Single-variable short-term photovoltaic power generation prediction method based on enhanced local correlation features
By using sliding window treatment, overlapping feature enhancement method and Lagrangian insertion method in short-term photovoltaic power generation prediction, combined with linearly separable convolutional residual network, the problem of insufficient learning of local features and complex nonlinear relationships in the existing technology is solved, and the prediction accuracy and stability are significantly improved.
Patent Information
- Application Number
- CN202411649815.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-02-25
AI Technical Summary
The prior art has limitations in capturing local features and learning of complex nonlinear relationships in short-term photovoltaic power generation prediction, especially in the absence of local changes and correlations.
A univariate short-term photovoltaic power generation prediction method based on enhanced local correlation characteristics is adopted, and the feature information volume of data is increased through sliding window processing and overlapping feature enhancement method, and the local time characteristics are enhanced by Lagrangian insertion method. Combined with linear separable convolutional residual networks, feature training is performed to optimize the prediction model.
It significantly improves the accuracy and stability of prediction, can more effectively capture local features and complex nonlinear relationships in the data, and improves the accuracy of short-term photovoltaic power generation prediction.
Smart Images
Figure CN119513495A_ABST
Abstract
Description
Technical Field
[0001] The invention discloses a single variable short-term photovoltaic power generation prediction method based on enhanced local correlation characteristics, and relates to the technical field of photovoltaic power generation prediction. Background Art
[0002] With the widespread application of photovoltaic power generation in the energy field, short-term photovoltaic power generation forecasting has become an important part of grid dispatching and optimization. Photovoltaic power generation has significant intermittency and volatility, so accurate short-term forecasting is crucial to the stability and efficiency of the grid. At present, the methods for photovoltaic power generation forecasting mainly include physical model methods, statistical model methods, machine learning methods, and deep learning methods.
[0003] The physical model method is based on the physical laws of solar radiation and meteorological data, but its effect is limited when dealing with complex meteorological conditions. The statistical model method uses the statistical characteristics of historical data for prediction, but its ability to handle nonlinear relationships is insufficient. Machine learning methods can capture nonlinear relationships in data in a data-driven way and have been widely used in recent years. Deep learning models further improve the ability to learn complex data patterns, especially in large-scale data processing.
[0004] Although deep learning methods have improved prediction accuracy, they still have limitations in capturing local features and learning complex nonlinear relationships, especially in handling local changes and correlations in short-term photovoltaic power generation forecasts. Summary of the invention
[0005] In view of the problems of the prior art, the present invention provides a single variable short-term photovoltaic power generation prediction method and device based on enhanced local correlation features. The technical solution adopted is: In a first aspect, a univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features comprises: S1, dividing the original photovoltaic power generation data into a training set and a test set, and preprocessing the original photovoltaic power generation data; S2, performing sliding window processing on the training set, and applying overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the training set; S3, obtaining enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; S4, performing feature training on a photovoltaic power generation prediction model using a linearly separable convolutional residual network according to the local time features to obtain training weights and an optimization model; S5, performing sliding window processing on the test set, and applying overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the test set; S6, obtaining enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; S7, importing the feature data in the current sliding window into the optimization model, and performing prediction using the training weights.
[0006] In some implementations, according to the Lagrange interpolation method, interpolation is performed between two adjacent data in each set of data in the sliding window, specifically including: The Lagrange interpolation method is performed four times on each group of data. Specifically, interpolation is performed between two adjacent data in each group, and three new interpolation data are inserted at equal intervals between two real values.
[0007] Among them, the Lagrange interpolation basis function is:
[0008] The interpolation polynomial expression is:
[0009] In some implementations, the S4 specifically includes: S41, introducing an improved linearly separable convolution kernel according to the photovoltaic power generation prediction model, and capturing key linear relationships and key linear features through the training weights; S42, according to the photovoltaic power generation prediction model, using a residual block to perform model training with ResNet50 as the main network combined with a deep residual learning mechanism; S43, repeat the training according to S42 to obtain the optimized model.
[0010] In some implementations, the S7 specifically includes: S71, inputting the sliding window into the optimization model to obtain a target value corresponding to the prediction; S72, slide the sliding window by one data point, repeat step S71, and obtain the target value corresponding to the prediction; S73, slide the sliding window to the last data point to complete the prediction of the training weight.
[0011] In a second aspect, an embodiment of the present invention provides a univariate short-term photovoltaic power generation prediction device based on enhanced local correlation features, the device comprising: A data processing module, used for dividing the original photovoltaic power generation data into a training set and a test set, and preprocessing the original photovoltaic power generation data; A first sliding window processing module is used to perform sliding window processing on the training set, and apply overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the training set; A first feature processing module, configured to obtain enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to a Lagrange interpolation method; A model optimization module, used to perform feature training on a photovoltaic power generation prediction model using a linearly separable convolutional residual network according to the local time features to obtain training weights and an optimization model; A second sliding window processing module is used to perform sliding window processing on the test set, and apply overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the test set; A second feature processing module is used to obtain enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; The power generation prediction module is used to import the characteristic data in the current sliding window into the optimization model and perform prediction through the training weights.
[0012] In some implementations, according to the Lagrange interpolation method, interpolation is performed between two adjacent data in each set of data in the sliding window, specifically including: The Lagrange interpolation method is performed four times on each group of data. Specifically, interpolation is performed between two adjacent data in each group, and three new interpolation data are inserted at equal intervals between two real values.
[0013] Among them, the Lagrange interpolation basis function is:
[0014] The interpolation polynomial expression is:
[0015] In some implementations, the model optimization module specifically includes: A weight training unit, used for introducing an improved linearly separable convolution kernel according to the photovoltaic power generation prediction model, and capturing key linear relationships and key linear features through the training weights; A model training unit, used to perform model training according to the photovoltaic power generation prediction model by using a residual block with ResNet50 as the main network combined with a deep residual learning mechanism; The model generating unit is used to perform repeated training according to the model training unit to obtain the optimized model.
[0016] In some implementations, the power generation prediction module specifically includes: A target acquisition unit, used to input the sliding window into the optimization model to obtain a target value corresponding to the prediction; A sliding window moving unit, used to slide the sliding window by one data point, repeat the step operation of the target acquisition unit, and obtain the target value corresponding to the prediction; The prediction completion unit is used to slide the sliding window to the last data point to complete the prediction of the training weight.
[0017] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein when the one or more computer instructions are executed by the processor, the method described in the first aspect above is implemented.
[0018] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the method described in the first aspect.
[0019] One or more embodiments of the present invention can bring at least the following beneficial effects: The method of the present invention achieves the purpose of improving the accuracy and stability of prediction by strengthening the local correlation between data and improving the model's ability to capture local features. First, the method of the present invention adopts interpolation prediction technology, which enhances the correlation between fluctuating data through interpolation methods and fully mines local related information. Secondly, the enhanced data is preprocessed to ensure that the model can better capture local features in the data. Thirdly, an optimized prediction model is used, which has a stronger ability to capture local features and can accurately predict the enhanced data. Finally, through comprehensive processing, a more accurate photovoltaic power generation prediction result is obtained, which significantly improves the overall prediction performance and can solve the limitations of traditional models in dealing with complex fluctuating data. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0021] Figure 1 This is an implementation flow chart of a single variable short-term photovoltaic power generation prediction method with enhanced local correlation characteristics provided by an embodiment of the present invention. Figure 2 It is a structural diagram of an improved linearly separable convolution residual module in an embodiment of the present invention. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.
[0023] Embodiment 1: Figure 1 The implementation flow chart of a univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features is shown, as Figure 1 As shown, the univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features provided in this embodiment includes: S1, dividing the original photovoltaic power generation data into a training set and a test set, and preprocessing the original photovoltaic power generation data; S2, performing sliding window processing on the training set, and applying overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the training set; S3, obtaining enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; S4, performing feature training on a photovoltaic power generation prediction model using a linearly separable convolutional residual network according to the local time features to obtain training weights and an optimization model; S5, performing sliding window processing on the test set, and applying overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the test set; S6, obtaining enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; S7, importing the feature data in the current sliding window into the optimization model, and performing prediction using the training weights.
[0024] First, according to S1, the reasonable division of the original photovoltaic power generation data is the first step in photovoltaic power generation prediction modeling. The data for this experiment comes from the hourly photovoltaic power generation data of a certain region from January 1, 2021 to December 31, 2021 throughout the year. When dividing the data set, it is necessary to fully consider the characteristics of the time series to ensure that the data distribution of the training set and the test set can truly reflect the actual power generation situation and avoid deviations caused by seasonal changes or weather characteristics. Specifically, the hourly data for the 10 months from January 1, 2021 to October 31, 2021, totaling 7296 groups, are used as training sets; 1464 groups of data from November 1 to December 31, 2021 are used as test sets.
[0025] In addition, data preprocessing is required, including steps such as removing outliers, filling missing values, and normalization to improve the stability of the model and prediction accuracy. Reasonable division and processing of data sets helps ensure that the model performs well during training and evaluation, and provides a solid data foundation for subsequent photovoltaic power generation prediction. Therefore, scientifically dividing the original data into training sets and test sets is an important prerequisite for the construction of photovoltaic power generation prediction models, providing necessary support for model training and verification.
[0026] Next, according to S2, the training set data is subjected to sliding window processing, and the overlapping feature enhancement method is applied to the data in each sliding window. This step can effectively overlap local features and enhance the feature information content of the data set.
[0027] Specifically, first, according to S21, first, the training set is subjected to sliding window processing, and the sliding window will gradually move during the processing. For example, given a set of training data sequences:
[0028] The sliding window contains 8 data, that is, the 7 power data from 0:00 to 6:00 on January 1, 2020 are used as features, and the power data at 7:00 on January 1, 2020 is used as a label. The window starts from the first data in the sequence and slides 1 data each time until all the data are traversed. Among them, The sliding window formula is as follows:
[0029] Secondly, according to S21, in each sliding window, the first 7 feature data will be divided into 3 groups, each group contains 5 data, and the groups slide with a step size of 1 data, as follows:
[0030] Finally, according to S23, the three sets of feature data that have been processed are combined with the label data in the sliding window, as follows:
[0031] Next, according to S3, the fourth Lagrange interpolation method is used to process the data in the sliding window. This is the key step of this method. This step can effectively enhance the local correlation features.
[0032] First, according to S31, each set of data is processed using the quartic Lagrange interpolation method. Specifically, interpolation is performed between two adjacent data in each set, and three new interpolation data are inserted at equal intervals between two real values.
[0033] The Lagrange interpolation basis function is:
[0034] The interpolation polynomial expression is:
[0035] By using this formula, we can and Three new data points are inserted at equal intervals in each interval between 0:00 and 0:00 on January 1, 2020, which is equivalent to adding the power data at 0:15, 0:30, and 0:45. Each set of processed data will be expanded from 5 to 17, that is, 5 real values plus 12 interpolated data. Through interpolation, the change trend between data can be refined and the local time characteristics can be enhanced.
[0036] Each set of data in the sliding window is specifically expanded as follows:
[0037] Next, according to S32, the three sets of interpolated feature data and the label data in the sliding window are combined, as follows:
[0038] In this way, each sliding window finally generates 52 data, of which the last data is the 8th data in the sliding window, i.e., the label data, which is used as the predicted target value. This step helps to build closer temporal correlation and contextual dependency in the time series.
[0039] Finally, according to S33, the sliding window moves back one data each time, repeats the above steps, and performs the same interpolation and combination processing on the new sliding window. The whole process will continue to loop until all data in the training set have completed the sliding window division and interpolation processing. This sliding window method ensures the coherence of the data and the consistency of the time characteristics, and generates rich samples for model training.
[0040] Next, according to S4, the photovoltaic power generation prediction model using linearly separable convolutional residual network can more efficiently capture the key features of the data processed by Lagrange interpolation and overlapping feature enhancement method, and finally obtain the optimal training weights. Specifically: First, according to S41, the model introduces an improved linear separable convolution kernel based on the characteristics of the processed data. The traditional point-by-point convolution layer is replaced by a linear layer, that is, a fully connected layer, so as to accurately capture the key linear relationships and features in the processed data.
[0041] Secondly, according to S42, the residual block is optimized in the model architecture. The original large-size convolution kernel is replaced by a linearly separable convolution kernel. This improvement makes the model better adapt to the interpolated data structure and effectively improves the efficiency of feature extraction.
[0042] The entire training process uses ResNet50 as the backbone network, combined with a deep residual learning mechanism, to ensure that the model captures complex data patterns while maintaining training stability and efficiency. Through this design, the model can fully utilize the rich features of the interpolated data and improve the prediction accuracy of photovoltaic power generation.
[0043] Finally, according to S43, the model successfully obtained the ideal training weights after multiple rounds of training and tuning, providing efficient and accurate PV power generation prediction capabilities for practical applications.
[0044] Next, according to S5, the test set data is processed by sliding window, and the overlapping feature enhancement method is applied to the data in each sliding window. This step can effectively overlap the local features and enhance the feature information of the data set, and use the same data processing method to correspond to the training weights. Specifically, it includes: First, according to S51, the test set is processed with a sliding window, and the sliding window moves gradually during the processing. For example, given a set of test data sequences:
[0045] The sliding window contains 7 feature data, that is, the 7 power data from 0:00 to 6:00 on November 1, 2020 are used as features. The window starts from the first data in the sequence and slides 1 data each time until all the data are traversed. Among them, The sliding window formula is as follows:
[0046] Secondly, according to S52, in each sliding window, the 7 feature data will be divided into 3 groups, each group contains 5 data, and the groups slide with a step size of 1 data, as follows:
[0047] Finally, according to S53, the three sets of feature data that have been processed are combined as follows:
[0048] Next, according to S6, the fourth Lagrange interpolation method is used to process the data in the sliding window. This is the key step of this method. This step can effectively enhance the local correlation features, and the same data processing method is used to correspond to the training weights. Specifically, it includes: First, according to S61, each set of data is processed using the quartic Lagrange interpolation method. Specifically, interpolation is performed between two adjacent data in each set, and three new interpolation data are inserted at equal intervals between two real values.
[0049] The Lagrange interpolation basis function is:
[0050] The interpolation polynomial expression is:
[0051] By using this formula, we can and Three new data points are inserted at equal intervals in each interval between 0:00 and 0:00 on January 1, 2020, which is equivalent to adding the power data at 0:15, 0:30, and 0:45. Each set of processed data will be expanded from 5 to 17, that is, 5 real values plus 12 interpolated data. Through interpolation, the change trend between data can be refined and the local time characteristics can be enhanced.
[0052] Each set of data in the sliding window is specifically expanded as follows:
[0053] Finally, according to S62, the three sets of feature data after interpolation are combined, as follows:
[0054] Each sliding window generates 51 feature data for the prediction task.
[0055] Next, according to S7, the feature data in the current sliding window is imported into the linear separable convolutional residual network photovoltaic power generation prediction model, and the training weights are used for prediction. Each time a sliding window is input, the corresponding target value can be predicted. Subsequently, the sliding window is moved back by one data point, the same interpolation and feature processing steps are repeated, and the model is input again for prediction. The whole process is repeated until all prediction tasks are completed, achieving continuous and efficient photovoltaic power generation prediction.
[0056] The method embodiment of the present invention has the following advantages: The sliding window processing and overlapping feature enhancement method are used to significantly improve the feature information content of the data set. Specifically, this method divides the original time series data into multiple overlapping windows, so that the data in each window can be fully utilized, thereby generating more feature combinations. Each sliding window is continuous in time, allowing the model to learn between adjacent time points, thereby more effectively identifying short-term patterns in time series. At the same time, the introduction of this overlapping feature enables the model to capture long-term trends, because the overlap between multiple sliding windows provides additional contextual information, which helps to reveal potential patterns and periodic changes. In addition, the design of overlapping features improves the diversity of data and further enhances the generalization ability of the model during training, enabling it to better adapt to different prediction scenarios and needs.
[0057] 2. The fourth Lagrange interpolation method effectively enriches the local time features by inserting new interpolation data in data processing and inserting new data points at equal intervals between two adjacent data points. This method not only fills the information gaps in the original data, but also enhances the ability to express the trend of data changes by smoothing the change curve of the data. Specifically, the Lagrange interpolation method can estimate the value of unobserved points based on known data points, thereby generating a continuous and smooth curve, which is crucial for capturing small changes. This meticulous local feature extraction enables the model to respond more sensitively to abnormal events or emergencies in the data, improving the accuracy and reliability of its predictions. Especially when faced with complex time series data, the interpolation method effectively reduces the prediction errors that may be caused by data sparsity, thereby providing a more robust input for the model.
[0058] 3. When using the linearly separable convolutional residual network model, an improved convolutional layer design is made based on the characteristics of the processed data. This design makes the model more efficient in feature extraction by optimizing the structure and size of the convolution kernel. Specifically, the linearly separable convolution significantly reduces the number of parameters by splitting the standard convolution into channel-by-channel convolution and point-by-point convolution, while retaining the expressive power of the model, thereby improving computational efficiency and reducing the risk of overfitting. In addition, the improved convolutional layer can more accurately capture local patterns and complex structures in the data, making the model more flexible and adaptable when extracting features, ensuring that it can handle diverse input data.
[0059] 4. Combined with the characteristics of the residual network, the model can effectively solve the gradient vanishing problem in deep networks. By introducing skip connections, the residual network allows the gradient to be transmitted through a shorter path during back propagation, thereby maintaining the stability of the training process. This feature enables the model to maintain effective learning in deeper network layers when capturing complex patterns.
[0060] Embodiment 2: An embodiment of the present invention provides a single variable short-term photovoltaic power generation prediction device based on enhanced local correlation features, the device comprising: A data processing module, used for dividing the original photovoltaic power generation data into a training set and a test set, and preprocessing the original photovoltaic power generation data; A first sliding window processing module is used to perform sliding window processing on the training set, and apply overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the training set; A first feature processing module, configured to obtain enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to a Lagrange interpolation method; A model optimization module, used to perform feature training on a photovoltaic power generation prediction model using a linearly separable convolutional residual network according to the local time features to obtain training weights and an optimization model; A second sliding window processing module is used to perform sliding window processing on the test set, and apply overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the test set; A second feature processing module is used to obtain enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; The power generation prediction module is used to import the characteristic data in the current sliding window into the optimization model and perform prediction through the training weights.
[0061] Specifically, according to the Lagrange interpolation method, interpolation is performed between two adjacent data in each set of data in the sliding window, specifically including: The Lagrange interpolation method is performed four times on each group of data. Specifically, interpolation is performed between two adjacent data in each group, and three new interpolation data are inserted at equal intervals between two real values.
[0062] Among them, the Lagrange interpolation basis function is:
[0063] The interpolation polynomial expression is:
[0064] Specifically, the model optimization module specifically includes: A weight training unit, used for introducing an improved linearly separable convolution kernel according to the photovoltaic power generation prediction model, and capturing key linear relationships and key linear features through the training weights; A model training unit, used to perform model training according to the photovoltaic power generation prediction model by using a residual block with ResNet50 as the main network combined with a deep residual learning mechanism; The model generating unit is used to perform repeated training according to the model training unit to obtain the optimized model.
[0065] Specifically, the power generation prediction module specifically includes: A target acquisition unit, used to input the sliding window into the optimization model to obtain a target value corresponding to the prediction; A sliding window moving unit, used to slide the sliding window by one data point, repeat the step operation of the target acquisition unit, and obtain the target value corresponding to the prediction; The prediction completion unit is used to slide the sliding window to the last data point to complete the prediction of the training weight.
[0066] Embodiment three: This embodiment further provides an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions implement the method of embodiment 1 when executed by the processor; In practical applications, the processor can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller unit (MCU), a microprocessor or other electronic components to execute the methods in the above embodiments.
[0067] The method implemented in this embodiment is as shown in the content of the first embodiment.
[0068] Embodiment 4: This embodiment further provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by one or more processors, the method of the first embodiment is implemented; Among them, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.
[0069] The method implemented in this embodiment is as shown in the content of the first embodiment.
[0070] In the several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed system and method can also be implemented in other ways. The system and method embodiments described above are merely illustrative.
[0071] It should be noted that, in this article, the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. The terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0072] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.
Claims
1. A univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features, characterized in that: The method comprises: S1, dividing the original photovoltaic power generation data into a training set and a test set, and preprocessing the original photovoltaic power generation data; S2, performing sliding window processing on the training set, and applying overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the training set; S3, obtaining enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; S4, performing feature training on a photovoltaic power generation prediction model using a linearly separable convolutional residual network according to the local time features to obtain training weights and an optimization model; S5, performing sliding window processing on the test set, and applying overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the test set; S6, obtaining enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; S7, importing the feature data in the current sliding window into the optimization model, and performing prediction using the training weights.
2. The univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features according to claim 1 is characterized in that: According to the Lagrange interpolation method, interpolation is performed between two adjacent data in each set of data in the sliding window, specifically including: The Lagrange interpolation method is performed four times on each set of data. Specifically, interpolation is performed between two adjacent data in each set, and three new interpolation data are inserted at equal intervals between two real values. Among them, the Lagrange interpolation basis function is: ; The interpolation polynomial expression is: .
3. The univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features according to claim 1 is characterized in that: The S4 specifically includes: S41, introducing an improved linearly separable convolution kernel according to the photovoltaic power generation prediction model, and capturing key linear relationships and key linear features through the training weights; S42, according to the photovoltaic power generation prediction model, the model is trained by using a residual block with ResNet50 as the main network combined with a deep residual learning mechanism; S43, repeat the training according to S42 to obtain the optimized model.
4. The univariate short-term photovoltaic power generation prediction method based on enhanced local correlation features according to claim 1 is characterized in that: The S7 specifically includes: S71, inputting the sliding window into the optimization model to obtain a target value corresponding to the prediction; S72, slide the sliding window by one data point, repeat step S71, and obtain the target value corresponding to the prediction; S73, slide the sliding window to the last data point to complete the prediction of the training weight.
5. A univariate short-term photovoltaic power generation prediction device based on enhanced local correlation features, characterized in that: The device comprises: A data processing module, used for dividing the original photovoltaic power generation data into a training set and a test set, and preprocessing the original photovoltaic power generation data; A first sliding window processing module is used to perform sliding window processing on the training set, and apply overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the training set; A first feature processing module, configured to obtain enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to a Lagrange interpolation method; A model optimization module, used to perform feature training on a photovoltaic power generation prediction model using a linearly separable convolutional residual network according to the local time features to obtain training weights and an optimization model; A second sliding window processing module is used to perform sliding window processing on the test set, and apply overlapping feature enhancement method to the data in each sliding window to obtain enhanced feature information of the test set; A second feature processing module is used to obtain enhanced local time features by interpolating between two adjacent data in each set of data in the sliding window according to the Lagrange interpolation method; The power generation prediction module is used to import the characteristic data in the current sliding window into the optimization model and perform prediction through the training weights.
6. The single variable short-term photovoltaic power generation prediction device with enhanced local correlation characteristics according to claim 5 is characterized in that: According to the Lagrange interpolation method, interpolation is performed between two adjacent data in each set of data in the sliding window, specifically including: The Lagrange interpolation method is performed four times on each set of data. Specifically, interpolation is performed between two adjacent data in each set, and three new interpolation data are inserted at equal intervals between two real values. Among them, the Lagrange interpolation basis function is: ; The interpolation polynomial expression is: .
7. The single variable short-term photovoltaic power generation prediction device with enhanced local correlation characteristics according to claim 5 is characterized in that: The model optimization module specifically includes: A weight training unit, used for introducing an improved linearly separable convolution kernel according to the photovoltaic power generation prediction model, and capturing key linear relationships and key linear features through the training weights; A model training unit, used to perform model training according to the photovoltaic power generation prediction model by using a residual block with ResNet50 as the main network combined with a deep residual learning mechanism; The model generating unit is used to perform repeated training according to the model training unit to obtain the optimized model.
8. The single variable short-term photovoltaic power generation prediction device with enhanced local correlation characteristics according to claim 5, characterized in that: The power generation prediction module specifically includes: A target acquisition unit, used to input the sliding window into the optimization model to obtain a target value corresponding to the prediction; A sliding window moving unit, used to slide the sliding window by one data point, repeat the step operation of the target acquisition unit, and obtain the target value corresponding to the prediction; The prediction completion unit is used to slide the sliding window to the last data point to complete the prediction of the training weight.
9. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein when the one or more computer instructions are executed by the processor, a univariate short-term photovoltaic power generation prediction method with enhanced local correlation features as described in any one of claims 1 to 4 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the single variable short-term photovoltaic power generation prediction method with enhanced local correlation characteristics as described in any one of claims 1 to 4 above.