes a power generation prediction model, introduces the cloud image feature-irradiance correlation model into the power generation prediction model for fusion, and builds a cloud image feature-irradiance-power generation prediction model; Prediction module 14: Predicts power generation based on the cloud image feature-irradiance-power generation prediction model and outputs the target power generation prediction value.

[0035] Furthermore, the fitting module 12 is used to perform the following method: A cloud image data preprocessing step is constructed, which includes format conversion, geometric correction, filtering and denoising, and image enhancement. The cloud image data stream of the site foundation is filtered according to the cloud image data preprocessing step to obtain a usable site foundation cloud image data stream. Edge detection and cloud image segmentation are performed on the usable site foundation cloud image data stream to obtain foreground region information of the foundation cloud layer. A convolutional neural network is used to extract cloud layer correlation features from the foreground region information of the foundation cloud layer to obtain a key feature set of the cloud image.

[0036] Furthermore, the fitting module 12 is used to perform the following method: The key feature set of the cloud map and the solar irradiance data of the field station are paired and integrated according to the collection time to obtain cloud map irradiance data sample pairs; each key feature of the cloud map in the cloud map irradiance data sample pair is used as the independent variable and the solar irradiance of the field station is used as the dependent variable to perform multivariate regression fitting to obtain an initial cloud map irradiance fitting model; the performance of the initial cloud map irradiance fitting model is verified and the parameters are optimized to establish a cloud map feature-irradiance correlation model.

[0037] Furthermore, the fusion module 13 is used to perform the following methods: Collect historical power generation data of the target power station, which includes solar irradiance data, equipment operation data, and corresponding actual power generation data; extract and screen the power impact features of the historical power generation data to obtain a power generation impact feature sample set; use an LSTM network to train and validate the power generation impact feature sample set to pre-establish a power generation prediction model.

[0038] Furthermore, the fusion module 13 is used to perform the following methods: Irradiance prediction is corrected by combining the cloud map feature dataset with the cloud map feature-irradiance correlation model to obtain an irradiance correction dataset; the cloud map feature dataset and the irradiance correction dataset are then introduced into the power generation prediction model for data splicing and fusion to obtain a cloud map feature-corrected irradiance-power generation fusion dataset; prediction training and optimization are performed based on the cloud map feature-corrected irradiance-power generation fusion dataset to build a cloud map feature-irradiance-power generation prediction model.

[0039] Furthermore, the fusion module 13 is used to perform the following methods: The irradiance correction dataset is introduced into the power generation prediction model to predict power output, resulting in a power generation prediction dataset. The cloud map feature dataset, the irradiance correction dataset, and the power generation prediction dataset are then time-aligned and fused according to a preset time window to obtain a cloud map feature-corrected irradiance-power generation fusion dataset.

[0040] Furthermore, the fusion module 13 is used to perform the following methods: Based on the cloud map feature-corrected irradiance-power generation fusion dataset, model prediction training and actual testing and evaluation are performed to obtain the power generation prediction model performance parameters; based on the power generation prediction model performance parameters, a power model optimization strategy is constructed; based on the power model optimization strategy, the model parameters are iteratively optimized to build the cloud map feature-irradiance-power generation prediction model.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting power generation based on ground-based cloud map data, characterized in that, The method includes: An all-sky imager is deployed at the target site to collect ground-based cloud image data streams. At the same time, an irradiance measurement device is deployed to collect solar irradiance data of the site. Feature extraction is performed on the ground cloud map data stream of the site to obtain a key feature set of the cloud map. Based on the key feature set of the cloud map and the solar irradiance data of the site, a correlation fitting is performed to establish a cloud map feature-irradiance correlation model. A power generation prediction model is pre-established, and the cloud map feature-irradiance correlation model is introduced into the power generation prediction model for fusion to build a cloud map feature-irradiance-power generation prediction model. Based on the cloud map feature-irradiance-power generation prediction model, power generation is predicted, and the target power generation prediction value is output.

2. The power generation prediction method based on ground-based cloud map data as described in claim 1, characterized in that, Obtain the key feature set of the cloud map, including: The cloud image data preprocessing steps include format conversion, geometric correction, filtering and denoising, and image enhancement. The cloud map data stream of the site foundation is filtered according to the cloud map data preprocessing steps to obtain an usable site foundation cloud map data stream; Edge detection and cloud image segmentation are performed on the available site foundation cloud image data stream to obtain foreground area information of the foundation cloud layer; A convolutional neural network is used to extract cloud-related features from the foreground region information of the ground-based cloud layer, thereby obtaining a key feature set of the cloud map.

3. The power generation prediction method based on ground-based cloud map data as described in claim 1, characterized in that, Establish a cloud image feature-irradiance correlation model, including: The key feature set of the cloud map and the solar irradiance data of the station are paired and integrated according to the collection time to obtain cloud map irradiance data sample pairs; Using the key features of each cloud map in the cloud map irradiance data sample pair as independent variables and the solar irradiance of the station as dependent variable, a multiple regression fitting was performed to obtain the initial cloud map irradiance fitting model. The performance of the initial cloud map irradiance fitting model was verified and the parameters were optimized to establish a cloud map feature-irradiance correlation model.

4. The power generation prediction method based on ground-based cloud map data as described in claim 1, characterized in that, A pre-established power generation prediction model includes: Collect historical power generation data of the target power station, which includes solar irradiance data, equipment operation data and corresponding actual power generation data; The historical power generation dataset is subjected to power impact feature extraction, filtering and labeling to obtain a power generation impact feature sample set; The power generation prediction model is pre-established by using an LSTM network to train, validate, and optimize the power generation influence feature sample set.

5. The power generation prediction method based on ground-based cloud map data as described in claim 1, characterized in that, A cloud image feature-irradiance-power generation prediction model is constructed, including: Irradiance prediction is corrected by combining the cloud map feature dataset with the cloud map feature-irradiance correlation model to obtain the irradiance correction dataset; The cloud map feature dataset and the irradiance correction dataset are introduced into the power generation prediction model for data splicing and fusion to obtain the cloud map feature-corrected irradiance-power generation fusion dataset; Based on the cloud map feature-corrected irradiance-power generation fusion dataset, prediction training and optimization are performed to build a cloud map feature-irradiance-power generation prediction model.

6. The power generation prediction method based on ground-based cloud map data as described in claim 5, characterized in that, Obtain a cloud map feature-corrected irradiance-power generation fusion dataset, including: The irradiance correction dataset is introduced into the power generation prediction model to predict power output, thus obtaining the power generation prediction dataset. The cloud map feature dataset, the irradiance correction dataset, and the power generation prediction dataset are time-aligned and spliced ​​together according to a preset time window to obtain a cloud map feature-corrected irradiance-power generation fusion dataset.

7. The power generation prediction method based on ground-based cloud map data as described in claim 5, characterized in that, Based on the aforementioned cloud map feature-corrected irradiance-power generation fusion dataset, prediction training and optimization are performed to build a cloud map feature-irradiance-power generation prediction model, including: Based on the cloud map feature-corrected irradiance-power generation fusion dataset, the model prediction training and actual test evaluation are carried out to obtain the power generation prediction model performance parameters. Based on the performance parameters of the power generation prediction model, a power model optimization strategy is constructed. Based on the power model optimization strategy, the model parameters are iteratively optimized to build a cloud map feature-irradiance-power generation prediction model.

8. A power generation prediction system based on ground-based cloud map data, characterized in that, The system is used to implement the power generation prediction method based on ground-based cloud map data according to any one of claims 1-7, the system comprising: Data acquisition module: Deploy an all-sky imager at the target site to collect ground-based cloud image data streams from the all-sky imager, and simultaneously deploy an irradiance measurement device to collect solar irradiance data from the site; Fitting module: Extracts features from the ground cloud map data stream of the site to obtain a key feature set of the cloud map, and performs correlation fitting based on the key feature set of the cloud map and the solar irradiance data of the site to establish a cloud map feature-irradiance correlation model; Fusion module: A pre-established power generation prediction model is introduced into the cloud map feature-irradiance correlation model and fused to the power generation prediction model to build a cloud map feature-irradiance-power generation prediction model; Prediction module: Based on the cloud map feature-irradiance-power generation prediction model, predicts power generation and outputs the target power generation prediction value.

Images