Photovoltaic power generation power prediction method and system, electronic equipment and storage medium

Abstract

The invention discloses a photovoltaic power generation power prediction method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining photovoltaic historical power data and sky cloud image data, and carrying out the preprocessing of the photovoltaic historical power data and the sky cloud image data; performing feature extraction on the preprocessed sky cloud image based on a pre-trained feature extraction model, and splicing the extracted image features and the preprocessed photovoltaic historical power data to form a combined feature vector; and establishing a time sequence model of the combined feature vectors, and predicting the photovoltaic power generation power based on the time sequence model. According to the method, the fine-grained features of the sky image are extracted through the VGG19, then splicing processing is carried out in combination with photovoltaic historical data, time sequence modeling is carried out by utilizing a self-attention mechanism of iTransform, accurate prediction of photovoltaic power is realized, dynamic changes of environmental factors can be effectively coped with, and the accuracy and efficiency of online monitoring and energy management of a photovoltaic system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation system monitoring, and in particular to a photovoltaic power generation power prediction method, system, electronic equipment and storage medium. Background Art

[0002] The large-scale deployment of solar energy not only reduces carbon emissions and helps address climate change, but also improves energy security, reduces dependence on fossil fuels, and promotes a green and sustainable economic transition. However, the volatility and intermittency of solar power generation are major challenges. Power generation capacity is affected by multiple factors, including weather and climate, which complicates grid operations. Therefore, photovoltaic power forecasting is crucial. Accurate power forecasting not only helps manage photovoltaic power fluctuations and ensures stable grid operation, but also optimizes power dispatch and resource allocation, improving energy efficiency. Furthermore, accurate forecasting helps reduce power system operating costs, minimize backup power requirements, and promote the development of smart grids.

[0003] In the field of photovoltaic power forecasting, traditional machine learning methods such as linear regression, support vector machines, decision trees, and random forests are widely used. Each of these methods has its own advantages and disadvantages. Linear regression models are known for their simplicity and efficiency, making them suitable for linear relationships. However, they struggle with the complex nonlinear characteristics of photovoltaic power generation data. Support vector machines perform well with high-dimensional data, especially with small sample datasets, and have strong generalization capabilities. However, they require long training times and complex hyperparameter tuning. Decision trees offer good interpretability and are easy to understand and implement, but are prone to overfitting when dealing with complex data, leading to inaccurate predictions. Random forests improve prediction accuracy by integrating multiple decision trees, but this also increases computational complexity, requiring more computing resources for model training and prediction. In contrast, neural network models demonstrate superior performance in photovoltaic power forecasting. Long short-term memory networks (LSTMs) can capture long-term dependencies in time series data, making them suitable for handling the temporal variations of photovoltaic power generation data. However, they require long training times and are prone to overfitting when data volumes are insufficient. Convolutional neural networks significantly improve performance through feature extraction, especially when processing images and sequence data. However, they have high requirements for data preprocessing and require a large amount of computing resources. Deep neural networks have powerful modeling capabilities and can extract multi-level features from complex data, but the training process is relatively complex and prone to overfitting. To address the challenges in photovoltaic power prediction, the present invention proposes a photovoltaic power prediction model that combines the VGG19 convolutional neural network and the improved iTransformer model. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a photovoltaic power prediction method, system, electronic device and storage medium to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a photovoltaic power generation power prediction method, comprising:

[0007] Acquiring photovoltaic historical power data and sky cloud image data, and preprocessing the photovoltaic historical power data and sky cloud image data;

[0008] Performing feature extraction on the pre-processed sky cloud image based on a pre-trained feature extraction model, and concatenating the extracted image features with the pre-processed photovoltaic historical power data to form a combined feature vector;

[0009] A time series model of the combined feature vector is established, and photovoltaic power generation power is predicted based on the time series model.

[0010] As a preferred solution of the photovoltaic power generation prediction method of the present invention, the preprocessing of the sky cloud image data includes:

[0011] Standardize the sky cloud image sequence to make the data size of each image the same;

[0012] A quality threshold is preset, and when there is an abnormal image in the sky cloud image, if the quality index of the abnormal image is not less than the quality threshold, the abnormal image data is retained; if the quality index of the abnormal image is less than the quality threshold, the low-quality abnormal image is eliminated;

[0013] Downsample the image sequence in the temporal dimension.

[0014] As a preferred solution of the photovoltaic power generation prediction method of the present invention, the preprocessing of photovoltaic historical power data includes: presetting a threshold value to identify abnormal values ​​of the photovoltaic historical power data sequence;

[0015] If the absolute deviation of the photovoltaic power data value at any time point from the mean of the entire photovoltaic historical power data series exceeds a preset threshold, it will be marked as an outlier, and the outlier will be corrected, and the normal photovoltaic historical power data will be standardized.

[0016] As a preferred embodiment of the photovoltaic power generation prediction method of the present invention, before extracting features from the pre-processed sky cloud image based on the pre-trained feature extraction model, the method further comprises: synchronizing the sky cloud image sequence with the photovoltaic historical power data in the time dimension;

[0017] Assume that the time series of sky cloud images is I={I1,I2,…,I T}, the photovoltaic historical data is P = {p1, p2, ..., p T}, by matching the corresponding time points t to form the aligned input feature pairs (i t ,p t ).

[0018] As a preferred embodiment of the photovoltaic power generation prediction method of the present invention, the method includes: performing feature extraction on the pre-processed sky cloud image based on a pre-trained feature extraction model, and concatenating the extracted image features with the pre-processed photovoltaic historical power data to form a combined feature vector, including:

[0019] Input the sky cloud image into a feature extraction model using a 3×3 convolution kernel, a ReLU activation function, and a max pooling layer;

[0020] The feature extraction model outputs the feature vector F VGG19 =VGG19(I′ i ), and perform dimensionality reduction processing on the feature vector, and merge it with the photovoltaic historical power data features to form a combined feature vector.

[0021] The beneficial effect of this preferred technical solution is that by extracting fine-grained features of sky images and combining them with photovoltaic historical data for splicing processing, the accuracy of photovoltaic power prediction can be improved.

[0022] As a preferred embodiment of the photovoltaic power generation prediction method of the present invention, the method of establishing a time series model of the combined feature vector and predicting the photovoltaic power generation based on the time series model includes:

[0023] Performing dimensionality reduction on the combined feature vector and calculating the importance weight of each time step through a self-attention mechanism;

[0024] Enhanced time series feature extraction is based on multiple stacked model units, each including a self-attention layer and a feedforward neural network. A preferred embodiment of the photovoltaic power prediction method of the present invention includes: establishing a time series model of the combined feature vectors, predicting photovoltaic power based on the time series model, and further comprising: processing the spliced ​​time series features through the time series model to predict photovoltaic power, and outputting the photovoltaic power prediction result.

[0025] The beneficial effect of this preferred technical solution is that it uses the self-attention mechanism to perform time series modeling to achieve accurate prediction of photovoltaic power, effectively respond to dynamic changes in environmental factors, and improve the accuracy and efficiency of online monitoring and energy management of photovoltaic systems.

[0026] In a second aspect, the present invention provides a photovoltaic power generation power prediction system, comprising:

[0027] A preprocessing module is used to obtain photovoltaic historical power data and sky cloud image data, and preprocess the photovoltaic historical power data and sky cloud image data;

[0028] An image feature extraction module is used to extract features from the preprocessed sky cloud image based on a pretrained feature extraction model, and to combine the extracted image features with the preprocessed photovoltaic historical power data to form a combined feature vector;

[0029] The time series reasoning module is used to establish a time series model of the combined feature vector and predict photovoltaic power generation based on the time series model.

[0030] In a third aspect, the present invention provides an electronic device, comprising:

[0031] memory and processor;

[0032] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the photovoltaic power generation power prediction method are implemented.

[0033] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the photovoltaic power generation power prediction method.

[0034] Compared with the existing technology, the beneficial effects of the present invention are as follows: the present invention extracts fine-grained features of sky images through VGG19, combines them with photovoltaic historical data for splicing processing, and uses the self-attention mechanism of iTransformer for time series modeling to achieve accurate prediction of photovoltaic power, which can effectively respond to dynamic changes in environmental factors and improve the accuracy and efficiency of online monitoring and energy management of photovoltaic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them:

[0036] Figure 1 A schematic diagram of a method flow of a photovoltaic power prediction method, system, electronic device, and storage medium according to an embodiment of the present invention;

[0037] Figure 2 The present invention is a schematic diagram of a system flow of a photovoltaic power prediction method, system, electronic device and storage medium according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0039] Example 1, reference Figure 1 , is an embodiment of the present invention, which provides a photovoltaic power generation power prediction method, comprising:

[0040] S100: Acquire photovoltaic historical power data and sky cloud image data, and preprocess the photovoltaic historical power data and sky cloud image data;

[0041] S200: extracting features from the pre-processed sky cloud image based on a pre-trained feature extraction model, and combining the extracted image features with the pre-processed photovoltaic historical power data to form a combined feature vector;

[0042] S300: Establish a time series model of the combined feature vector, and predict photovoltaic power generation power based on the time series model.

[0043] It should be noted that the stability of photovoltaic power generation is significantly affected by the weather type, and the cloud cover determines the amount of solar irradiance received by the photovoltaic panel. Therefore, the image-based method has obvious advantages in prediction, because the cloud change information in the sky image can more intuitively reflect the changing trend of photovoltaic power than numerical meteorological data. By analyzing the sky image, the movement changes of the clouds can be detected and more fine-grained information can be obtained, which is crucial to improving the accuracy of photovoltaic power prediction. This embodiment is based on VGG19 and iTransformer combined with the sky cloud map modeling strategy to establish a model with historical sky image sequence data and photovoltaic power generation historical power data as input and future photovoltaic power generation power prediction values ​​as output, ensuring that the association between image features and photovoltaic power is effectively modeled; the fine-grained features of the sky image are extracted through the VGG19 model and spliced ​​with the historical photovoltaic data; based on the time series modeling capability of Transformer, the spliced ​​features are subjected to time dependency analysis and feature modeling to ensure accurate prediction of future photovoltaic power changes.

[0044] The VGG19 model is a deep convolutional neural network designed to extract deep features from images. The image first passes through a series of convolutional and pooling layers, extracting features at different levels layer by layer. After the final few convolutional layers of VGG19, the output is a high-dimensional feature representation that has undergone dimensionality reduction. These features contain fine-grained information related to photovoltaic power generation, such as changes in cloud cover and sky brightness in the image.

[0045] Furthermore, the VGG19 model was pre-trained on the ImageNet dataset. During training, images were normalized and data augmented. The model used a 3×3 convolution kernel, ReLU activation function, and max pooling layers, with classification performed through three fully connected layers. Optimization used SGD (learning rate decay, momentum 0.9, L2 regularization), with a batch size of 256 and training for 74 epochs.

[0046] In the embodiment of the present application, the pre-processing of the sky cloud image data in step S100 includes:

[0047] Standardize the sky cloud image sequence to make the data size of each image the same;

[0048] Specifically, the sky cloud image sequence is standardized and all images are adjusted to a fixed size S×S to meet the requirements of model input and ensure continuity on the time axis.

[0049] A quality threshold is preset. When there is an abnormal image in the sky cloud image, if the quality index of the abnormal image is not less than the quality threshold, the abnormal image data will be retained; if the quality index of the abnormal image is less than the quality threshold, the low-quality abnormal image will be eliminated;

[0050] Downsample the image sequence in the temporal dimension.

[0051] For example, in order to reduce the amount of calculation and redundant information, downsampling is performed on the time axis. Assuming that the image sequence is I = {I1, I2, ..., I T}, after downsampling we obtain a new sequence I′={I1,I k+1 ,I 2k+1 ,…,I T}, where k is the sampling interval.

[0052] In an optional embodiment, the pre-processing of the sky cloud image data in step S100 may further include: using Gaussian filtering and mean filtering methods to reduce random noise in the image, thereby improving the accuracy of photovoltaic prediction analysis.

[0053] In another optional embodiment, the preprocessing of the sky cloud image data in step S100 may include: applying adaptive histogram equalization technology to highlight important features in the image so that the model can more easily identify and learn key features. In order to increase the diversity of the training set and prevent overfitting, the image may be subjected to transformation operations such as rotation, flipping, scaling, and brightness adjustment without affecting the authenticity.

[0054] In the embodiment of the present application, the pre-processing of the photovoltaic historical power data in step S100 includes: presetting a threshold value to identify abnormal values ​​in the photovoltaic historical power data sequence;

[0055] If the absolute deviation of the photovoltaic power data value at any time point from the mean of the entire photovoltaic historical power data series exceeds a preset threshold, it will be marked as an outlier, and the outlier will be corrected, and the normal photovoltaic historical power data will be standardized.

[0056] For example, the photovoltaic historical power data is preprocessed to detect and process abnormal values ​​first to ensure the continuity and stability of the data. Assume that the historical power data is a sequence P = {p1, p2, ..., p T}, where the outlier value p1 at a certain moment satisfies the condition ( is the mean, σ is the standard deviation, and α is the preset threshold), the outlier p can be corrected by interpolation or mean replacement. t to be processed.

[0057] In addition, to eliminate the influence of different numerical dimensions, the photovoltaic power data is standardized. Assume that the standardized power data is P′={p′1,p′2,…,p′ T}, the available formula Map the data to the [0,1] interval to reduce numerical fluctuations during model training.

[0058] In an optional embodiment, preprocessing the photovoltaic historical power data in step S100 may further include: decomposing the photovoltaic historical power data into trend components, seasonal components and residual components using a time series decomposition method to divide the data patterns and process different components separately.

[0059] In another optional embodiment, preprocessing the photovoltaic historical power data in step S100 may include: applying a sliding average smoothing technique to reduce short-term fluctuations in the data while retaining long-term trends and removing noise to make the model prediction more accurate.

[0060] In the embodiment of the present application, before extracting features from the pre-processed sky cloud image based on the pre-trained feature extraction model, the method further includes: synchronizing the sky cloud image sequence with the photovoltaic historical power data in the time dimension;

[0061] Assume that the time series of sky cloud images is I={I1,I2,…,I T}, the photovoltaic historical data is P = {p1, p2, ..., p T}, by matching the corresponding time points t to form the aligned input feature pairs (i t ,p t ).

[0062] In the embodiment of the present application, in step S200, feature extraction is performed on the pre-processed sky cloud image based on the pre-trained feature extraction model, and the extracted image features are spliced ​​with the pre-processed photovoltaic historical power data to form a combined feature vector, including:

[0063] The sky cloud image is fed into a feature extraction model using a 3×3 convolution kernel, ReLU activation function, and max pooling layer.

[0064] The feature extraction model outputs the feature vector F VGG19 =VGG19(I′ i ), and perform dimensionality reduction processing on the feature vector, which is then merged with the features of the photovoltaic historical power data to form a combined feature vector.

[0065] Specifically, the eigenvector F′ after dimensionality reduction VGG19 It can be expressed as:

[0066] F′ VGG19 =g(F VGG19 )

[0067] Here, g is a dimensionality reduction function that compresses feature dimensions while retaining key information. These features are then combined with historical PV data as model input.

[0068] Furthermore, the spliced ​​image features F img and photovoltaic historical data characteristics F pv Fusion is performed to form a combined feature vector F concat =[F img ,F pv ].

[0069] In the embodiment of the present application, step S300 establishes a time series model of the combined feature vectors, and predicts photovoltaic power generation based on the time series model, including:

[0070] Perform dimensionality reduction on the combined feature vector and calculate the importance weight of each time step through the self-attention mechanism;

[0071] Enhanced temporal feature extraction is based on multiple stacked model units, each of which includes a self-attention layer and a feed-forward neural network.

[0072] It should be noted that the reduced feature vector is input into the iTransformer model, leveraging its self-attention mechanism to dynamically model dependencies within time series data. This self-attention mechanism enables iTransformer to identify and capture correlations between input features at different time steps, effectively learning the long- and short-term dependencies of photovoltaic power data. Through the layered processing of multiple iTransformerBlocks, the model is able to gradually extract deep temporal information about features. Each iTransformerBlock incorporates a self-attention mechanism and a feedforward neural network module, which collaboratively capture the complex patterns of photovoltaic power time series data.

[0073] It should also be noted that the Embedding layer transforms the high-dimensional feature vector F concat Converted into low-dimensional feature representation F embed =E(F concat ), this process not only reduces the dimension of the feature, but also retains the key information of the data. The feature F after dimensionality reduction embed This adapts to the input format of the time series model and provides a foundation for subsequent iTransformer modeling. This allows the model to reduce computational complexity while ensuring effective processing and learning of time series data.

[0074] In an optional embodiment, the time series model of the combined feature vector is established in step S300, and the photovoltaic power generation power is predicted based on the time series model, which may also include: concatThe input is sent to the Embedding layer. The main function of the Embedding layer is to map the high-dimensional feature vector to a lower-dimensional space, reducing the computational burden while retaining the key information of the data. This mapping process can be represented by the function E(·):

[0075] F embed =E(F concat )

[0076] Dynamically focus on the input feature F through the iTransformer self-attention mechanism embed The importance of different time steps in the self-attention. The output A of the self-attention is calculated as:

[0077]

[0078] Where Q, K, and V are query, key, and value matrices, respectively, and d k This mechanism enables the model to dynamically weight features based on their correlations and identify the most important time points for future power prediction.

[0079] iTransformer consists of multiple stacked iTransformerBlocks, each of which mainly includes a self-attention layer and a feedforward neural network. The output of layer l can be expressed as:

[0080] Z l =LayerNorm(Z l-1 +Attention(Z l-1 ))

[0081] Z l =LayerNorm(Z l +FeedForward(Z l ))

[0082] Among them, Z l-1 It is the output of the previous layer. LayerNorm is used for regularization to improve the stability of training. FeedForward is a fully connected feedforward network that can introduce nonlinear relationships.

[0083] After processing through multiple iTransformerBlocks, the output O is the final feature representation, which can effectively capture the time series information in photovoltaic power forecasting. Output O will be further used to predict photovoltaic power generation.

[0084] In another optional embodiment, a time series model of the combined feature vector is established in step S300, and photovoltaic power generation power is predicted based on the time series model. The dependency relationship in the long time series can also be obtained through a long short-term memory network to predict the photovoltaic power generation power.

[0085] In an embodiment of the present application, a time series model of the combined feature vector is established in step S300, and photovoltaic power generation power is predicted based on the time series model. It also includes: processing the spliced ​​time series features through the time series model to predict photovoltaic power generation power, and outputting the photovoltaic power prediction result.

[0086] In an optional embodiment, step S300 establishes a time series model of the combined feature vectors, and predicts the photovoltaic power generation based on the time series model, which may also include: processing the spliced ​​features O through the iTransformer model to predict the photovoltaic power generation, and the final output layer is a linear layer W, which is used to map the features to the predicted power value P pred :

[0087] P pred =W·O+b

[0088] Where b is the bias term. The final output predicted power P pred .

[0089] By utilizing sky cloud maps, this invention successfully addresses the difficulty of photovoltaic power prediction caused by complex environmental variables in photovoltaic power generation systems, significantly improves the accuracy of photovoltaic system online monitoring and energy management, and enhances the stability of system operation.

[0090] Example 2, reference Figure 2 , is an embodiment of the present invention. This embodiment is different from the first embodiment in that it provides a photovoltaic power generation power prediction system, including:

[0091] A preprocessing module is used to obtain photovoltaic historical power data and sky cloud image data, and preprocess the photovoltaic historical power data and sky cloud image data;

[0092] An image feature extraction module is used to extract features from pre-processed sky cloud images based on a pre-trained feature extraction model, and to combine the extracted image features with pre-processed photovoltaic historical power data to form a combined feature vector.

[0093] The time series reasoning module is used to establish a time series model of the combined feature vector and predict the photovoltaic power generation power based on the time series model.

[0094] Specifically, when executed, each module of the photovoltaic power generation prediction system of this embodiment implements the steps of the photovoltaic power generation prediction method in Example 1, for example:

[0095] In one embodiment, the photovoltaic power generation prediction system may perform the following steps:

[0096] Standardize the sky cloud image sequence to make the data size of each image the same;

[0097] A quality threshold is preset. When there is an abnormal image in the sky cloud image, if the quality index of the abnormal image is not less than the quality threshold, the abnormal image data will be retained; if the quality index of the abnormal image is less than the quality threshold, the low-quality abnormal image will be eliminated;

[0098] Downsample the image sequence in the temporal dimension.

[0099] Preset thresholds to identify abnormal values ​​in the photovoltaic historical power data series;

[0100] If the absolute deviation of the photovoltaic power data value at any time point from the mean of the entire photovoltaic historical power data series exceeds a preset threshold, it will be marked as an outlier, and the outlier will be corrected, and the normal photovoltaic historical power data will be standardized.

[0101] Synchronize sky cloud image sequences with historical photovoltaic power data in the time dimension;

[0102] Assume that the time series of sky cloud images is I={I1,I2,…,I T}, the photovoltaic historical data is P = {p1, p2, ..., p T}, by matching the corresponding time points t to form the aligned input feature pairs (i t ,p t ).

[0103] The sky cloud image is fed into a feature extraction model using a 3×3 convolution kernel, ReLU activation function, and max pooling layer.

[0104] The feature extraction model outputs the feature vector F VGG19 =VGG19(I′ i ), and perform dimensionality reduction processing on the feature vector, which is then merged with the features of the photovoltaic historical power data to form a combined feature vector.

[0105] Perform dimensionality reduction on the combined feature vector and calculate the importance weight of each time step through the self-attention mechanism;

[0106] Enhanced temporal feature extraction is based on multiple stacked model units, each of which includes a self-attention layer and a feed-forward neural network.

[0107] The spliced ​​time series features are processed through a time series model to predict photovoltaic power generation and output photovoltaic power prediction results.

[0108] This embodiment further provides an electronic device applicable to the photovoltaic power prediction method, including:

[0109] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the photovoltaic power generation power prediction method proposed in the above embodiment.

[0110] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for predicting photovoltaic power generation proposed in the above embodiment is implemented.

[0111] The storage medium proposed in this embodiment and the photovoltaic power generation power prediction method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0112] 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, which can be stored in a computer-readable storage medium, such as a computer's 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 a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0113] Example 3 is an embodiment of the present invention. This embodiment verifies the beneficial effects of the present invention through comparative experiments.

[0114] This embodiment conducted a comparative test on the SKIPP'D dataset. Compared with other models, the model of the present invention showed obvious advantages in the photovoltaic power prediction task, as shown in Table 1:

[0115] Table 1 Model indicator parameters

[0116] Model Name RMSE MAE <![CDATA[R 2 ]]> Model of the present invention 2.5370kW 1.6818kW 0.9037 FEDFormer 8.6007kW 7.1921kW -0.1072 Autoformer 6.7633kW 5.0445kW 0.3154 Dlinear 2.6354kW 1.7735kW 0.8895

[0117] It can be seen from Table 1 that compared with the FEDFormer model, the RMSE and MAE of the proposed model are lower than those of the FEDFormer. 2The RMSE and MAE of this model are also lower than those of Autoformer. 2 Higher, showing stronger accuracy and generalization ability; although compared with Dlinear, the gap between the two is smaller, but this model is still slightly better, with lower RMSE and MAE, R 2 Slightly higher than Dlinear R 2 , showing stronger prediction performance and stability overall.

[0118] Experimental results demonstrate that the proposed method outperforms traditional models across multiple test metrics. By using the VGG19 network to extract fine-grained features from sky images, the proposed method overcomes the feature extraction limitations of traditional convolutional neural networks. iTransformer is also used to model time series data, capturing the complex relationship between historical photovoltaic power generation data and weather variations. Combining these two approaches, the model not only effectively reflects the impact of weather conditions on photovoltaic power generation, but also improves forecast accuracy and provides a more reliable basis for grid scheduling and resource allocation.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A photovoltaic power generation power prediction method, characterized in that: include: Acquiring photovoltaic historical power data and sky cloud image data, and preprocessing the photovoltaic historical power data and sky cloud image data; Performing feature extraction on the pre-processed sky cloud image based on a pre-trained feature extraction model, and concatenating the extracted image features with the pre-processed photovoltaic historical power data to form a combined feature vector; A time series model of the combined feature vector is established, and photovoltaic power generation power is predicted based on the time series model.

2. The photovoltaic power generation prediction method according to claim 1, wherein: Preprocessing the sky cloud image data includes: Standardize the sky cloud image sequence to make the data size of each image the same; A quality threshold is preset, and when there is an abnormal image in the sky cloud image, if the quality index of the abnormal image is not less than the quality threshold, the abnormal image data is retained; if the quality index of the abnormal image is less than the quality threshold, the low-quality abnormal image is eliminated; Downsample the image sequence in the temporal dimension.

3. The photovoltaic power generation prediction method according to claim 2, wherein: Preprocessing the photovoltaic historical power data includes: presetting thresholds to identify abnormal values ​​in the photovoltaic historical power data series; If the absolute deviation of the photovoltaic power data value at any time point from the mean of the entire photovoltaic historical power data series exceeds a preset threshold, it will be marked as an outlier, and the outlier will be corrected, and the normal photovoltaic historical power data will be standardized.

4. The photovoltaic power generation prediction method according to claim 3, wherein: Before extracting features from the pre-processed sky cloud image based on the pre-trained feature extraction model, the method further includes: synchronizing the sky cloud image sequence with the photovoltaic historical power data in the time dimension; Assume that the time series of sky cloud images is I={I1,I2,…,I T }, the photovoltaic historical data is P = {p1, p2, ..., p T }, by matching the corresponding time points t to form the aligned input feature pairs (i t ,p t ).

5. The photovoltaic power generation prediction method according to claim 3 or 4, characterized in that: Performing feature extraction on the pre-processed sky cloud image based on a pre-trained feature extraction model, and splicing the extracted image features with the pre-processed photovoltaic historical power data to form a combined feature vector, including: Input the sky cloud image into a feature extraction model using a 3×3 convolution kernel, a ReLU activation function, and a max pooling layer; The feature extraction model outputs the feature vector F VGG19 =VGG19(I i ′ ), and perform dimensionality reduction processing on the feature vector, and merge it with the photovoltaic historical power data features to form a combined feature vector.

6. The photovoltaic power generation prediction method according to claim 5, wherein: Establishing a time series model of the combined feature vector and predicting photovoltaic power generation based on the time series model includes: Performing dimensionality reduction on the combined feature vector and calculating the importance weight of each time step through a self-attention mechanism; Enhanced temporal feature extraction is based on multiple stacked model units, each of which includes a self-attention layer and a feed-forward neural network.

7. The photovoltaic power generation prediction method according to claim 6, wherein: Establishing a time series model of the combined feature vector and predicting photovoltaic power generation based on the time series model also includes: processing the spliced ​​time series features through the time series model to predict photovoltaic power generation and outputting the photovoltaic power prediction result.

8. A photovoltaic power generation power prediction system, characterized in that: include: A preprocessing module is used to obtain photovoltaic historical power data and sky cloud image data, and preprocess the photovoltaic historical power data and sky cloud image data; An image feature extraction module is used to extract features from the preprocessed sky cloud image based on a pretrained feature extraction model, and to combine the extracted image features with the preprocessed photovoltaic historical power data to form a combined feature vector; The time series reasoning module is used to establish a time series model of the combined feature vector and predict photovoltaic power generation based on the time series model.

9. An electronic device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the photovoltaic power generation power prediction method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the photovoltaic power generation power prediction method according to any one of claims 1 to 7.

Images