Offshore photovoltaic short-term power prediction method, device, equipment, medium and product

Abstract

The invention discloses an offshore photovoltaic short-term power prediction method, device and equipment, a medium and a product. The method comprises the steps of obtaining multi-mode original meteorological data of an area where a target photovoltaic power station is located; adopting a combined deep learning model to extract spatial-temporal features of corresponding modals in the original meteorological data; performing space-time alignment and fusion on the space-time features by adopting an attention mechanism to obtain a multi-modal feature vector; and predicting according to the multi-modal feature vector to obtain a short-term power prediction result of the target photovoltaic power station. The feature extraction capability of deep learning is utilized, a combined deep learning model is adopted to process time and space features at the same time, and multi-modal features are effectively aligned and fused in combination with an attention mechanism, so that the limitation of an existing scheme in coping with a complex environment and multi-source heterogeneous data is solved, and the accuracy of the multi-modal feature fusion is improved. Therefore, short-term power prediction of the offshore photovoltaic power station can be carried out more accurately and stably so as to better adapt to actual operation requirements of an offshore photovoltaic system.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of photovoltaic power generation technology, and in particular to a method, device, equipment, medium and product for offshore photovoltaic short-term power prediction. Background Art

[0002] In the field of photovoltaic power forecasting, the most basic technical solution is to use a single type of prediction model (such as autoregressive models, linear regression models, etc.) to predict the power output of photovoltaic power plants. These models typically only utilize a single type of data source (such as historical power output or a single meteorological factor). Due to the lack of support for multi-source meteorological information, the prediction results are often less than ideal when dealing with complex meteorological conditions and nonlinear relationships between multiple variables. Traditional machine learning models (such as decision trees, random forests, support vector machines, and gradient boosting trees) have been introduced for prediction. However, due to the lack of deep modeling capabilities for spatiotemporal characteristics and the inability to naturally process image and time series data, these models still have limited performance in complex scenarios.

[0003] With the development of deep learning technology, existing solutions have begun to use neural network models to predict photovoltaic power. However, these solutions typically rely solely on a single neural network model, making it difficult to fully exploit the correlations between multimodal data. For complex scenarios such as sudden weather changes and volatile ocean environments, photovoltaic power forecasting involves multiple meteorological parameters and historical data, each of which has complex interactions. Existing single neural network models are unable to effectively process and integrate the temporal and spatial characteristics of these data, thus failing to fully utilize multimodal data and affecting overall forecasting performance. Summary of the Invention

[0004] Embodiments of the present invention provide a method, apparatus, device, medium, and product for offshore photovoltaic short-term power prediction to address the limitations of existing solutions in dealing with complex environments and multi-source heterogeneous data, thereby improving the accuracy and robustness of short-term power prediction for offshore photovoltaic power stations.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting offshore photovoltaic short-term power, the method comprising:

[0006] Obtain multi-modal raw meteorological data for the area where the target PV power station is located;

[0007] The combined deep learning model is used to extract the spatiotemporal features of the corresponding modes in the original meteorological data;

[0008] An attention mechanism is used to align and fuse the spatiotemporal features to obtain a multimodal feature vector.

[0009] A short-term power prediction result of the target photovoltaic power station is obtained according to the multimodal feature vector prediction.

[0010] Optionally, the combined deep learning model is used to extract the spatiotemporal features of the corresponding modality in the original meteorological data, including:

[0011] A convolutional neural network is used to extract spatial features from the original meteorological data;

[0012] The spatial features are converted into feature vector sequences and input into a long short-term memory network to obtain the temporal features in the original meteorological data.

[0013] Optionally, the original meteorological data includes satellite cloud image data and marine lidar scanning data; the method further includes:

[0014] Obtaining numerical weather forecast data for the area where the target photovoltaic power station is located;

[0015] The numerical weather forecast data is fused with the spatial features extracted based on the satellite cloud image data to obtain enhanced cloud image features.

[0016] Optionally, the use of an attention mechanism to perform spatiotemporal alignment and fusion on the spatiotemporal features to obtain a multimodal feature vector includes:

[0017] Inputting the spatiotemporal features of each modality into different fully connected layers to generate a query-key-value matrix;

[0018] Calculating the dot product similarity between the query and the key based on the query-key-value matrix;

[0019] Determining the association weights between the spatiotemporal features of each modality according to the dot product similarity;

[0020] The spatiotemporal features of each modality are weighted and summed according to the association weights to obtain the multimodal feature vector.

[0021] Optionally, the obtaining a short-term power prediction result of the target photovoltaic power station according to the multimodal feature vector prediction includes:

[0022] Performing feature enhancement on the multimodal feature vector through a convolutional neural network;

[0023] The enhanced multimodal feature vector is temporally modeled using a long short-term memory network, and the short-term power prediction result is predicted based on the modeling result.

[0024] Optionally, the method further includes:

[0025] Obtaining station-side meteorological data for the area where the target photovoltaic power station is located;

[0026] Using a classification model to identify abnormal weather based on the station-side meteorological data and the original meteorological data;

[0027] The feature weights between the spatiotemporal features of each mode during fusion are adjusted according to the abnormal weather, or the model used for short-term power prediction is switched according to the abnormal weather.

[0028] In a second aspect, an embodiment of the present invention further provides an offshore photovoltaic short-term power prediction device, the device comprising:

[0029] A meteorological data acquisition module is used to obtain multi-modal raw meteorological data of the area where the target photovoltaic power station is located;

[0030] A spatiotemporal feature extraction module, configured to extract the spatiotemporal features of the corresponding modes in the original meteorological data using a combined deep learning model;

[0031] A spatiotemporal feature fusion module is used to perform spatiotemporal alignment and fusion of the spatiotemporal features using an attention mechanism to obtain a multimodal feature vector;

[0032] The short-term power prediction module is used to obtain a short-term power prediction result of the target photovoltaic power station according to the multimodal feature vector prediction.

[0033] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0034] one or more processors;

[0035] a memory for storing one or more programs;

[0036] When the one or more programs are executed by the one or more processors, the one or more processors implement the offshore photovoltaic short-term power prediction method provided by any embodiment of the present invention.

[0037] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the offshore photovoltaic short-term power prediction method provided by any embodiment of the present invention.

[0038] In a fifth aspect, an embodiment of the present invention further provides a computer program product, which includes a computer program, and when the program is executed by a processor, it implements the offshore photovoltaic short-term power prediction method provided by any embodiment of the present invention.

[0039] An embodiment of the present invention provides a method for offshore photovoltaic short-term power prediction, which first obtains multi-modal original meteorological data of the area where the target photovoltaic power station is located, then uses a combined deep learning model to extract the spatiotemporal features of the corresponding modality in the original meteorological data, and then uses an attention mechanism to perform spatiotemporal alignment and fusion on the extracted spatiotemporal features to obtain a multi-modal feature vector, and then predicts the short-term power prediction result of the target photovoltaic power station based on the multi-modal feature vector. The offshore photovoltaic short-term power prediction method provided by the embodiment of the present invention solves the limitations of existing solutions in dealing with complex environments and multi-source heterogeneous data by utilizing the feature extraction capabilities of deep learning, and using a combined deep learning model to simultaneously process temporal and spatial features, and then combining the attention mechanism to effectively align and fuse multi-modal features, thereby solving the limitations of existing solutions in dealing with complex environments and multi-source heterogeneous data, thereby being able to more accurately and stably perform short-term power prediction of offshore photovoltaic power stations to better adapt to the actual operation needs of offshore photovoltaic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Flowchart of the offshore photovoltaic short-term power prediction method provided in Example 1 of the present invention;

[0041] Figure 2 A schematic diagram of the structure of an offshore photovoltaic short-term power prediction device provided in the second embodiment of the present invention;

[0042] Figure 3 This is a structural diagram of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0044] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the steps as sequential processes, many of the steps can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0045] Example 1

[0046] Figure 1This is a flow chart of the offshore photovoltaic short-term power prediction method provided in Example 1 of the present invention. This embodiment is applicable to situations where accurate short-term power prediction of offshore photovoltaic power stations is required in complex marine environments. This method can be performed by the offshore photovoltaic short-term power prediction device provided in this embodiment of the present invention. This device can be implemented using hardware and / or software and can generally be integrated into electronic equipment.

[0047] like Figure 1 As shown, the specific steps include:

[0048] S11. Acquire multi-modal original meteorological data of the area where the target photovoltaic power station is located.

[0049] S12. Use a combined deep learning model to extract the spatiotemporal features of the corresponding modal in the original meteorological data.

[0050] S13. Use the attention mechanism to perform spatiotemporal alignment and fusion on the spatiotemporal features to obtain a multimodal feature vector.

[0051] S14. Obtain a short-term power prediction result of the target photovoltaic power station according to the multimodal feature vector prediction.

[0052] Specifically, raw meteorological data can include satellite cloud imagery and marine LiDAR scans. Satellite cloud imagery provides a macroscopic overview of weather conditions, including dynamic cloud information such as cloud cover, cloud thickness, cloud morphology, spatial distribution, and movement speed. These characteristics are crucial for photovoltaic power forecasting, as changes in cloud cover directly affect solar irradiance, and thus the power output of photovoltaic power plants. Satellite cloud imagery has high spatial resolution and can cover wide areas, providing continuous, real-time weather images. LiDAR scans can capture atmospheric vertical profile information, including vertical profiles of low- and high-altitude wind speed, wind direction, temperature, and humidity. This data accurately reflects local meteorological conditions at sea, particularly atmospheric stability and boundary layer conditions, which are also critical for short-term power forecasting. Specifically, satellite cloud imagery data can be collected at a frame rate of every 15 minutes, and marine LiDAR scan data can be collected at a rate of once per minute, with the collected data pre-processed and standardized.

[0053] After obtaining multimodal raw meteorological data, a combined deep learning model can be used to extract temporal and spatial features from each modality's raw meteorological data, yielding the corresponding modal's spatiotemporal features. Feature extraction converts raw data into features that are more easily usable for modeling, helping prediction models better understand the data.

[0054] Optionally, the use of a combined deep learning model to extract the spatiotemporal features of the corresponding modality in the original meteorological data includes: using a convolutional neural network to extract the spatial features in the original meteorological data; converting the spatial features into a feature vector sequence and inputting them into a long short-term memory network to obtain the temporal features in the original meteorological data.

[0055] Among them, the Convolutional Neural Network (CNN) is a deep learning model that excels at extracting local features of images (such as edges and textures). It extracts spatial features of images through convolution operations and reduces the dimensionality of images through pooling operations. Specifically, convolutional neural networks can be used to extract spatial features from satellite cloud image data and marine lidar scanning data. Information such as cloud structure, cloud amount, thickness, density, and movement direction in satellite cloud images, as well as layered structures in lidar scan images, can all be extracted and expressed through the multi-layer convolutional network of CNN. Through these feature extractions, the original image data can be converted into feature vectors that are meaningful for prediction tasks, enabling the prediction model to understand the changing trends of the cloud layer and its impact on solar irradiation, thereby providing reliable spatial information support for photovoltaic power prediction.

[0056] The Long Short-Term Memory (LSTM) network is a variant of the Recurrent Neural Network (RNN) specifically designed to address the vanishing and exploding gradient problems encountered by standard RNNs when processing long-term dependencies. LSTM controls the flow of information through its unique cell state and input, forget, and output gates, effectively capturing important features in long time series. Specifically, LSTMs can be used to extract temporal features from satellite cloud imagery and marine LiDAR scan data. For example, changes in cloud cover in satellite cloud images and vertical meteorological profiles obtained from LiDAR scans are both time-dependent information. Using LSTM, these time series features can be modeled to capture the long-term dynamic characteristics of the data, such as the temporal variations in meteorological elements such as wind speed and temperature, particularly the vertical gradient and dynamic characteristics of wind speed at different altitudes. This makes the model more robust in responding to complex weather changes.

[0057] Specifically, for each modality of raw meteorological data, spatial features are first extracted using a convolutional neural network. These spatial features are then converted into a sequence of feature vectors and fed into a long-short-term memory network to extract temporal features. By jointly extracting features from raw meteorological data using a convolutional neural network and a long-short-term memory network, the team fully captures the spatiotemporal characteristics of multi-source heterogeneous data.

[0058] For satellite cloud imagery data, a convolutional neural network can include multiple convolutional layers, each with a kernel size of 3×3 and a stride of 1, with the number of kernels gradually increasing from 32 to 128. The convolution operation can extract spatial features from satellite cloud images, such as cloud texture, dense areas, and edge information. A ReLU activation function is applied after each convolutional layer to introduce nonlinearity. The ReLU activation function effectively addresses the vanishing gradient problem and improves the model's ability to fit nonlinear features. Every two convolutional layers are followed by a max pooling layer with a 2×2 kernel size to reduce dimensionality while retaining key features, thereby reducing computational effort and enhancing model generalization. Batch normalization can also be added after the convolutional layers to prevent vanishing or exploding gradients and accelerate model convergence. A long short-term memory network can include two LSTM layers, each with 64 hidden units, to model the temporal dynamics of cloud morphology and thickness. This combined deep learning model can generate spatiotemporal feature vectors for satellite cloud images, incorporating both the spatial structure and temporal variations of the clouds.

[0059] For marine lidar scan data, the convolutional neural network is similar to the one described above, with the convolution kernel size adjusted to 1×3 to accommodate the characteristics of lidar scan data, specifically extracting vertical atmospheric profile features. Specifically, three layers of convolution are used to extract vertical features of wind speed, temperature, and humidity. Each convolutional layer is followed by a ReLU activation function and a max pooling layer with a pooling kernel size of 1×2 to retain the most important vertical meteorological features. The long short-term memory network, consisting of two LSTM layers, each with 128 hidden units, is used to capture the temporal evolution of the atmospheric vertical profile, particularly changes in wind speed, temperature, and humidity over different time periods. This combined deep learning model generates spatiotemporal feature vectors for marine lidar scan data, combining profile features with temporal dynamics. Furthermore, by combining multiple convolutional and LSTM layers, comprehensive cloud dynamics and atmospheric profile features are captured.

[0060] After obtaining the spatiotemporal features of each modality, feature fusion can be performed to form a comprehensive multimodal feature vector. This fully captures the complex factors affecting PV power output, thereby improving the prediction model's performance in complex tasks. Because multimodal data originates from different sensors or models and has different temporal and spatial distributions, spatiotemporal alignment of the features is necessary during fusion to ensure that the correlations between the multimodal data are fully captured. This allows the prediction model to fully utilize the various feature information, thereby improving the accuracy of subsequent prediction models. The attention mechanism is a deep learning technique that assigns different weights to input features, enabling the model to focus on the most important information, thereby improving task accuracy. Specifically, the attention mechanism can be used to align multimodal spatiotemporal features. By learning the importance relationships between features from different modalities, the attention mechanism can achieve flexible and effective alignment. During fusion, it dynamically assigns importance to different features and gives key features higher weights during the fusion process, thereby improving overall prediction accuracy and enhancing the robustness and adaptability of the prediction model to complex marine environments. For example, when marine lidar scanning data (such as the atmospheric boundary layer) shows significant changes over a certain period of time, or when satellite cloud image data (such as local cloud cover) changes rapidly, the attention mechanism can automatically increase its weight accordingly, allowing the model to pay more attention to this specific feature, which is particularly helpful for improving power prediction results under abnormal weather conditions.

[0061] Optionally, the attention mechanism is used to perform spatiotemporal alignment and fusion of the spatiotemporal features to obtain a multimodal feature vector, including: inputting the spatiotemporal features of each modality into different fully connected layers to generate a query-key-value matrix; calculating the dot product similarity between the query and the key according to the query-key-value matrix; determining the association weight between the spatiotemporal features of each modality according to the dot product similarity; and performing weighted summation of the spatiotemporal features of each modality according to the association weight to obtain the multimodal feature vector.

[0062] Specifically, a query-key-value (QKV) mechanism can be employed, where the spatiotemporal features of satellite cloud images and marine lidar scans are fed into separate fully connected layers to generate corresponding query (Q), key (K), and value (V) matrices. The dot product similarity between the query and key is then calculated to determine the association weights between the spatiotemporal features of different modalities. Based on these association weights, the spatiotemporal features of different modalities are weighted and summed to produce a fused multimodal feature vector. This vector incorporates the spatiotemporal information relationships between the features of different modalities and comprehensively considers the spatial variations of large-scale clouds and the temporal variations of local meteorological conditions. A higher weight indicates a greater impact of the feature on power prediction at the current moment. Furthermore, to capture the multifaceted correlations between features, a multi-head attention mechanism can be introduced, where multiple attention heads are run in parallel and their outputs are concatenated to form a fused high-dimensional feature representation, which serves as the desired multimodal feature vector. Satellite cloud images provide wide-area weather dynamics information, and marine lidar scans provide detailed vertical structure. By integrating satellite cloud images and marine lidar scans for precise modeling of multimodal spatiotemporal characteristics, we can simultaneously capture macro-weather evolution and subtle changes in local meteorological conditions, making full use of the complementarity of data from different sources. This allows us to establish a more comprehensive offshore photovoltaic power station power output prediction model to improve the accuracy of photovoltaic power prediction.

[0063] After obtaining the multimodal feature vector, a pre-trained prediction model can be used to predict the short-term power of the target photovoltaic power station based on the multimodal feature vector to obtain a short-term power prediction result. Optionally, obtaining the short-term power prediction result of the target photovoltaic power station based on the multimodal feature vector includes: enhancing the multimodal feature vector using a convolutional neural network; performing temporal modeling on the enhanced multimodal feature vector using a long short-term memory network; and predicting the short-term power prediction result based on the modeling result.

[0064] Specifically, a prediction model combining CNN and LSTM can be used. After obtaining the multimodal feature vector, CNN is first used for further feature enhancement. Specifically, the multimodal feature vector can be input into a convolutional layer to extract deep spatial features. This can, for example, identify cloud and wind speed variations that may have been missed during the multimodal feature fusion process. LSTM time series modeling can then be performed based on the enhanced multimodal feature vector. Specifically, a single LSTM layer with 128 hidden units can be used for temporal modeling to capture power output trends across time periods, thereby predicting the short-term power forecast for the target PV power plant. During the time series feature prediction stage, a temporal attention mechanism can be further introduced to dynamically weight the feature importance at different time points to increase attention to important time segments, further improving the accuracy of power forecasts.

[0065] Further model evaluation and optimization can be performed. Specifically, evaluation indicators such as mean square error (MSE), mean absolute percentage error (MAPE), and coefficient of determination (R 2 ), etc. Model hyperparameters can then be adjusted through grid search or Bayesian optimization, including the number of layers and hidden units in the LSTM model, the convolution kernel size in the CNN model, and the learning rate, to minimize errors in evaluation metrics. K-fold cross-validation can also be performed on the training data to ensure model stability and adaptability to different datasets, thereby achieving high accuracy and robustness in complex maritime environments.

[0066] In an optional embodiment, the method further includes: obtaining station-side meteorological data of the area where the target photovoltaic power station is located; using a classification model to identify abnormal weather based on the station-side meteorological data and the original meteorological data; adjusting the feature weights between the spatiotemporal features of each modality during fusion according to the abnormal weather, or switching the model used for short-term power prediction according to the abnormal weather.

[0067] Specifically, in photovoltaic power forecasting, abnormal weather often has a significant impact on the forecast results. Therefore, anomaly weather handling can be added to correct the power forecast value according to the identified abnormal weather, thereby improving the robustness and prediction accuracy of the model in the corresponding situation and enhancing the superiority of power forecast performance under abnormal weather conditions. Specifically, a classification model can be used to identify abnormal weather by combining station-side meteorological data (which can provide ground meteorological conditions), satellite cloud image data, and marine lidar scanning data (all types of data can be standardized in advance). Among them, the classification model can specifically use a random forest, and the above-mentioned types of data can be used as input to train a random forest classifier. The classifier outputs whether abnormal weather exists, such as typhoons, heavy rains, etc. The output classification results can be used to determine whether special forecasting strategies need to be adopted. A support vector machine (SVM) model can also be used to assist in classification as a supplement to the random forest to be suitable for processing small samples and weather characteristics with nonlinear boundaries.

[0068] When abnormal weather is identified, the prediction strategy can be adjusted accordingly. For example, based on the above fusion results, an attention mechanism can be used to redistribute the feature weights when fusing multimodal spatiotemporal features. For example, when high wind speed conditions are detected, the radar wind speed feature is given a higher weight to better reflect the impact of weather changes on power output. Different prediction models can also be pre-trained for various types of abnormal weather. When abnormal weather is identified, the corresponding model branch can be selected to switch to. This model branch can then be specifically designed to handle the corresponding extreme weather conditions, with enhanced robustness and adaptability, ensuring high prediction accuracy even in extreme environments.

[0069] In an optional embodiment, the original meteorological data includes satellite cloud image data and marine lidar scanning data; the method also includes: obtaining numerical weather forecast data for the area where the target photovoltaic power station is located; fusing the numerical weather forecast data with the spatial features extracted based on the satellite cloud image data to obtain enhanced cloud image features.

[0070] Among them, Numerical Weather Prediction (NWP) is a mathematical model based on physical equations (such as atmospheric dynamics equations, thermodynamics equations, etc.), which predicts future weather by numerical simulation of atmospheric state. Numerical weather forecast data can be generated by a regional climate model, which is a regional-level meteorological simulation that can provide more comprehensive background meteorological data for photovoltaic power prediction, such as wind speed, wind direction, temperature, humidity and air pressure, which can directly affect solar irradiance and aerodynamic conditions. It usually has a high spatial and temporal resolution, so that the prediction model can make more accurate predictions based on actual weather conditions. The regional meteorological model is a high-resolution meteorological simulation model based on a specific area, usually presented in the form of grid points, which can provide refined regional meteorological forecast data to support power prediction and decision-making in a specific area. For example, under certain extreme weather conditions, changes in specific meteorological parameters in the numerical weather forecast may directly affect the power generation of the photovoltaic system. By combining this information, the reliability and effectiveness of the prediction model can be significantly improved. Or in a specific area, due to the influence of topography or ocean climate, the weather characteristics may be unique to a certain extent. The regional meteorological model can provide more realistic meteorological input for the power prediction model through detailed simulation of regional characteristics.

[0071] Specifically, after extracting spatial features from satellite cloud image data, these spatial features can be deeply extracted and enhanced by combining them with numerical weather forecast data for the region where the target PV power station is located. The regional meteorological large-scale model provides a forecast of future weather conditions. By combining these features with satellite cloud imagery, a more comprehensive understanding of cloud dynamics and their impact on PV power output can be achieved, thereby improving forecast accuracy. This not only helps capture microscopic changes in regional weather but also allows for better responses to abnormal weather and extreme conditions. Convolutional neural networks can be used to extract spatial features from satellite cloud imagery data, such as cloud cover, cloud thickness, and cloud morphology. These features primarily reflect current weather conditions and lack the ability to predict future trends. Once these spatial features are obtained, the numerical weather forecast data output by the regional meteorological large-scale model can be fused with these spatial features to form enhanced cloud image features, which serve as new spatial features for subsequent spatiotemporal feature fusion and prediction. This results in a feature modeling approach that combines data-driven and physical models, enabling the model to adapt to complex ocean environments and changing meteorological conditions, achieving higher adaptability and forecast accuracy. This fusion process not only takes into account the characteristics of the clouds at the current moment, but also combines the regional meteorological model's predictions of future meteorological changes, making the model more superior in capturing dynamic meteorological changes, thereby enabling more accurate predictions of future cloud movement and its impact on photovoltaic power.

[0072] Furthermore, the spatial features of numerical weather forecast data and satellite cloud imagery data can be fused in multiple deep convolutional layers. By incorporating regional meteorological elements as auxiliary signals in different convolutional layers, the model's perception capabilities can be enhanced at each convolutional layer. At the same time, the feature weights of each meteorological element can be distributed and adaptively adjusted in different convolutional layers to ensure that the numerical weather forecast data can effectively guide the deep learning model to extract more physically meaningful features, thereby enhancing the perception of future cloud evolution and changing meteorological conditions, and improving the model's flexibility and generalization capabilities. For example, in the initial convolutional layers, the fused meteorological elements can help the model identify cloud morphological characteristics under different weather conditions. For example, fusing wind speed and air pressure from weather forecasts with edge detection features from satellite cloud imagery can improve the ability to resolve cloud morphology. In deeper convolutional layers, the fused meteorological elements can adjust the prediction of cloud dynamics, ensuring that the model accurately models cloud evolution. For example, by incorporating temperature change forecasts for the next few hours, the understanding of cloud movement and deformation can be enhanced. By combining this with numerical weather forecast data, temporal and spatial correlations can be enhanced, enabling the model to identify relationships between specific cloud dynamics and meteorological factors. For example, wind speed forecasts can help the model understand the direction and speed of cloud movement, while humidity forecasts can help the model identify possible precipitation conditions, further influencing PV power predictions.

[0073] The technical solution provided by the embodiment of the present invention first obtains the multimodal raw meteorological data of the area where the target photovoltaic power station is located, then uses a combined deep learning model to extract the spatiotemporal features of the corresponding modalities in the raw meteorological data, and then uses an attention mechanism to perform spatiotemporal alignment and fusion of the extracted spatiotemporal features to obtain a multimodal feature vector, and then predicts the short-term power prediction result of the target photovoltaic power station based on the multimodal feature vector. By utilizing the feature extraction capability of deep learning, and using a combined deep learning model to simultaneously process temporal and spatial features, and then combining the attention mechanism to effectively align and fuse multimodal features, the limitations of existing solutions in dealing with complex environments and multi-source heterogeneous data are resolved, thereby enabling more accurate and stable short-term power prediction of offshore photovoltaic power stations to better meet the actual operating needs of offshore photovoltaic systems.

[0074] Example 2

[0075] Figure 2 This is a schematic diagram of the structure of the offshore photovoltaic short-term power prediction device provided in the second embodiment of the present invention. The device can be implemented by hardware and / or software and can generally be integrated into electronic devices to execute the offshore photovoltaic short-term power prediction method provided in any embodiment of the present invention. Figure 2 As shown, the device includes:

[0076] The meteorological data acquisition module 21 is used to obtain multi-modal original meteorological data of the area where the target photovoltaic power station is located;

[0077] A spatiotemporal feature extraction module 22 is configured to extract the spatiotemporal features of the corresponding modalities in the original meteorological data using a combined deep learning model;

[0078] A spatiotemporal feature fusion module 23 is configured to perform spatiotemporal alignment and fusion of the spatiotemporal features using an attention mechanism to obtain a multimodal feature vector;

[0079] The short-term power prediction module 24 is configured to obtain a short-term power prediction result of the target photovoltaic power station based on the multimodal feature vector prediction.

[0080] The technical solution provided by the embodiment of the present invention first obtains the multimodal raw meteorological data of the area where the target photovoltaic power station is located, then uses a combined deep learning model to extract the spatiotemporal features of the corresponding modalities in the raw meteorological data, and then uses an attention mechanism to perform spatiotemporal alignment and fusion of the extracted spatiotemporal features to obtain a multimodal feature vector, and then predicts the short-term power prediction result of the target photovoltaic power station based on the multimodal feature vector. By utilizing the feature extraction capability of deep learning, and using a combined deep learning model to simultaneously process temporal and spatial features, and then combining the attention mechanism to effectively align and fuse multimodal features, the limitations of existing solutions in dealing with complex environments and multi-source heterogeneous data are resolved, thereby enabling more accurate and stable short-term power prediction of offshore photovoltaic power stations to better meet the actual operating needs of offshore photovoltaic systems.

[0081] On the basis of the above technical solution, optionally, the spatiotemporal feature extraction module 22 is specifically used to:

[0082] A convolutional neural network is used to extract spatial features from the original meteorological data;

[0083] The spatial features are converted into feature vector sequences and input into a long short-term memory network to obtain the temporal features in the original meteorological data.

[0084] Based on the above technical solution, optionally, the original meteorological data includes satellite cloud image data and marine laser radar scanning data; the device further includes:

[0085] A forecast data acquisition module, configured to acquire numerical weather forecast data for the area where the target photovoltaic power station is located;

[0086] The cloud image feature enhancement module is used to fuse the numerical weather forecast data with the spatial features extracted based on the satellite cloud image data to obtain enhanced cloud image features.

[0087] On the basis of the above technical solution, optionally, the spatiotemporal feature fusion module 23 is specifically used to:

[0088] Inputting the spatiotemporal features of each modality into different fully connected layers to generate a query-key-value matrix;

[0089] Calculating the dot product similarity between the query and the key based on the query-key-value matrix;

[0090] Determining the association weights between the spatiotemporal features of each modality according to the dot product similarity;

[0091] The spatiotemporal features of each modality are weighted and summed according to the association weights to obtain the multimodal feature vector.

[0092] On the basis of the above technical solution, optionally, the short-term power prediction module 24 is specifically used to:

[0093] Performing feature enhancement on the multimodal feature vector through a convolutional neural network;

[0094] The enhanced multimodal feature vector is temporally modeled using a long short-term memory network, and the short-term power prediction result is predicted based on the modeling result.

[0095] On the basis of the above technical solution, optionally, the device further includes:

[0096] A station-side meteorological data acquisition module, configured to acquire station-side meteorological data of the area where the target photovoltaic power station is located;

[0097] an abnormal weather identification module, configured to identify abnormal weather based on the station-side meteorological data and the original meteorological data using a classification model;

[0098] A model adjustment module is used to adjust the feature weights between the spatiotemporal features of each mode during fusion according to the abnormal weather, or to switch the model used for short-term power prediction according to the abnormal weather.

[0099] The offshore photovoltaic short-term power prediction device provided by the embodiment of the present invention can execute the offshore photovoltaic short-term power prediction method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0100] It is worth noting that in the above-mentioned embodiment of the offshore photovoltaic short-term power prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0101] Example 3

[0102] Figure 3 This is a structural diagram of an electronic device provided in accordance with a third embodiment of the present invention, showing a block diagram of an exemplary electronic device suitable for implementing the embodiments of the present invention. Figure 3 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention. Figure 3 As shown, the electronic device includes a processor 31, a memory 32, an input device 33 and an output device 34; the number of processors 31 in the electronic device can be one or more. Figure 3 Taking a processor 31 as an example, the processor 31, memory 32, input device 33 and output device 34 in the electronic device can be connected through a bus or other means. Figure 3 The bus connection is taken as an example.

[0103] Memory 32, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the offshore photovoltaic short-term power forecasting method in the embodiments of the present invention (for example, the meteorological data acquisition module 21, spatiotemporal feature extraction module 22, spatiotemporal feature fusion module 23, and short-term power forecasting module 24 in the offshore photovoltaic short-term power forecasting device). Processor 31 executes the software programs, instructions, and modules stored in memory 32 to execute various functional applications and data processing of the electronic device, thereby implementing the aforementioned offshore photovoltaic short-term power forecasting method.

[0104] The memory 32 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the electronic device, etc. Furthermore, the memory 32 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, the memory 32 may further include a memory remotely located relative to the processor 31, and these remote memories may be connected to the electronic device via a network. Examples of the aforementioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0105] The input device 33 may be used to obtain multimodal meteorological data and generate key signal inputs related to user settings and function control of the electronic device, etc. The output device 34 may include a display screen, which may be used to display real-time forecast results, etc. to the user.

[0106] Example 4

[0107] A fourth embodiment of the present invention further provides a storage medium containing computer-executable instructions. When the computer-executable instructions are executed by a computer processor, the computer-executable instructions are used to perform a method for predicting offshore photovoltaic short-term power. The method includes:

[0108] Obtain multi-modal raw meteorological data for the area where the target PV power station is located;

[0109] The combined deep learning model is used to extract the spatiotemporal features of the corresponding modes in the original meteorological data;

[0110] An attention mechanism is used to align and fuse the spatiotemporal features to obtain a multimodal feature vector.

[0111] A short-term power prediction result of the target photovoltaic power station is obtained according to the multimodal feature vector prediction.

[0112] The storage medium can be any of various types of memory devices or storage devices. The term "storage medium" is intended to include: installation media, such as CD-ROMs, floppy disks, or tape devices; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (such as hard disks or optical storage); registers or other similar types of memory elements, etc. The storage medium may also include other types of memory or combinations thereof. In addition, the storage medium may be located in the computer system in which the program is executed, or may be located in a different second computer system that is connected to the computer system via a network (such as the Internet). The second computer system may provide program instructions to the computer for execution. The term "storage medium" may include two or more storage media that may reside in different locations (e.g., in different computer systems connected via a network). The storage medium may store program instructions (e.g., embodied as a computer program) that may be executed by one or more processors.

[0113] Of course, the storage medium containing computer-executable instructions provided in an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in the offshore photovoltaic short-term power prediction method provided in any embodiment of the present invention.

[0114] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0115] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0116] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0117] Example 5

[0118] Embodiment 5 of the present invention also provides a computer program product, which includes a computer program (also referred to as code, instructions). The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it is used to execute the offshore photovoltaic short-term power prediction method provided in any of the above embodiments, and has the corresponding beneficial effects of the execution method.

[0119] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for predicting offshore photovoltaic short-term power, characterized in that: include: Obtain multi-modal raw meteorological data for the area where the target PV power station is located; The combined deep learning model is used to extract the spatiotemporal features of the corresponding modes in the original meteorological data; An attention mechanism is used to align and fuse the spatiotemporal features to obtain a multimodal feature vector. A short-term power prediction result of the target photovoltaic power station is obtained according to the multimodal feature vector prediction.

2. The offshore photovoltaic short-term power prediction method according to claim 1, characterized in that: The combined deep learning model is used to extract the spatiotemporal features of the corresponding modes in the original meteorological data, including: A convolutional neural network is used to extract spatial features from the original meteorological data; The spatial features are converted into feature vector sequences and input into a long short-term memory network to obtain the temporal features in the original meteorological data.

3. The offshore photovoltaic short-term power prediction method according to claim 2, characterized in that: The original meteorological data includes satellite cloud image data and marine laser radar scanning data; the method further includes: Obtaining numerical weather forecast data for the area where the target photovoltaic power station is located; The numerical weather forecast data is fused with the spatial features extracted based on the satellite cloud image data to obtain enhanced cloud image features.

4. The offshore photovoltaic short-term power prediction method according to claim 1, characterized in that: The attention mechanism is used to perform spatiotemporal alignment and fusion on the spatiotemporal features to obtain a multimodal feature vector, including: Inputting the spatiotemporal features of each modality into different fully connected layers to generate a query-key-value matrix; Calculating the dot product similarity between the query and the key based on the query-key-value matrix; Determining the association weights between the spatiotemporal features of each modality according to the dot product similarity; The spatiotemporal features of each modality are weighted and summed according to the association weights to obtain the multimodal feature vector.

5. The offshore photovoltaic short-term power prediction method according to claim 1, characterized in that: The obtaining of a short-term power prediction result of the target photovoltaic power station according to the multimodal feature vector prediction includes: Performing feature enhancement on the multimodal feature vector through a convolutional neural network; The enhanced multimodal feature vector is temporally modeled using a long short-term memory network, and the short-term power prediction result is predicted based on the modeling result.

6. The offshore photovoltaic short-term power prediction method according to claim 1, characterized in that: The method further comprises: Obtaining station-side meteorological data for the area where the target photovoltaic power station is located; Using a classification model to identify abnormal weather based on the station-side meteorological data and the original meteorological data; The feature weights between the spatiotemporal features of each mode during fusion are adjusted according to the abnormal weather, or the model used for short-term power prediction is switched according to the abnormal weather.

7. An offshore photovoltaic short-term power prediction device, characterized in that: include: A meteorological data acquisition module is used to obtain multi-modal raw meteorological data of the area where the target photovoltaic power station is located; A spatiotemporal feature extraction module, configured to extract the spatiotemporal features of the corresponding modes in the original meteorological data using a combined deep learning model; A spatiotemporal feature fusion module is used to perform spatiotemporal alignment and fusion of the spatiotemporal features using an attention mechanism to obtain a multimodal feature vector; The short-term power prediction module is used to obtain a short-term power prediction result of the target photovoltaic power station according to the multimodal feature vector prediction.

8. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the offshore photovoltaic short-term power prediction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the offshore photovoltaic short-term power prediction method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the offshore photovoltaic short-term power prediction method according to any one of claims 1 to 6.

Images