Photovoltaic power adaptive confidence interval prediction method and device, medium and equipment

Abstract

The invention discloses a photovoltaic power adaptive confidence interval prediction method and device, a medium and equipment, and relates to the technical field of power prediction, and the method comprises the steps: obtaining historical photovoltaic power generation data and historical meteorological data; performing correlation analysis on the historical photovoltaic power generation data and the historical meteorological data by adopting a Pearson correlation coefficient analysis method to obtain a key meteorological data sequence; performing model training on the deep learning network model based on the CNN-Informer hybrid architecture by using the key meteorological data sequence and the historical photovoltaic power generation data as training samples to obtain a photovoltaic power adaptive confidence interval prediction model; the photovoltaic power adaptive confidence interval prediction model is adopted to predict photovoltaic power generation data and meteorological data obtained in real time, and a prediction interval is obtained; and based on the prediction interval upper limit and the prediction interval lower limit of the prediction interval as reference boundaries of the charge-discharge power of the energy storage system, controlling the charge-discharge state of the energy storage system. According to the invention, the precision and reliability of interval prediction can be improved.

Description

Technical Field

[0001] This invention relates to the field of power prediction technology, and in particular to a method, apparatus, medium, and device for predicting photovoltaic power adaptive confidence intervals. Background Technology

[0002] Photovoltaic power generation, as an important component of clean energy, is widely used globally due to its renewable, low-pollution, and low-carbon characteristics. However, photovoltaic power output is significantly affected by various natural factors such as solar irradiance, temperature, and cloud cover, exhibiting significant volatility, intermittency, and uncertainty. With the continuous expansion of new energy installed capacity and its accelerated replacement of traditional fossil fuels, the random fluctuations in power output pose a severe challenge to the real-time balance, safe and stable operation, and power quality of the power system. At the distribution network level, the integration of high-proportion distributed photovoltaic systems can easily lead to problems such as voltage exceeding limits, line overload, and power backfeeding. At the system dispatch level, the unpredictability of photovoltaic output increases the demand for reserve capacity, raises system operating costs, and the phenomenon of "curtailment" still occurs frequently, resulting in resource waste.

[0003] However, existing methods still suffer from several bottlenecks, limiting the effectiveness of prediction results in directly serving downstream advanced applications such as energy storage dispatch and grid optimization. The loss function design of some interval prediction models struggles to achieve an effective balance between "interval coverage" and "interval width," or suffers from non-differentiability issues, affecting training stability and final interval quality. For multi-step prediction tasks, existing models still have room for improvement in capturing long-term dependencies, complex periodic patterns, and abrupt changes in photovoltaic power time series, restricting the reliability and practicality of prediction intervals over long time scales. Crucially, the output of most current interval prediction methods fails to adequately and seamlessly integrate with downstream decision-making models such as energy storage system optimization control and grid risk dispatch. These downstream applications typically require prediction results to provide clear, reliable, and dynamically changing power fluctuation boundaries as key constraints for their optimization models. The deficiencies in interval quality and adaptability of existing methods make it difficult to provide directly reliable decision boundaries, thus impacting the final effectiveness of the entire "prediction-control" chain. Summary of the Invention

[0004] In view of this, the present invention provides a photovoltaic power adaptive confidence interval prediction method, device, medium and equipment, the main purpose of which is to solve the problem that the existing photovoltaic power prediction methods ignore the random fluctuations of power generation, resulting in strong uncertainty in the point prediction results and making it difficult to directly use them for precise control of energy storage systems.

[0005] To address the above problems, this application provides a photovoltaic power adaptive confidence interval prediction method, comprising: Acquire historical photovoltaic power generation data and historical meteorological data for different time periods; The Pearson correlation coefficient analysis method was used to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. The key meteorological data sequence and the historical photovoltaic power generation data are used as training samples to train a deep learning network model based on the CNN-Informer hybrid architecture, so as to obtain a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions. The photovoltaic power adaptive confidence interval prediction model is used to predict the photovoltaic power generation data and meteorological data acquired in real time to obtain the prediction interval of photovoltaic power for future periods. The upper and lower limits of the prediction interval are used as reference boundaries for the charging and discharging power of the energy storage system to control the charging and discharging state of the energy storage system.

[0006] Optionally, the step of using Pearson correlation coefficient analysis to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences strongly correlated with photovoltaic power generation specifically includes: Data preprocessing is performed on the historical photovoltaic power generation data and the historical meteorological data; Using the preprocessed historical meteorological data as the independent variable and the historical photovoltaic power generation data as the dependent variable, the Pearson correlation coefficients between the historical meteorological data and the historical photovoltaic power generation data in different dimensions were calculated. The absolute values ​​of each Pearson correlation coefficient are calculated to obtain the absolute Pearson correlation coefficient corresponding to each Pearson correlation coefficient. The historical meteorological data corresponding to the first predetermined number of Pearson correlation coefficients, arranged in descending order of absolute Pearson correlation coefficient values, are identified as the key meteorological data sequence.

[0007] Optionally, the step of using the key meteorological data sequence and the historical photovoltaic power generation data as training samples to train a deep learning network model based on a CNN-Informer hybrid architecture to obtain a photovoltaic power adaptive confidence interval prediction model that meets preset conditions specifically includes: Step 1: Use the CNN feature extraction layer of the deep learning network model to extract features from the key meteorological data sequence and the historical photovoltaic power generation data to obtain the historical high-dimensional feature tensor; Step 2: The encoder of the Informer layer of the deep learning network model is used to encode the historical high-dimensional feature tensor to obtain the historical context vector sequence. Step 3: The decoder of the Informer layer of the deep learning network model decodes the historical context vector sequence and the future time step to obtain the future sequence representation of historical photovoltaic power. Step 4: The output layer of the Informer layer of the deep learning network model is used to perform parallel computation on the future sequence representation of the historical photovoltaic power to obtain the predicted historical photovoltaic power values ​​corresponding to different quantiles; Step 5: Update the model parameters of the deep learning network model based on the historical photovoltaic power prediction value and the actual photovoltaic power corresponding to the future time step using a preset quantile loss function and a preset model loss function; Step 6: Repeat steps 1 to 5 to update the updated deep learning network model until the model converges, thus obtaining a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions.

[0008] Optionally, the encoder using the Informer layer of the deep learning network model encodes the historical high-dimensional feature tensor to obtain a historical context vector sequence, specifically including: The historical high-dimensional feature tensor is subjected to sine-cosine position encoding to obtain the first encoding vector; The first encoding vector is processed by a probabilistic sparse self-attention method to obtain the second encoding vector; The second encoding vector is distilled using convolution and max pooling to obtain the third encoding vector; The third encoding vector is subjected to residual calculation and normalization to obtain the historical context vector sequence.

[0009] Optionally, the decoder using the Informer layer of the deep learning network model decodes based on the historical context vector sequence and the future time step to obtain a future sequence representation of historical photovoltaic power, specifically including: Step 1: Use the masked multi-head self-attention method to determine the historical prediction data and starting token before the current prediction time point; Step 2: Using the future representation of photovoltaic power at the current time point to be predicted as the query target, query the key-value pairs of the historical context vector sequence to obtain the predicted historical power representation at the current time point to be predicted; Step 3: Use a feedforward neural network to perform nonlinear transformation on the predicted historical power representation to obtain the predicted historical power feature representation for the current time point to be predicted; Step 4: Perform residual calculation and layer normalization on the predicted historical power feature representation to obtain the historical prediction value corresponding to the current prediction time point. The historical prediction value includes the first historical prediction value corresponding to the target median, the second historical prediction value corresponding to the pessimistic quantile, and the third historical prediction value corresponding to the optimistic quantile. Step 5: Fill the masked future positions corresponding to the current time point to be predicted in the historical context vector sequence with the historical predicted values ​​to update the historical context vector sequence; Step 6: Repeat steps 1 to 5 to predict the future representation of photovoltaic power at the next time point to be predicted based on the updated historical context vector sequence, so as to obtain the future sequence representation of historical photovoltaic power.

[0010] Optionally, before controlling the charging and discharging state of the energy storage system based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system, the method further includes: performing a performance evaluation on the prediction interval for a future time period. The performance evaluation of the prediction interval for the future time period specifically includes: Collect actual photovoltaic power values ​​for future periods; Based on the actual photovoltaic power value and the predicted interval, evaluation indicators are calculated to obtain multi-dimensional evaluation indicators; the multi-dimensional evaluation indicators include interval coverage rate indicator, interval average width indicator, and interval coverage width indicator. The prediction interval is evaluated based on the multidimensional evaluation indicators.

[0011] Optionally, the step of using the photovoltaic power adaptive confidence interval prediction model to predict real-time photovoltaic power generation data and meteorological data to obtain the prediction interval for photovoltaic power in future periods specifically includes: The CNN feature extraction layer of the photovoltaic power adaptive confidence interval prediction model is used to extract features from the photovoltaic power generation data and the meteorological data to obtain the target high-dimensional feature tensor; The encoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model is used to encode the target high-dimensional feature tensor to obtain the target context vector sequence. The decoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model decodes the target context vector sequence and the future time step to obtain the future sequence representation of the target photovoltaic power. The output layer of the photovoltaic power adaptive confidence interval prediction model is used to perform parallel computation processing on the future sequence representation of the target photovoltaic power to obtain the photovoltaic power prediction value corresponding to different target quantiles, so as to obtain the prediction interval of photovoltaic power in the future time period.

[0012] To address the aforementioned problems, this application provides a photovoltaic power adaptive confidence interval prediction device, comprising: The acquisition module is used to acquire historical photovoltaic power generation data and historical meteorological data for different time periods; The correlation analysis module is used to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data using the Pearson correlation coefficient analysis method to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. The model training module is used to train a deep learning network model based on a CNN-Informer hybrid architecture using the key meteorological data sequence and the historical photovoltaic power generation data as training samples, so as to obtain a photovoltaic power adaptive confidence interval prediction model that meets preset conditions. The prediction module is used to predict the real-time photovoltaic power generation data and meteorological data using the photovoltaic power adaptive confidence interval prediction model to obtain the prediction interval of photovoltaic power for future periods. The control module is used to control the charging and discharging state of the energy storage system based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system.

[0013] To address the aforementioned problems, this application provides a storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned photovoltaic power adaptive confidence interval prediction method.

[0014] To address the aforementioned problems, this application provides an electronic device, comprising at least a memory and a processor. The memory stores a computer program, and the processor, when executing the computer program in the memory, implements the steps of the aforementioned photovoltaic power adaptive confidence interval prediction method.

[0015] The beneficial effects of this application are as follows: This application captures complex spatiotemporal features through a CNN-Informer hybrid model and directly outputs probability predictions by combining it with quantile regression, significantly improving the accuracy and reliability of interval predictions. The CNN-Informer hybrid model effectively captures the complex spatiotemporal features and long-range dependence of photovoltaic power, and the combination with quantile regression directly outputs probability predictions, significantly improving the accuracy and reliability of interval predictions. The provided adaptive confidence interval can dynamically and quantitatively reflect the fluctuation range and prediction uncertainty of photovoltaic output, providing key inputs for advanced applications such as power system dispatching and risk assessment. The output prediction interval can be directly used as boundary conditions for the optimized control of flexible resources such as energy storage systems, helping to improve the stable operation and economic dispatch level of the power grid.

[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating an adaptive confidence interval prediction method for photovoltaic power provided in an embodiment of this application is shown. Figure 2 This illustration shows a schematic diagram of the model structure of the deep learning network model with the CNN-Informer hybrid architecture provided in an embodiment of this application; Figure 3 A flowchart illustrating an adaptive confidence interval prediction method for photovoltaic power according to another embodiment of this application is shown; Figure 4 A schematic diagram illustrating the use of interval prediction results provided in an embodiment of this application for energy storage optimization control is shown. Figure 5 A structural block diagram of a photovoltaic power adaptive confidence interval prediction method provided in another embodiment of this application is shown. Detailed Implementation

[0018] Various embodiments and features of this application are described herein with reference to the accompanying drawings.

[0019] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this application will be apparent to those skilled in the art.

[0020] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0021] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0022] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application.

[0023] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0024] Specific embodiments of this application are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in a variety of substantially any suitable detailed structures.

[0025] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.

[0026] This application provides a photovoltaic power adaptive confidence interval prediction method, such as... Figure 1 As shown, it includes: Step S101: Obtain historical photovoltaic power generation data and historical meteorological data for different time periods; In the specific implementation process, this step involves loading photovoltaic power generation and related meteorological data from reliable data sources. Data sources can include local monitoring systems of photovoltaic power plants, meteorological data servers of meteorological departments, and historical photovoltaic power generation data, including but not limited to historical photovoltaic power generation data for different time periods; historical meteorological data includes but is not limited to meteorological parameters such as irradiance, temperature, and humidity.

[0027] Step S102: Use the Pearson correlation coefficient analysis method to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. In this step, the historical photovoltaic power generation data and the historical meteorological data are preprocessed. The preprocessed historical meteorological data is used as the independent variable, and the historical photovoltaic power generation data is used as the dependent variable. Pearson correlation coefficients between the historical meteorological data and the historical photovoltaic power generation data in different dimensions are calculated. The absolute values ​​of each Pearson correlation coefficient are calculated to obtain the absolute Pearson correlation coefficients corresponding to each Pearson correlation coefficient. The historical meteorological data corresponding to the first predetermined number of Pearson correlation coefficients, arranged in descending order of absolute Pearson correlation coefficient values, are determined as the key meteorological data sequence.

[0028] Step S103: Use the key meteorological data sequence and the historical photovoltaic power generation data as training samples to train the deep learning network model based on the CNN-Informer hybrid architecture, and obtain a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions. In the specific implementation process, step one involves using the CNN feature extraction layer of the deep learning network model to extract features from the key meteorological data sequence and the historical photovoltaic power generation data, obtaining a historical high-dimensional feature tensor; step two involves using the encoder of the Informer layer of the deep learning network model to encode the historical high-dimensional feature tensor, obtaining a historical context vector sequence; step three involves using the decoder of the Informer layer of the deep learning network model to decode based on the historical context vector sequence and the future time step, obtaining a future sequence representation of historical photovoltaic power; step four involves using the output layer of the Informer layer of the deep learning network model to perform parallel computation processing on the future sequence representation of historical photovoltaic power, obtaining historical photovoltaic power prediction values ​​corresponding to different quantiles; step five involves updating the model parameters of the deep learning network model based on the historical photovoltaic power prediction values ​​and the actual photovoltaic power corresponding to the future time step using a preset quantile loss function; and step six involves repeating steps one to five to update the updated deep learning network model until the model converges, obtaining a photovoltaic power adaptive confidence interval prediction model that meets preset conditions. Figure 2 The diagram shown is a schematic of the deep learning network model of the CNN-Informer hybrid architecture of this application.

[0029] Step S104: Use the photovoltaic power adaptive confidence interval prediction model to predict the real-time photovoltaic power generation data and meteorological data to obtain the prediction interval of photovoltaic power for future periods; In this step, the CNN feature extraction layer of the photovoltaic power adaptive confidence interval prediction model is used to extract features from the photovoltaic power generation data and the meteorological data to obtain a target high-dimensional feature tensor. The encoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model encodes the target high-dimensional feature tensor to obtain a target context vector sequence. The decoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model decodes the target context vector sequence and the future time step to obtain a future sequence representation of the target photovoltaic power. The output layer of the photovoltaic power adaptive confidence interval prediction model performs parallel computation on the future sequence representation of the target photovoltaic power to obtain photovoltaic power prediction values ​​corresponding to different target quantiles, so as to obtain the prediction interval of photovoltaic power for future periods.

[0030] Step S105: Based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system, control the charging and discharging state of the energy storage system.

[0031] In the specific implementation process, the upper and lower limits of the output photovoltaic power prediction range are directly used as the reference boundaries for the charging and discharging power of the energy storage system to smooth out photovoltaic fluctuations.

[0032] This application captures complex spatiotemporal features using a CNN-Informer hybrid model and combines it with quantile regression to directly output probabilistic predictions, significantly improving the accuracy and reliability of interval predictions. The CNN-Informer hybrid model effectively captures the complex spatiotemporal features and long-range dependencies of photovoltaic power, and the combination with quantile regression to directly output probabilistic predictions significantly enhances the accuracy and reliability of interval predictions. The provided adaptive confidence interval can dynamically and quantitatively reflect the fluctuation range and prediction uncertainty of photovoltaic output, providing key input for advanced applications such as power system dispatching and risk assessment. The output prediction interval can be directly used as boundary conditions for the optimized control of flexible resources such as energy storage systems, helping to improve the stable operation and economic dispatch level of the power grid.

[0033] Another embodiment of this application provides a different method for adaptive confidence interval prediction of photovoltaic power, such as... Figure 3 As shown, it includes: Step S201: Obtain historical photovoltaic power generation data and historical meteorological data for different time periods; In the specific implementation process, this step involves loading photovoltaic power generation and related meteorological data from reliable data sources. Data sources can include local monitoring systems of photovoltaic power plants, meteorological data servers of meteorological departments, and historical photovoltaic power generation data, including but not limited to historical photovoltaic power generation data for different time periods; historical meteorological data includes but is not limited to meteorological parameters such as irradiance, temperature, and humidity.

[0034] Step S202: Use the Pearson correlation coefficient analysis method to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. In this step, data preprocessing is performed on the historical photovoltaic power generation data and the historical meteorological data. Data preprocessing methods include, but are not limited to, missing value handling, outlier handling, and data standardization. Using the preprocessed historical meteorological data as the independent variable and the historical photovoltaic power generation data as the dependent variable, Pearson correlation coefficients are calculated between the historical meteorological data and the historical photovoltaic power generation data across different dimensions. For each preprocessed time series and photovoltaic power generation, the correlation between them is calculated using the Pearson correlation coefficient method. r The calculation formula is:

[0035] in, x Represents photovoltaic power generation p Other than historical meteorological data variables; and Representing variables respectively x and photovoltaic power generation p In the i The value of seconds, and Representing variables respectively x and photovoltaic power generation p exist n The mean value within a second time window is calculated; the absolute values ​​of each Pearson correlation coefficient are processed to obtain the absolute Pearson correlation coefficient corresponding to each Pearson correlation coefficient; the historical meteorological data corresponding to the first predetermined number of Pearson correlation coefficients, arranged in descending order of absolute Pearson correlation coefficient values, are determined as the key meteorological data sequence. Variables corresponding to correlation coefficients with relatively large absolute values ​​are selected, i.e., parameter data strongly correlated with photovoltaic power generation.

[0036] Step S203: Use the key meteorological data sequence and the historical photovoltaic power generation data as training samples to train the deep learning network model based on the CNN-Informer hybrid architecture, and obtain a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions. This step, in its specific implementation, includes the following steps: Step 1: The CNN feature extraction layer of the deep learning network model is used to extract features from the key meteorological data sequence and the historical photovoltaic power generation data to obtain a historical high-dimensional feature tensor. Specifically, the CNN feature extraction layer takes three-dimensional time-series data of shape (N, T, F) as input, where N is the batch size, T is the historical time step, and F is the feature dimension, which includes photovoltaic power and selected key meteorological factors. The processing involves using multiple one-dimensional convolution kernels to perform sliding convolution operations on the input sequence along the time axis, with ReLU as the activation function. The convolution kernels automatically learn local fluctuation patterns and related features in the data. Output: Historical high-dimensional feature tensor (N, T', D), where T' is the length of the sequence after convolution, and D is the number of convolution kernels, also known as the feature mapping dimension.

[0037] Step 2: The encoder of the Informer layer of the deep learning network model is used to encode the historical high-dimensional feature tensor to obtain the historical context vector sequence. Specifically, the historical high-dimensional feature tensor is subjected to sine-cosine positional encoding to obtain a first encoded vector; a probabilistic sparse self-attention method is used to perform depth calculation on the first encoded vector to obtain a second encoded vector; convolution and max pooling are used to distill the second encoded vector to obtain a third encoded vector; residual calculation and normalization are performed on the third encoded vector to obtain the historical context vector sequence. Positional encoding is added at each time step, injecting temporal sequence information. Probabilistic sparse self-attention calculates the probability distribution of query-key pairs, selecting only a predetermined number of the most relevant query-key pairs for attention calculation, significantly reducing computational complexity. Distillation compresses the sequence length layer by layer through convolutional layers and max pooling operations, retaining key information and reducing redundancy. Residual connections and layer normalization are applied to the output of each layer to stabilize the training process. The output is the encoded context vector sequence, capturing the global long-range dependencies of the photovoltaic data input sequence.

[0038] Step 3: The decoder of the Informer layer of the deep learning network model decodes the historical context vector sequence and the future time step to obtain the future sequence representation of historical photovoltaic power. Specifically, it includes the following steps: Step 31: Use the masked multi-head self-attention method to determine the historical prediction data and starting token before the current prediction time point; Step 32: Using the future representation of photovoltaic power at the current time point to be predicted as the query target, query the key-value pairs of the historical context vector sequence to obtain the predicted historical power representation at the current time point to be predicted; Step 33: Use a feedforward neural network to perform nonlinear transformation on the predicted historical power representation to obtain the predicted historical power feature representation for the current time point to be predicted; Step 34: Perform residual calculation and layer normalization on the predicted historical power feature representation to obtain the historical prediction value corresponding to the current prediction time point. The historical prediction value includes the first historical prediction value corresponding to the target median, the second historical prediction value corresponding to the pessimistic quantile, and the third historical prediction value corresponding to the optimistic quantile. Step 35: Fill the masked future position corresponding to the current time point to be predicted in the historical context vector sequence with the historical predicted value to update the historical context vector sequence; Step 36: Repeat steps 31 to 36 to predict the future representation of photovoltaic power at the next time point to be predicted based on the updated historical context vector sequence, so as to obtain the future sequence representation of historical photovoltaic power.

[0039] Masked multi-head self-attention is a technique where the decoder's self-attention layer uses a masking mechanism to ensure that the current position only focuses on previous positions. Specifically, encoder-decoder attention performs attention calculations on decoder queries and encoder key-value pairs, focusing on relevant parts of the input sequence. A feedforward neural network performs a nonlinear transformation independently for each position through two fully connected layers. Finally, residual connections and layer normalization are applied to each layer, ensuring that each layer also uses residual connections and layer normalization. Its output is the decoded future sequence representation of photovoltaic power.

[0040] Step 4: The output layer of the Informer layer of the deep learning network model is used to perform parallel computation on the future sequence representation of the historical photovoltaic power to obtain the predicted historical photovoltaic power values ​​corresponding to different quantiles; Specifically, the historical photovoltaic power future sequence representation output by the decoder is used as input, and multiple fully connected sub-networks are connected in parallel. Each sub-network corresponds to a target quantile τ (τ=0.05, 0.5, 0.95), and the predicted value at that quantile is calculated respectively. The target median can be τ=0.5, the pessimistic quantile can be τ=0.05, and the optimistic quantile can be τ=0.95; thus obtaining the first historical predicted value corresponding to the target median, the second historical predicted value corresponding to the pessimistic quantile, and the third historical predicted value corresponding to the optimistic quantile.

[0041] The historical photovoltaic power prediction values ​​corresponding to different quantiles were obtained.

[0042] Step 5: Update the model parameters of the deep learning network model based on the historical photovoltaic power prediction value and the actual photovoltaic power corresponding to the future time step using a preset quantile loss function and a preset model loss function; Specifically, the mathematical expression of the preset quantile loss function is as follows:

[0043] Where y is the actual value, Let τ be the predicted value and τ be the quantile (0 < τ < 1). By adjusting the value of τ, prediction results at different quantiles can be obtained. For each future time step, multiple predicted values ​​corresponding to quantiles are output, forming a prediction interval. When τ = 0.05 and τ = 0.95, a 90% confidence interval is output. The mathematical expression for the preset model loss function is as follows:

[0044] in, The median loss is the weighting factor, representing the weight of the median loss in the total loss. The function is set to 1 if the quantile is in the set {0.05, 0.95}, and 0 otherwise. τ represents the quantile loss weight; τ is the quantile level, with a value of 0.05 or 0.95. Quantile loss; i For the first i The sample at the quantile The predicted value below; τ is the mean squared error, used to measure the accuracy of the median (τ=0.5) prediction, and its mathematical expression is as follows:

[0045] in, For the first i The predicted value of the median (τ=0.5) of a sample.

[0046] By updating all network parameters using the backpropagation algorithm, the model can accurately predict both the conditional median and conditional quantiles simultaneously. After training, for a new input sequence, the model directly outputs multiple quantile predictions without subsequent calibration.

[0047] Step 6: Repeat steps 1 to 5 to update the updated deep learning network model until the model converges, thus obtaining a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions.

[0048] Specifically, when the difference between the loss function values ​​in two adjacent execution processes is less than or equal to the preset loss threshold, a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions is obtained.

[0049] Step S204: Use the CNN feature extraction layer of the photovoltaic power adaptive confidence interval prediction model to extract features from the photovoltaic power generation data and the meteorological data to obtain the target high-dimensional feature tensor; In this step, the CNN feature extraction layer of the photovoltaic power adaptive confidence interval prediction model is used. For the three-dimensional time series data of photovoltaic power generation data and meteorological data with input shape (N, T, F), multiple one-dimensional convolution kernels are used to perform sliding convolution operation on the input sequence along the time axis. The activation function is ReLU. The local fluctuation patterns and related features in the data are automatically learned through the convolution kernels, and the target high-dimensional feature tensor is output.

[0050] Step S205: The encoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model is used to encode the target high-dimensional feature tensor to obtain the target context vector sequence; In this step, the target high-dimensional feature tensor is sine-cosine positional encoded to obtain a first target encoding vector; the first target encoding vector is processed by depth calculation using a probabilistic sparse self-attention method to obtain a second target encoding vector; the second target encoding vector is distilled using convolution and max pooling to obtain a third target encoding vector; and the third target encoding vector is processed by residual calculation and normalization to obtain the target context vector sequence.

[0051] Step S206: The decoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model decodes the target context vector sequence and the future time step to obtain the future sequence representation of the target photovoltaic power. In its specific implementation, step S206 includes the following steps: Step S2061: Use the masked multi-head self-attention method to determine the historical prediction data and starting token before the current prediction time point; Step S2062: Using the future representation of photovoltaic power at the current time point to be predicted as the query target, query the key-value pairs of the historical context vector sequence to obtain the predicted historical power representation at the current time point to be predicted; Step S2063: Use a feedforward neural network to perform nonlinear transformation on the predicted historical power representation to obtain the predicted historical power feature representation of the current time point to be predicted; Step S2064: Perform residual calculation and layer normalization on the predicted historical power feature representation to obtain the historical prediction value corresponding to the current prediction time point. The historical prediction value includes the first historical prediction value corresponding to the target median, the second historical prediction value corresponding to the pessimistic quantile, and the third historical prediction value corresponding to the optimistic quantile. Step S2065: Fill the masked future position corresponding to the current time point to be predicted in the historical context vector sequence with the historical predicted value to update the historical context vector sequence; Step S2066: Repeat steps S2061 to S2066 to predict the future representation of photovoltaic power at the next time point to be predicted based on the updated historical context vector sequence, so as to obtain the future sequence representation of historical photovoltaic power.

[0052] Step S207: The output layer of the photovoltaic power adaptive confidence interval prediction model is used to perform parallel computation processing on the future sequence representation of the target photovoltaic power to obtain the photovoltaic power prediction value corresponding to different target quantiles, so as to obtain the prediction interval of photovoltaic power in the future time period; In this step, the target photovoltaic power future sequence representation output by the decoder is used as input, and multiple fully connected sub-networks are connected in parallel. Each sub-network corresponds to a target quantile τ (τ=0.05, 0.5, 0.95), and the predicted value at that quantile is calculated respectively. The photovoltaic power predicted values ​​corresponding to different quantiles are obtained, thus obtaining the prediction interval of photovoltaic power for future time periods.

[0053] Step S208: Perform a performance evaluation on the prediction interval for the future time period; In this step, the actual photovoltaic power value for a future period is collected; based on the actual photovoltaic power value and the prediction interval, evaluation indicators are calculated to obtain multi-dimensional evaluation indicators; these multi-dimensional evaluation indicators include interval coverage rate indicators, interval average width indicators, and interval coverage width indicators; the prediction interval is then evaluated based on these multi-dimensional evaluation indicators. Specifically, the formula for calculating interval coverage rate (PICP) is:

[0054] Where n is the number of samples. For indicator functions, when the actual value Falling within the prediction range At that time, ;otherwise, ; The PICP value reflects the accuracy of model uncertainty quantification: If PICP ≥ the target confidence level, it indicates that the interval construction is reasonable; If PICP is too low, it indicates that the range underestimates uncertainty; If PICP is too high, it indicates that the range is overly conservative.

[0055] The formula for calculating the average width of the interval (PINAW) is as follows:

[0056] Where R is the range of the actual value, i.e., R = max(y). min(y). For the i-th sample at the quantile The predicted value corresponds to the upper limit of the prediction interval in the loss function; This is the predicted value (lower limit of the prediction interval) for the i-th sample at the quantile τ.

[0057] The numerical value of PINAW reflects the granularity of uncertainty estimation: The smaller the width, the more accurate the information provided by the model (ideal state); However, it needs to be balanced with PICP (too narrow a range may sacrifice coverage).

[0058] The formula for calculating the coverage width (CWC) is as follows:

[0059] in, and The penalty coefficient is... For target coverage probability; CWC values ​​resolve the contradiction of a single indicator: when coverage meets the target, When the exponent term is very small, CWC ≈ PINAW; when the coverage is insufficient, exponent term Rapidly increase, impose significant penalties on CWC, and force an increase in coverage.

[0060] Based on the evaluation results, the photovoltaic power generation data and the meteorological data are optimized to obtain a prediction range of photovoltaic power for future periods that meets the index conditions.

[0061] Step S209: Based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system, control the charging and discharging state of the energy storage system.

[0062] In the specific implementation process, this step directly uses the upper and lower limits of the output photovoltaic power prediction range as the reference boundaries for the charging and discharging power of the energy storage system to smooth out photovoltaic fluctuations. For example... Figure 4The diagram illustrates the application of the interval prediction results of this application to energy storage optimization control. The explicit, reliable, and dynamically changing power fluctuation boundary provided in this application is used as a key constraint for energy storage optimization, thereby improving photovoltaic absorption rate, ensuring stable grid operation, and providing necessary data for optimizing energy storage control strategies.

[0063] This application captures complex spatiotemporal features using a CNN-Informer hybrid model and combines it with quantile regression to directly output probabilistic predictions, significantly improving the accuracy and reliability of interval predictions. The CNN-Informer hybrid model effectively captures the complex spatiotemporal features and long-range dependencies of photovoltaic power, and the combination with quantile regression to directly output probabilistic predictions significantly enhances the accuracy and reliability of interval predictions. The provided adaptive confidence interval can dynamically and quantitatively reflect the fluctuation range and prediction uncertainty of photovoltaic output, providing key input for advanced applications such as power system dispatching and risk assessment. The output prediction interval can be directly used as boundary conditions for the optimized control of flexible resources such as energy storage systems, helping to improve the stable operation and economic dispatch level of the power grid.

[0064] Another embodiment of this application provides a photovoltaic power adaptive confidence interval prediction device, as shown in Figure 5, including: Module 1 is used to acquire historical photovoltaic power generation data and historical meteorological data for different time periods; Correlation analysis module 2 is used to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data using the Pearson correlation coefficient analysis method to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. Model training module 3 is used to train a deep learning network model based on a CNN-Informer hybrid architecture using the key meteorological data sequence and the historical photovoltaic power generation data as training samples, so as to obtain a photovoltaic power adaptive confidence interval prediction model that meets preset conditions. Prediction module 4 is used to predict the real-time photovoltaic power generation data and meteorological data using the photovoltaic power adaptive confidence interval prediction model to obtain the prediction interval of photovoltaic power for future periods. The control module 5 is used to control the charging and discharging state of the energy storage system based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system.

[0065] In the specific implementation process, the correlation analysis module 2 is specifically used to: preprocess the historical photovoltaic power generation data and the historical meteorological data; use the preprocessed historical meteorological data as the independent variable and the historical photovoltaic power generation data as the dependent variable to calculate the Pearson correlation coefficients between the historical meteorological data and the historical photovoltaic power generation data in different dimensions; perform absolute value calculation on each Pearson correlation coefficient to obtain the absolute Pearson correlation coefficient corresponding to each Pearson correlation coefficient; and determine the historical meteorological data corresponding to the first predetermined number of Pearson correlation coefficients in descending order of absolute Pearson correlation coefficient values ​​as the key meteorological data sequence.

[0066] In the specific implementation process, the model training module 3 is specifically used for the following steps: Step 1: Using the CNN feature extraction layer of the deep learning network model to extract features from the key meteorological data sequence and the historical photovoltaic power generation data to obtain a historical high-dimensional feature tensor; Step 2: Using the encoder of the Informer layer of the deep learning network model to encode the historical high-dimensional feature tensor to obtain a historical context vector sequence; Step 3: Using the decoder of the Informer layer of the deep learning network model to decode based on the historical context vector sequence and the future time step to obtain a future sequence representation of historical photovoltaic power; Step 4: Using the output layer of the Informer layer of the deep learning network model to perform parallel computation processing on the future sequence representation of historical photovoltaic power to obtain the predicted historical photovoltaic power value corresponding to different quantiles; Step 5: Updating the model parameters of the deep learning network model based on the predicted historical photovoltaic power value and the actual photovoltaic power corresponding to the future time step using a preset quantile loss function and a preset model loss function; Step 6: Repeating steps 1 to 5 to update the updated deep learning network model until the model converges, obtaining a photovoltaic power adaptive confidence interval prediction model that meets preset conditions.

[0067] In the specific implementation process, the model training module 3 is further used to: perform sine-cosine position encoding on the historical high-dimensional feature tensor to obtain a first encoding vector; perform depth calculation processing on the first encoding vector using a probabilistic sparse self-attention method to obtain a second encoding vector; perform distillation operation on the second encoding vector using convolution and max pooling to obtain a third encoding vector; and perform residual calculation and normalization processing on the third encoding vector to obtain the historical context vector sequence.

[0068] In the specific implementation process, the model training module 3 is further used for: Step 1, using a masked multi-head self-attention method to determine the historical prediction data and starting token before the current prediction time point; Step 2, using the future representation of photovoltaic power at the current prediction time point as the query target, querying the key-value pairs of the historical context vector sequence to obtain the predicted historical power representation at the current prediction time point; Step 3, using a feedforward neural network to perform nonlinear transformation processing on the predicted historical power representation to obtain the predicted historical power feature representation at the current prediction time point; Step 4, performing residual calculation and layer normalization processing on the predicted historical power feature representation to obtain... Step 5: Fill the masked future positions corresponding to the current time point to be predicted with the historical predicted values, which include a first historical predicted value corresponding to the target median, a second historical predicted value corresponding to the pessimistic quantile, and a third historical predicted value corresponding to the optimistic quantile; Step 6: Repeat steps 1 to 5 to predict the future representation of photovoltaic power at the next time point to be predicted based on the updated historical context vector sequence, so as to obtain the future sequence representation of historical photovoltaic power.

[0069] In specific implementation, the device further includes an evaluation module, which is specifically used to perform performance evaluation on the prediction interval for future periods. Specifically, this includes: collecting the actual photovoltaic power value for future periods; calculating evaluation indicators based on the actual photovoltaic power value and the prediction interval to obtain multi-dimensional evaluation indicators; the multi-dimensional evaluation indicators include interval coverage rate indicators, interval average width indicators, and interval coverage width indicators; and evaluating the prediction interval based on the multi-dimensional evaluation indicators.

[0070] In the specific implementation process, the prediction module 4 is specifically used to extract features from the photovoltaic power generation data and the meteorological data using the CNN feature extraction layer of the photovoltaic power adaptive confidence interval prediction model to obtain a target high-dimensional feature tensor; to encode the target high-dimensional feature tensor using the encoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model to obtain a target context vector sequence; to decode the target context vector sequence and the future time step using the decoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model to obtain a future sequence representation of the target photovoltaic power; and to perform parallel computation processing on the future sequence representation of the target photovoltaic power using the output layer of the photovoltaic power adaptive confidence interval prediction model to obtain photovoltaic power prediction values ​​corresponding to different target quantiles, so as to obtain the prediction interval of photovoltaic power for future periods.

[0071] This application captures complex spatiotemporal features using a CNN-Informer hybrid model and combines it with quantile regression to directly output probabilistic predictions, significantly improving the accuracy and reliability of interval predictions. The CNN-Informer hybrid model effectively captures the complex spatiotemporal features and long-range dependencies of photovoltaic power, and the combination with quantile regression to directly output probabilistic predictions significantly enhances the accuracy and reliability of interval predictions. The provided adaptive confidence interval can dynamically and quantitatively reflect the fluctuation range and prediction uncertainty of photovoltaic output, providing key input for advanced applications such as power system dispatching and risk assessment. The output prediction interval can be directly used as boundary conditions for the optimized control of flexible resources such as energy storage systems, helping to improve the stable operation and economic dispatch level of the power grid.

[0072] Another embodiment of this application provides a storage medium storing a computer program, which, when executed by a processor, implements the following method steps: Step 1: Obtain historical photovoltaic power generation data and historical meteorological data for different time periods; Step 2: Use the Pearson correlation coefficient analysis method to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. Step 3: Using the key meteorological data sequence and the historical photovoltaic power generation data as training samples, train the deep learning network model based on the CNN-Informer hybrid architecture to obtain a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions. Step 4: Use the photovoltaic power adaptive confidence interval prediction model to predict the real-time photovoltaic power generation data and meteorological data to obtain the prediction interval of photovoltaic power for future periods. Step 5: Based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system, control the charging and discharging state of the energy storage system.

[0073] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0074] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0075] The specific implementation process of the above method steps can be found in the embodiment of the above arbitrary photovoltaic power adaptive confidence interval prediction method, which will not be repeated here.

[0076] This application captures complex spatiotemporal features using a CNN-Informer hybrid model and combines it with quantile regression to directly output probabilistic predictions, significantly improving the accuracy and reliability of interval predictions. The CNN-Informer hybrid model effectively captures the complex spatiotemporal features and long-range dependencies of photovoltaic power, and the combination with quantile regression to directly output probabilistic predictions significantly enhances the accuracy and reliability of interval predictions. The provided adaptive confidence interval can dynamically and quantitatively reflect the fluctuation range and prediction uncertainty of photovoltaic output, providing key input for advanced applications such as power system dispatching and risk assessment. The output prediction interval can be directly used as boundary conditions for the optimized control of flexible resources such as energy storage systems, helping to improve the stable operation and economic dispatch level of the power grid.

[0077] Another embodiment of this application provides an electronic device, which can be a server. The electronic device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. The program of the electronic device, executed by the processor, can implement the functions or steps of a photovoltaic power adaptive confidence interval prediction method on the server side.

[0078] In one embodiment, an electronic device is provided, which can be a client. The electronic device includes a processor, memory, a network interface, a display screen, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with an external server via a network connection. When the program of the electronic device is executed by the processor, it implements the functions or steps of a photovoltaic power adaptive confidence interval prediction method on the client side.

[0079] Another embodiment of this application provides an electronic device, including at least a memory and a processor. The memory stores a computer program, and the processor, when executing the computer program in the memory, performs the following method steps: Step 1: Obtain historical photovoltaic power generation data and historical meteorological data for different time periods; Step 2: Use the Pearson correlation coefficient analysis method to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. Step 3: Using the key meteorological data sequence and the historical photovoltaic power generation data as training samples, train the deep learning network model based on the CNN-Informer hybrid architecture to obtain a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions. Step 4: Use the photovoltaic power adaptive confidence interval prediction model to predict the real-time photovoltaic power generation data and meteorological data to obtain the prediction interval of photovoltaic power for future periods. Step 5: Based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system, control the charging and discharging state of the energy storage system.

[0080] The specific implementation process of the above method steps can be found in the embodiment of the above arbitrary photovoltaic power adaptive confidence interval prediction method, which will not be repeated here.

[0081] This application captures complex spatiotemporal features using a CNN-Informer hybrid model and combines it with quantile regression to directly output probabilistic predictions, significantly improving the accuracy and reliability of interval predictions. The CNN-Informer hybrid model effectively captures the complex spatiotemporal features and long-range dependencies of photovoltaic power, and the combination with quantile regression to directly output probabilistic predictions significantly enhances the accuracy and reliability of interval predictions. The provided adaptive confidence interval can dynamically and quantitatively reflect the fluctuation range and prediction uncertainty of photovoltaic output, providing key input for advanced applications such as power system dispatching and risk assessment. The output prediction interval can be directly used as boundary conditions for the optimized control of flexible resources such as energy storage systems, helping to improve the stable operation and economic dispatch level of the power grid.

[0082] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A photovoltaic power adaptive confidence interval prediction method, characterized in that, include: Acquire historical photovoltaic power generation data and historical meteorological data for different time periods; The Pearson correlation coefficient analysis method was used to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. The key meteorological data sequence and the historical photovoltaic power generation data are used as training samples to train a deep learning network model based on the CNN-Informer hybrid architecture, so as to obtain a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions. The photovoltaic power adaptive confidence interval prediction model is used to predict the photovoltaic power generation data and meteorological data acquired in real time to obtain the prediction interval of photovoltaic power for future periods. The upper and lower limits of the prediction interval are used as reference boundaries for the charging and discharging power of the energy storage system to control the charging and discharging state of the energy storage system.

2. The method as described in claim 1, characterized in that, The Pearson correlation coefficient analysis method is used to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation, specifically including: Data preprocessing is performed on the historical photovoltaic power generation data and the historical meteorological data; Using the preprocessed historical meteorological data as the independent variable and the historical photovoltaic power generation data as the dependent variable, the Pearson correlation coefficients between the historical meteorological data and the historical photovoltaic power generation data in different dimensions were calculated. The absolute values ​​of each Pearson correlation coefficient are calculated to obtain the absolute Pearson correlation coefficient corresponding to each Pearson correlation coefficient. The historical meteorological data corresponding to the first predetermined number of Pearson correlation coefficients, arranged in descending order of absolute Pearson correlation coefficient values, are identified as the key meteorological data sequence.

3. The method as described in claim 1, characterized in that, The process of using the key meteorological data sequence and the historical photovoltaic power generation data as training samples to train a deep learning network model based on a CNN-Informer hybrid architecture to obtain a photovoltaic power adaptive confidence interval prediction model that meets preset conditions specifically includes: Step 1: Use the CNN feature extraction layer of the deep learning network model to extract features from the key meteorological data sequence and the historical photovoltaic power generation data to obtain the historical high-dimensional feature tensor; Step 2: The encoder of the Informer layer of the deep learning network model is used to encode the historical high-dimensional feature tensor to obtain the historical context vector sequence. Step 3: The decoder of the Informer layer of the deep learning network model decodes the historical context vector sequence and the future time step to obtain the future sequence representation of historical photovoltaic power. Step 4: The output layer of the Informer layer of the deep learning network model is used to perform parallel computation on the future sequence representation of the historical photovoltaic power to obtain the predicted historical photovoltaic power values ​​corresponding to different quantiles; Step 5: Update the model parameters of the deep learning network model based on the historical photovoltaic power prediction value and the actual photovoltaic power corresponding to the future time step using a preset quantile loss function and a preset model loss function; Step 6: Repeat steps 1 to 5 to update the updated deep learning network model until the model converges, thus obtaining a photovoltaic power adaptive confidence interval prediction model that meets the preset conditions.

4. The method as described in claim 3, characterized in that, The encoder using the Informer layer of the deep learning network model encodes the historical high-dimensional feature tensor to obtain a historical context vector sequence, specifically including: The historical high-dimensional feature tensor is subjected to sine-cosine position encoding to obtain the first encoding vector; The first encoding vector is processed by a probabilistic sparse self-attention method to obtain the second encoding vector; The second encoding vector is distilled using convolution and max pooling to obtain the third encoding vector; The third encoding vector is subjected to residual calculation and normalization to obtain the historical context vector sequence.

5. The method as described in claim 3, characterized in that, The decoder of the Informer layer of the deep learning network model decodes based on the historical context vector sequence and the future time step to obtain a future sequence representation of historical photovoltaic power, specifically including: Step 1: Use the masked multi-head self-attention method to determine the historical prediction data and starting token before the current prediction time point; Step 2: Using the future representation of photovoltaic power at the current time point to be predicted as the query target, query the key-value pairs of the historical context vector sequence to obtain the predicted historical power representation at the current time point to be predicted; Step 3: Use a feedforward neural network to perform nonlinear transformation on the predicted historical power representation to obtain the predicted historical power feature representation for the current time point to be predicted; Step 4: Perform residual calculation and layer normalization on the predicted historical power feature representation to obtain the historical prediction value corresponding to the current prediction time point. The historical prediction value includes the first historical prediction value corresponding to the target median, the second historical prediction value corresponding to the pessimistic quantile, and the third historical prediction value corresponding to the optimistic quantile. Step 5: Fill the masked future positions corresponding to the current time point to be predicted in the historical context vector sequence with the historical predicted values ​​to update the historical context vector sequence; Step 6: Repeat steps 1 to 5 to predict the future representation of photovoltaic power at the next time point to be predicted based on the updated historical context vector sequence, so as to obtain the future sequence representation of historical photovoltaic power.

6. The method as described in claim 1, characterized in that, Before controlling the charging and discharging state of the energy storage system based on the upper and lower limits of the predicted interval as reference boundaries for the charging and discharging power of the energy storage system, the method further includes: performing a performance evaluation on the predicted interval for future time periods. The performance evaluation of the prediction interval for the future time period specifically includes: Collect actual photovoltaic power values ​​for future periods; Based on the actual photovoltaic power value and the predicted interval, evaluation indicators are calculated to obtain multi-dimensional evaluation indicators; the multi-dimensional evaluation indicators include interval coverage rate indicator, interval average width indicator, and interval coverage width indicator. The prediction interval is evaluated based on the multidimensional evaluation indicators.

7. The method as described in claim 1, characterized in that, The photovoltaic power adaptive confidence interval prediction model is used to predict real-time photovoltaic power generation data and meteorological data to obtain the prediction interval for photovoltaic power in future periods, specifically including: The CNN feature extraction layer of the photovoltaic power adaptive confidence interval prediction model is used to extract features from the photovoltaic power generation data and the meteorological data to obtain the target high-dimensional feature tensor; The encoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model is used to encode the target high-dimensional feature tensor to obtain the target context vector sequence. The decoder of the Informer layer of the photovoltaic power adaptive confidence interval prediction model decodes the target context vector sequence and the future time step to obtain the future sequence representation of the target photovoltaic power. The output layer of the photovoltaic power adaptive confidence interval prediction model is used to perform parallel computation processing on the future sequence representation of the target photovoltaic power to obtain the photovoltaic power prediction value corresponding to different target quantiles, so as to obtain the prediction interval of photovoltaic power in the future time period.

8. A photovoltaic power adaptive confidence interval prediction device, characterized in that, include: The acquisition module is used to acquire historical photovoltaic power generation data and historical meteorological data for different time periods; The correlation analysis module is used to perform correlation analysis on the historical photovoltaic power generation data and the historical meteorological data using the Pearson correlation coefficient analysis method to obtain a predetermined number of key meteorological data sequences that are strongly correlated with photovoltaic power generation. The model training module is used to train a deep learning network model based on a CNN-Informer hybrid architecture using the key meteorological data sequence and the historical photovoltaic power generation data as training samples, so as to obtain a photovoltaic power adaptive confidence interval prediction model that meets preset conditions. The prediction module is used to predict the real-time photovoltaic power generation data and meteorological data using the photovoltaic power adaptive confidence interval prediction model to obtain the prediction interval of photovoltaic power for future periods. The control module is used to control the charging and discharging state of the energy storage system based on the upper and lower limits of the prediction interval as reference boundaries for the charging and discharging power of the energy storage system.

9. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the photovoltaic power adaptive confidence interval prediction method according to any one of claims 1-7.

10. An electronic device, characterized in that, It includes at least a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program in the memory, implements the steps of the photovoltaic power adaptive confidence interval prediction method according to any one of claims 1-7.

Images