Method and system for predicting short-term photovoltaic power generation power of photovoltaic power station

Abstract

The invention discloses a photovoltaic power station short-term photovoltaic power generation power prediction method and system, and the method comprises the steps: carrying out the normalization processing of the historical power data and meteorological data of a photovoltaic system, and screening out the meteorological characteristics with high correlation; secondly, decomposing non-stationary time sequence data into a plurality of stationary subsequences by using VMD, and optimizing decomposition parameters of the VMD through WOA so as to ensure optimal feature expression; then, combining a subsequence after VMD decomposition with the screened meteorological features, and inputting the combined subsequence and the screened meteorological features into an SCINet model for feature extraction and prediction; and finally, through multi-level convolution operation of the SCINet model, outputting a short-term prediction value of the photovoltaic power. The method has the advantages that the accuracy of photovoltaic power generation power prediction is effectively improved, and reliable technical support is provided for operation management of a photovoltaic power station.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power generation technology, and in particular to a method and system for predicting short-term photovoltaic power generation in a photovoltaic power station based on a hybrid deep learning model. Background Art

[0002] In recent years, with the development of global industry and technological progress, energy demand has increased dramatically, and the consumption of traditional fossil energy has become increasingly prominent. As a clean and renewable energy source, solar energy has the advantages of less pollution and low cost, and has become an important choice to replace traditional energy. Photovoltaic power generation technology relies on solar energy to convert electricity. Due to its outstanding performance in efficiency, reliability and economy, it has been widely used in the global energy field. However, solar power generation is cyclical and volatile, and its power generation is greatly affected by factors such as weather and seasons, resulting in uncertainty in output power. This volatility has brought challenges to the stable operation of the power grid system. Therefore, how to accurately predict the power generation of photovoltaic power stations has become a key issue in the efficient operation of photovoltaic power stations and grid dispatching management.

[0003] There are many types of current photovoltaic power generation prediction methods, which can be divided into ultra-short-term prediction, short-term prediction, medium-term prediction and long-term prediction according to the time range. Prediction methods can also be divided into physical models, statistical models, traditional artificial intelligence methods, machine learning methods and deep learning methods. Physics-based models usually use photovoltaic module parameters and meteorological conditions to simulate the power generation process, but their ability to capture complex nonlinear features is limited. Data-driven models, such as support vector regression (SVR), long short-term memory network (LSTM) and recurrent neural network (RNN), perform well in processing nonlinear time series data, but due to the complexity of photovoltaic power generation data, a single model is difficult to accurately capture all features. In addition, frequency domain decomposition methods such as empirical mode decomposition (EEMD), complete ensemble empirical mode decomposition (CEEMDAN) and variational mode decomposition (VMD) are gradually being applied to the processing and prediction of photovoltaic time series. However, these methods are prone to introduce noise or over-decomposition during data reconstruction, which affects the accuracy of prediction.

[0004] In recent years, optimization algorithms have also begun to be widely used in photovoltaic power generation prediction to reduce the noise introduced in frequency domain decomposition. For example, the Sparrow Search Algorithm (SSA), the GreyWolf Optimizer (GWO), and the Whale Optimization Algorithm (WOA) have made significant progress in improving prediction accuracy. However, the existing frequency domain decomposition method and optimization algorithm combination still have certain limitations in practical applications, such as insufficient redundant processing of data features and limited ability to capture time series features, which in turn affects the accuracy of photovoltaic power prediction. Summary of the invention

[0005] In view of the defects of the prior art, the present invention provides a method and system for predicting short-term photovoltaic power generation in a photovoltaic power station. The variational mode decomposition (VMD) is used for data preprocessing to solve the nonlinear and non-stationary problems in the photovoltaic time series; at the same time, the whale optimization algorithm (WOA) is introduced to optimize the VMD decomposition process to reduce data noise and redundant information; finally, the sample convolutional interactive neural network model (SCINet) is used to capture the complex features in the time series data, which greatly improves the prediction accuracy of photovoltaic power generation and the robustness of the model.

[0006] In order to achieve the above invention object, the technical solution adopted by the present invention is as follows:

[0007] A short-term photovoltaic power generation prediction method for photovoltaic power stations is based on the deep learning framework Pytorch, combined with Pearson correlation coefficient analysis, variational mode decomposition VMD, whale optimization algorithm WOA and sample convolutional interactive neural network SCINet to perform short-term photovoltaic power generation prediction. The specific steps are as follows:

[0008] S1 Data preprocessing: normalize the historical power data and meteorological data of the photovoltaic system, and select meteorological characteristic parameters with high correlation with photovoltaic power generation as the input of the model through Pearson correlation coefficient analysis;

[0009] S2 data decomposition: Use VMD method to decompose non-stationary photovoltaic power time series data into multiple stationary subsequences with different frequencies; use whale optimization algorithm WOA to optimize VMD decomposition parameters, dynamically determine the optimal decomposition number K and corresponding penalty parameters, and ensure that each subsequence has the best feature expression ability;

[0010] S3 feature optimization and prediction model training: Each subsequence after VMD decomposition is combined with the meteorological feature sequence after Pearson correlation screening and input into the SCINet model. SCINet extracts multi-scale features in the time series through its internal interactive convolution structure SCI-Block, and performs residual connection and optimization on the data through the binary tree structure to enhance the modeling ability of complex time series;

[0011] S4 prediction output: Through the multi-level convolution operation of the SCINet model, the short-term prediction value of photovoltaic power is output, and the Concat&Realign layer is used to integrate all feature information and generate the final photovoltaic power prediction result.

[0012] Furthermore, the historical power data and meteorological data include: global horizontal radiation, diffuse horizontal radiation, temperature, humidity and wind speed.

[0013] Furthermore, the variational mode decomposition (VMD) obtains the center frequency and bandwidth of the modal component through an iterative optimization method to achieve a smooth decomposition of the photovoltaic power data.

[0014] Furthermore, the whale optimization algorithm WOA is used to dynamically optimize the number of subsequences K and penalty parameters in VMD decomposition to prevent over-decomposition and ensure that the subsequences retain sufficient feature information.

[0015] Furthermore, the sample convolutional interactive neural network SCINet extracts multi-scale features in time series at different levels through convolution filters, and combines different SCI-Block modules through a binary tree structure for residual connection to generate the final prediction result.

[0016] Furthermore, the prediction results are evaluated by the indicators of root mean square error (RMSE) and mean absolute error (MAE), and the model parameters are adjusted according to the error feedback to optimize the prediction accuracy.

[0017] Furthermore, the structure of the SCINet model includes: an SC-Block module, a SCINet module and a stacked SCINet module.

[0018] The SC-Block module includes: the input feature data Z represents the pre-processed photovoltaic power generation related features. In the SC-Block module, the input data is downsampled and divided into two sub-sequences: odd sequence Z odd and even sequence Z even .

[0019] In order to reduce the information loss in the downsampling process, an interactive learning mechanism is introduced, which adjusts the affine transformation parameters between two subsequences through mutual learning.

[0020] Through interactive learning, two subsequences Z odd and Z even It can couple each other's information and significantly enhance the representation ability of learning features.

[0021] The SCINet module structure:

[0022] The SCINet module is composed of multiple SC-Block modules arranged in layers to form a binary tree structure. Each SC-Block extracts features through convolution operations and performs feature learning at low resolution.

[0023] After the input time series has completed all downsampling, convolution, and interaction operations, all low-resolution components are realigned and concatenated into a new sequence representation.

[0024] Residual connection: The reconnected low-resolution component is added to the original time series and predicted through residual connection. This connection method helps to retain the key information of the original data while incorporating the newly extracted features.

[0025] Output layer: Finally, a fully connected layer is used to integrate the processed features and output the prediction results of photovoltaic power generation.

[0026] The stacked SCINet module includes:

[0027] Through the relay supervision mechanism, multiple SCINet modules are stacked together to form a Stacked SCINet. Each stacked SCINet module can independently learn features at different levels and enhance the representation ability of features through information transfer between layers.

[0028] The Concat&Realign layer realigns and connects into a new sequence. The specific process is as follows:

[0029] Concat: Connect two subsequences Zeven and Zodd in a specific dimension to form a complete new sequence.

[0030] Realign: To compensate for the time series details that may be lost during the downsampling process, Realign performs data alignment operations based on the characteristics of the subsequences to ensure that the features of different time steps can be effectively combined, providing more accurate input for subsequent prediction layers.

[0031] Furthermore, the specific process of decomposition of the VMD method is as follows:

[0032] 1) Initialization and n = 0; represents the initial signal of the kth component; represents the initial value of the center frequency of each component; Represents the initial value of the penalty parameter, which is used to control the bandwidth during the decomposition process.

[0033] 2) n = n + 1, enter the loop;

[0034] 3) Basis and The update formula is updated until the number of decompositions reaches K and the inner loop is stopped;

[0035] 4) Update λ according to the update formula of λ;

[0036] 5) Given the accuracy ε, if the stopping condition is met

[0037]

[0038] 6) Stop the loop; otherwise, go to step 2) and continue the loop.

[0039] 7) Finally, we get multiple decomposed single components u k , each component corresponds to a different frequency component, and the obtained results can effectively capture the non-stationary characteristics in photovoltaic power data.

[0040] The present invention also discloses a photovoltaic power station short-term photovoltaic power generation prediction system, which can be used to implement the above photovoltaic power station short-term photovoltaic power generation prediction method, specifically including:

[0041] Data acquisition module: responsible for collecting historical power data and meteorological data of the photovoltaic system in real time.

[0042] Data preprocessing module: normalize the collected data to eliminate the dimension effect. Use Pearson correlation coefficient analysis to screen meteorological characteristic parameters with high correlation with photovoltaic power generation.

[0043] Data decomposition module: The variational mode decomposition (VMD) method is used to decompose the photovoltaic power time series data into multiple stationary subsequences with different frequencies.

[0044] The whale optimization algorithm WOA is used to optimize the decomposition parameters of VMD and determine the optimal number of decompositions and penalty parameters.

[0045] Feature optimization module: Each subsequence after VMD decomposition is combined with the screened meteorological feature sequence to form a feature input set.

[0046] Prediction model training module: Build a sample convolutional interactive neural network SCINet model, and extract multi-scale features of time series through its internal SCI-Block structure. Perform model training and optimization to enhance the modeling capabilities of complex time series.

[0047] Prediction output module: Use the trained SCINet model to output short-term photovoltaic power prediction values. All feature information is integrated through the Concat&Realign layer to generate the final photovoltaic power prediction result.

[0048] The present invention also discloses a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned photovoltaic power station short-term photovoltaic power generation prediction method when executing the program.

[0049] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned method for predicting short-term photovoltaic power generation power of a photovoltaic power station is implemented.

[0050] Compared with the prior art, the advantages of the present invention are:

[0051] 1. Improve prediction accuracy: The use of variational mode decomposition (VMD) and whale optimization algorithm (WOA) effectively reduces data noise and over-decomposition problems, significantly improving the accuracy of photovoltaic power prediction. By accurately decomposing nonlinear and non-stationary historical data, more representative feature information can be captured.

[0052] 2. Enhanced model adaptability: Combined with the sample convolutional interactive neural network (SCINet) model, the system can adaptively learn features at different time scales, thereby improving the modeling ability of complex photovoltaic power generation time series. The interactive structure of the model can capture multi-level temporal dynamic information and ensure the comprehensive extraction of complex features.

[0053] 3. Real-time and flexibility: The present invention supports real-time data update and dynamic parameter adjustment, so that the prediction of photovoltaic power generation can adapt to the rapidly changing environment and meteorological conditions. Through continuous learning and optimization, the system can remain efficient and flexible in practical applications.

[0054] 4. Optimize power station management: By accurately predicting photovoltaic power generation, power station managers can better dispatch grid operations, optimize power load management, and thus improve the economic benefits and operating efficiency of photovoltaic power stations. Effective prediction can also provide reliable data support for the safe operation of the power grid.

[0055] 5. Reduce the complexity of data processing: Introduce data preprocessing techniques such as Pearson correlation coefficient analysis to ensure the relevance and effectiveness of input features, reduce the complexity of traditional statistical methods in processing nonlinear and non-periodic data, and improve the efficiency of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flow chart of a VMD algorithm according to an embodiment of the present invention;

[0057] Figure 2 It is a structural diagram of SCINet according to an embodiment of the present invention;

[0058] Figure 3 It is an experimental flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

[0060] In the data preprocessing part, the present invention preprocesses the obtained data set, firstly, the original data set is processed by personx coefficient algorithm correlation analysis and its characteristics are analyzed;

[0061] In the data decomposition part, variational mode decomposition (VMD) is used to decompose the original photovoltaic power data with non-stationary characteristics into multiple stationary subsequences of different frequency scales; the whale optimization algorithm (WOA) is introduced into VMD to obtain the optimal decomposition number K and optimal parameter value of the subsequence, and obtain a subsequence with appropriate decomposition degree and no loss of feature information;

[0062] In the prediction part, each subsequence obtained is combined with each feature sequence after feature optimization, and then input into the sample convolutional interactive neural network (SCINet), using multiple convolution filters in SCI-Block, submodules of SCINet, global horizontal radiation (GHI, W / m2), diffuse horizontal radiation (DHI, W / m2), and radiation global tilt (RGI, %). The information is hidden between the three, and the time series information is interactively learned through the interactive learning of the submodule, and then the submodules are combined by SCI-Block using a binary tree. After all the operations of SCI-Block are completed, a scinet is obtained, and finally the photovoltaic power feature information group is merged into the original photovoltaic power time series for prediction of the Concat&Realign layer.

[0063] (a) The correlation coefficient was first designed by statistician Karl Pearson. It is a statistical indicator that studies the linear correlation between variables and is generally represented by the letter r. The Pearson correlation coefficient is used to measure whether two data sets are on the same line and to measure the linear relationship between variables. This embodiment uses the Pearson correlation coefficient to perform feature selection to evaluate the correlation between features and photovoltaic power generation. According to the results of the correlation analysis, features with a higher correlation with photovoltaic power generation are selected as inputs to the prediction model.

[0064] (II) Variational modal decomposition was proposed by Konstantin Dragomiretskiy in 2014. It can effectively suppress the modal aliasing phenomenon of the EMD method (avoiding aliasing by controlling the bandwidth). Different from the EMD principle, the VMD decomposition method uses iterative search for the optimal solution of the variational model to determine the center frequency and bandwidth of each decomposed component. It is a completely non-recursive model. The model searches for a set of modal components and their respective center frequencies, and each mode is smooth after demodulation to the baseband. Experimental results show that this method is more robust in terms of sampling and noise. Modal variational decomposition is an adaptive, completely non-recursive modal variation and signal processing method. This technology has the advantage of being able to determine the number of modal decompositions. Its adaptability is reflected in determining the number of modal decompositions of a given sequence according to actual conditions. In the subsequent search and solution process, the optimal center frequency and limited bandwidth of each mode can be adaptively matched, and the effective separation of the intrinsic modal component (IMF) and the frequency domain division of the signal can be achieved, thereby obtaining the effective decomposition components of the given signal, and finally obtaining the optimal solution to the variational problem.

[0065] like Figure 1 As shown in Figure 2, the specific process of VMD method decomposition is as follows:

[0066] 1) Initialization and n = 0;

[0067] 2) n = n + 1, enter the loop;

[0068] 3) Basis and The update formula is updated until the number of decompositions reaches K and the inner loop is stopped;

[0069] 4) Update λ according to the update formula of λ;

[0070] 5) Given the accuracy ε, if the stopping condition is met

[0071]

[0072] 6) Stop the loop; otherwise, go to step 2) and continue the loop.

[0073] in, is the decomposed single-component AM / FM signal; is the center of each single-component AM / FM signal.

[0074] (III) SCINet model structure Figure 2 As shown in the figure, it is divided into three modules: SC-Block module, SCINet module and stacked SCINet module. It can learn richer multi-scale feature representations and adaptively recalibrate data weights to improve the accuracy of photovoltaic power generation prediction.

[0075] The input feature data Z is downsampled into two subsequences in the SCI-Block module. The two subsequences obtained from the input are divided into odd elements Zeven and even elements Zodd; in order to reduce the information loss of downsampling, interactive learning is introduced between the two subsequences. The interactive learning updates the two subsequences by learning the parameters of the affine transformation from each other, and greatly enhances the representation ability of the learning features by coupling the information of the two complementary parts. Secondly, the SCINet module is a binary tree with SCI blocks as nodes arranged in layers by multiple SCI modules. When the input time series completes all downsampling-convolution-interaction operations, all low-resolution components are realigned and connected into a new sequence representation. It is added to the original time series for prediction through residual connection; finally, the output of SCINet is obtained using a fully connected layer. In order to extract more complex temporal dynamics, it is difficult to grasp the temporal dependency with only one SCINet component. At this point, multiple SCINets are stacked up through relay supervision to form a Stacked SCINet.

[0076] like Figure 3 As shown, the overall experimental process steps are as follows: This embodiment adopts a comprehensive method that combines data preprocessing (person coefficient correlation analysis), VMD decomposition and Yang-Ben convolutional interactive neural network (SCINet) for prediction.

[0077] First, hourly data were collected, including weather variables such as power output from the PV system, humidity, temperature, global horizontal radiation, diffuse horizontal radiation, and wind speed. These data were normalized to ensure that the data had a uniform scale across different variables.

[0078] Secondly, the person coefficient correlation analysis is used to perform feature engineering processing to select parameter time variables with high correlation with photovoltaic power generation.

[0079] Next, VMD is used to decompose the normalized PV power data into multiple intrinsic mode functions (IMFs), which can capture different frequency components and trends. Whale optimization algorithm is added to find the optimal mode K and penalty parameters to prevent over-decomposition, reduce data noise, and obtain the most appropriate PV power generation data information signal component.

[0080] Subsequently, we used the SCINet model to train and predict the decomposed IMFs and clustered weather variable data. The SCINet model is designed based on void random convolution. In each layer, multiple convolution filters are used to extract different but valuable temporal features from subsampled sequences or features. It overcomes the shortcomings of insufficient information extraction in the middle layer of the TCN (Bai et al., 2018) model, greatly reduces the prediction time cost, improves the prediction effect of the model, and has excellent prediction performance. It can effectively learn complex temporal dependencies and patterns in historical data. The data set is divided into 7:1.5:1.5. During training and validation, the model learns how to predict future photovoltaic power generation through historical weather conditions and previous photovoltaic power data.

[0081] The model performance evaluation is mainly completed by comparing the error indicators (such as root mean square error RMSE and mean absolute error MAE) between the predicted output and the actual photovoltaic power generation. We iteratively adjust the model structure and input features to optimize the prediction accuracy and ensure the robustness of the model under different scenarios.

[0082] The method integrates data decomposition, cluster analysis and deep learning models to provide a comprehensive and effective solution for photovoltaic power generation forecasting.

[0083] In another embodiment of the present invention, a short-term photovoltaic power generation prediction system for a photovoltaic power station is provided. The system can be used to implement the above-mentioned short-term photovoltaic power generation prediction method for a photovoltaic power station, specifically including:

[0084] Data acquisition module: responsible for collecting historical power data and meteorological data of the photovoltaic system in real time.

[0085] Data preprocessing module: normalize the collected data to eliminate the dimension effect. Use Pearson correlation coefficient analysis to screen meteorological characteristic parameters with high correlation with photovoltaic power generation.

[0086] Data decomposition module: The variational mode decomposition (VMD) method is used to decompose the photovoltaic power time series data into multiple stationary subsequences with different frequencies.

[0087] The whale optimization algorithm WOA is used to optimize the decomposition parameters of VMD and determine the optimal number of decompositions and penalty parameters.

[0088] Feature optimization module: Each subsequence after VMD decomposition is combined with the screened meteorological feature sequence to form a feature input set.

[0089] Prediction model training module: Build a sample convolutional interactive neural network SCINet model, and extract multi-scale features of time series through its internal SCI-Block structure. Perform model training and optimization to enhance the modeling capabilities of complex time series.

[0090] Prediction output module: Use the trained SCINet model to output short-term photovoltaic power prediction values. All feature information is integrated through the Concat&Realign layer to generate the final photovoltaic power prediction result.

[0091] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of a photovoltaic power station short-term photovoltaic power generation power prediction method, including the following steps:

[0092] S1 Data preprocessing: normalize the historical power data and meteorological data of the photovoltaic system, and select meteorological characteristic parameters with high correlation with photovoltaic power generation as the input of the model through Pearson correlation coefficient analysis;

[0093] S2 data decomposition: Use VMD method to decompose non-stationary photovoltaic power time series data into multiple stationary subsequences with different frequencies; use whale optimization algorithm WOA to optimize VMD decomposition parameters, dynamically determine the optimal decomposition number K and corresponding penalty parameters, and ensure that each subsequence has the best feature expression ability;

[0094] S3 feature optimization and prediction model training: Each subsequence after VMD decomposition is combined with the meteorological feature sequence after Pearson correlation screening and input into the SCINet model. SCINet extracts multi-scale features in the time series through its internal interactive convolution structure SCI-Block, and performs residual connection and optimization on the data through the binary tree structure to enhance the modeling ability of complex time series;

[0095] S4 prediction output: Through the multi-level convolution operation of the SCINet model, the short-term prediction value of photovoltaic power is output, and the Concat&Realign layer is used to integrate all feature information and generate the final photovoltaic power prediction result.

[0096] In another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0097] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the method for predicting short-term photovoltaic power generation power of a photovoltaic power station in the above embodiment; the processor may load and execute the following steps:

[0098] S1 Data preprocessing: normalize the historical power data and meteorological data of the photovoltaic system, and select meteorological characteristic parameters with high correlation with photovoltaic power generation as the input of the model through Pearson correlation coefficient analysis;

[0099] S2 data decomposition: Use VMD method to decompose non-stationary photovoltaic power time series data into multiple stationary subsequences with different frequencies; use whale optimization algorithm WOA to optimize VMD decomposition parameters, dynamically determine the optimal decomposition number K and corresponding penalty parameters, and ensure that each subsequence has the best feature expression ability;

[0100] S3 feature optimization and prediction model training: Each subsequence after VMD decomposition is combined with the meteorological feature sequence after Pearson correlation screening and input into the SCINet model. SCINet extracts multi-scale features in the time series through its internal interactive convolution structure SCI-Block, and performs residual connection and optimization on the data through the binary tree structure to enhance the modeling ability of complex time series;

[0101] S4 prediction output: Through the multi-level convolution operation of the SCINet model, the short-term prediction value of photovoltaic power is output, and the Concat&Realign layer is used to integrate all feature information and generate the final photovoltaic power prediction result.

[0102] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and should be understood that the protection scope of the present invention is not limited to such special statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for predicting short-term photovoltaic power generation in a photovoltaic power station, characterized in that: Based on the deep learning framework Pytorch, combined with Pearson correlation coefficient analysis, variational mode decomposition VMD, whale optimization algorithm WOA and sample convolutional interactive neural network SCINet, short-term prediction of photovoltaic power generation is performed. The specific steps are as follows: S1 Data preprocessing: normalize the historical power data and meteorological data of the photovoltaic system, and select meteorological characteristic parameters with high correlation with photovoltaic power generation as the input of the model through Pearson correlation coefficient analysis; S2 data decomposition: Use VMD method to decompose non-stationary photovoltaic power time series data into multiple stationary subsequences with different frequencies; use whale optimization algorithm WOA to optimize VMD decomposition parameters, dynamically determine the optimal decomposition number K and corresponding penalty parameters, and ensure that each subsequence has the best feature expression ability; S3 feature optimization and prediction model training: Each subsequence after VMD decomposition is combined with the meteorological feature sequence after Pearson correlation screening and input into the SCINet model; SCINet extracts multi-scale features in the time series through its internal interactive convolution structure SCI-Block, and performs residual connection and optimization on the data through the binary tree structure to enhance the modeling ability of complex time series; S4 prediction output: Through the multi-level convolution operation of the SCINet model, the short-term prediction value of photovoltaic power is output, and the Concat&Realign layer is used to integrate all feature information and generate the final photovoltaic power prediction result.

2. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The historical power data and meteorological data include: global horizontal radiation, diffuse horizontal radiation, temperature, humidity and wind speed.

3. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The variational mode decomposition (VMD) obtains the center frequency and bandwidth of the modal component through an iterative optimization method to achieve a smooth decomposition of photovoltaic power data.

4. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The whale optimization algorithm WOA is used to dynamically optimize the number of subsequences K and penalty parameters in VMD decomposition to prevent over-decomposition and ensure that the subsequences retain sufficient feature information.

5. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The sample convolutional interactive neural network SCINet extracts multi-scale features in time series at different levels through convolution filters, and combines different SCI-Block modules through a binary tree structure for residual connection to generate the final prediction result.

6. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The prediction results are evaluated by the indicators of root mean square error (RMSE) and mean absolute error (MAE), and the model parameters are adjusted according to the error feedback to optimize the prediction accuracy.

7. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The structure of the SCINet model includes: an SC-Block module, a SCINet module, and a stacked SCINet module; The SC-Block module includes: input feature data Z represents the pre-processed photovoltaic power generation related features; in the SC-Block module, the input data is downsampled and divided into two sub-sequences: odd sequence Z odd and even sequence Z even ; In order to reduce the information loss in the downsampling process, an interactive learning mechanism is introduced; this mechanism adjusts the affine transformation parameters between two subsequences through mutual learning; Through interactive learning, two subsequences Z odd and Z even Able to couple each other's information and significantly enhance the representation ability of learning features; The SCINet module structure: The SCINet module is composed of multiple SC-Block modules arranged in layers to form a binary tree structure. Each SC-Block extracts features through convolution operations and performs feature learning at low resolution. After the input time series has completed all downsampling, convolution and interaction operations, all low-resolution components are realigned and concatenated into a new sequence representation; Residual connection: The reconnected low-resolution component is added to the original time series and predicted through residual connection; this connection method helps to retain the key information of the original data while incorporating the newly extracted features; Output layer: Finally, the fully connected layer is used to integrate the processed features and output the prediction results of photovoltaic power generation; The stacked SCINet module includes: Through the relay supervision mechanism, multiple SCINet modules are stacked together to form a Stacked SCINet; each stacked SCINet module can independently learn features at different levels and enhance the representation ability of features through information transmission between layers; The Concat&Realign layer realigns and connects into a new sequence; the specific process is as follows: Concat: connects two subsequences Zeven and Zodd in a specific dimension to form a complete new sequence; Realign: To compensate for the time series details that may be lost during the downsampling process, Realign performs data alignment operations based on the characteristics of the subsequences to ensure that the features of different time steps can be effectively combined, providing more accurate input for subsequent prediction layers.

8. The method for predicting short-term photovoltaic power generation in a photovoltaic power station according to claim 1, characterized in that: The specific process of decomposition by the VMD method is as follows: 1) Initialization and n = 0; represents the initial signal of the kth component; represents the initial value of the center frequency of each component; represents the initial value of the penalty parameter, which is used to control the bandwidth during the decomposition process; 2) n = n + 1, enter the loop; + 3) Basis and The update formula is updated until the number of decompositions reaches K and the inner loop is stopped; 4) Update λ according to the update formula of λ; 5) Given the accuracy ε, if the stopping condition is met 6) Stop the cycle; otherwise, go to step 2) and continue the cycle; 7) Finally, we get multiple decomposed single components u k , each component corresponds to a different frequency component, and the obtained results can effectively capture the non-stationary characteristics in photovoltaic power data.

9. A short-term photovoltaic power generation prediction system for a photovoltaic power station, characterized in that: The system can be used to implement the method for predicting short-term photovoltaic power generation in a photovoltaic power station according to any one of claims 1 to 8, specifically comprising: Data acquisition module: responsible for collecting historical power data and meteorological data of the photovoltaic system in real time; Data preprocessing module: normalize the collected data to eliminate the dimension effect; use Pearson correlation coefficient analysis to screen meteorological characteristic parameters with high correlation with photovoltaic power generation; Data decomposition module: The variational mode decomposition (VMD) method is used to decompose the photovoltaic power time series data into multiple stationary subsequences with different frequencies; Use the whale optimization algorithm WOA to optimize the decomposition parameters of VMD and determine the optimal number of decompositions and penalty parameters; Feature optimization module: Combine each subsequence after VMD decomposition with the screened meteorological feature sequence to form a feature input set; Prediction model training module: build a sample convolutional interactive neural network SCINet model, extract the multi-scale features of the time series through its internal SCI-Block structure; perform model training and optimization to enhance the modeling ability of complex time series; Prediction output module: Use the trained SCINet model to output the short-term photovoltaic power prediction value; integrate all feature information through the Concat&Realign layer to generate the final photovoltaic power prediction result.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the method for predicting short-term photovoltaic power generation power of a photovoltaic power station as claimed in any one of claims 1 to 8 is implemented.

Images