A Photovoltaic Power Generation Prediction Method and System Based on Stationarity Correction

Abstract

The present invention discloses a photovoltaic power generation prediction method and system based on stationarity correction. The preprocessed photovoltaic power generation data is segmented according to the cycle length to obtain a photovoltaic power generation data block sequence composed of multiple data blocks. The photovoltaic power generation data block sequence is input into a deep feature extraction model to perform deep time feature extraction and deep space feature extraction respectively, obtaining time features and space features. The time features and space features are fused to obtain deep features. An autocorrelation matrix is introduced to perform stationarity correction on the deep features to obtain corrected features. The corrected features are input into a power generation prediction model for prediction to obtain a prediction result. The present invention designs a deep feature extraction model to extract deep features and corrects the extracted features, improving the prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and particularly to a photovoltaic power generation prediction method and system based on stationarity correction. Background Technique

[0002] The statements in this part only provide background technical information related to the present disclosure, and do not necessarily constitute prior art.

[0003] As the core carrier of clean energy, accurate prediction of photovoltaic power generation has become the key technical basis for power system optimal dispatching, energy storage capacity configuration, and market-based transactions. To ensure the efficient operation and stable power generation of photovoltaic power generation systems, accurate prediction of photovoltaic power generation has become an urgent problem to be solved.

[0004] The prediction of photovoltaic power generation is affected by various factors, including not only long-term time characteristics such as monthly and annual cycles in historical data, but also short-term local time patterns such as daily cycles. At the same time, external environmental factors such as sunlight intensity, ambient temperature, and wind speed also have an important impact on power generation. There are some obvious deficiencies in existing photovoltaic power generation prediction algorithms:

[0005] (1) Many common photovoltaic power generation prediction methods rely on shallow feature extraction models, failing to fully exploit deep temporal features and complex relationships between variables, resulting in low prediction accuracy.

[0006] (2) Traditional methods have not effectively addressed non-stationarity problems caused by factors such as trends, seasonal variations, or irregular fluctuations, and have ignored the time dependence within the data, limiting the accurate modeling and prediction of photovoltaic power generation characteristics.

[0007] (3) Existing temporal modeling methods usually process the entire historical time series as a whole, failing to effectively capture local time patterns at different levels, resulting in insufficient flexibility and accuracy in prediction. Summary of the Invention

[0008] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a photovoltaic power generation prediction method and system based on stationarity correction, designs a deep feature extraction model to extract deep features, and corrects the extracted features, improving the prediction accuracy.

[0009] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

[0010] In a first aspect, the present invention provides a photovoltaic power generation prediction method based on stationarity correction, including:

[0011] Obtain photovoltaic power generation data and perform preprocessing;

[0012] Chunk the preprocessed photovoltaic power generation data according to the cycle length to obtain a sequence of photovoltaic power generation data chunks composed of multiple data chunks;

[0013] Input the sequence of photovoltaic power generation data chunks into a deep feature extraction model to perform deep time feature extraction and deep spatial feature extraction respectively, obtain time features and spatial features, and fuse the time features and spatial features to obtain deep features;

[0014] Introduce an autocorrelation matrix to correct the stationarity of the deep features to obtain corrected features;

[0015] Input the corrected features into a power generation prediction model for prediction to obtain a prediction result.

[0016] A further technical solution is that chunking the preprocessed photovoltaic power generation data according to the cycle length is specifically: repeating the last data value of the photovoltaic power generation data multiple times and filling it at the end of the data, using the cycle length as the chunk length for division to obtain multiple data chunks.

[0017] A further technical solution is that the deep feature extraction model is composed of multiple perceptrons connected in sequence.

[0018] A further technical solution is that obtaining the time features is specifically: first perform local feature extraction within each data chunk, and then perform global feature extraction between different data chunks to obtain time features.

[0019] A further technical solution is that obtaining the spatial features is specifically: combining the meteorological data chunk sequence with the photovoltaic power generation data chunk sequence into a multivariate time series, inputting the multivariate time series into the deep feature extraction model for modeling, and performing feature extraction in the variable dimension to obtain spatial features.

[0020] A further technical solution is that obtaining the corrected features is specifically:

[0021] Calculate the mean, variance, and autocorrelation matrix of the photovoltaic power generation data chunk sequence respectively;

[0022] Calculate the stationarity correction coefficient to make the autocorrelation matrix after multiplying the stationarity correction coefficient by the deep features approximate the autocorrelation matrix of the photovoltaic power generation data chunk sequence, and use the fast Fourier transform to accelerate the calculation of the stationarity correction coefficient;

[0023] Use the stationarity correction coefficient to perform an affine transformation on the deep features to obtain corrected features.

[0024] A further technical solution is that the expression of the stationarity correction coefficient is:

[0025]

[0026]

[0027] Among them, represents the stationarity correction coefficient, represents the length of the input data, represents the photovoltaic power generation data block sequence The th lag sequence and the th lag sequence, the correlation coefficient between them, represents the deep feature The th lag sequence and the th lag sequence, the correlation coefficient between them.

[0028] In a second aspect, the present invention provides a photovoltaic power generation prediction system based on stationarity correction, including:

[0029] A data acquisition module, which is configured to: acquire photovoltaic power generation data and perform preprocessing;

[0030] A data block division module, which is configured to: divide the preprocessed photovoltaic power generation data according to the cycle length to obtain a photovoltaic power generation data block sequence composed of multiple data blocks;

[0031] A deep feature extraction module, which is configured to: input the photovoltaic power generation data block sequence into a deep feature extraction model, respectively perform deep time feature extraction and deep space feature extraction to obtain time features and space features, and fuse the time features and space features to obtain deep features;

[0032] A stationarity correction module, which is configured to: introduce an autocorrelation matrix to perform stationarity correction on the deep features to obtain corrected features;

[0033] A prediction output module, which is configured to: input the corrected features into a power generation prediction model for prediction to obtain a prediction result.

[0034] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in a photovoltaic power generation prediction method based on stationarity correction as described in the first aspect.

[0035] In a fourth aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in a photovoltaic power generation prediction method based on stationarity correction as described in the first aspect.

[0036] The above one or more technical solutions have the following beneficial effects:

[0037] Through sequence chunking, the present invention divides the photovoltaic power generation data by cycle, which can effectively capture local time features and improve the adaptability of the power generation prediction model to periodic changes; the deep time feature extractor is used to fully mine the local time pattern and global time pattern of the photovoltaic power generation data, and the spatial feature extractor is used to fully mine the complex relationship between the photovoltaic power generation data and the meteorological data, enhancing the feature learning ability of the power generation prediction model; the autocorrelation matrix is introduced to correct the stationarity of the extracted features, and by restoring the lost non-stationary information, the dependence of the data and the stability of the model are further ensured; finally, the LSTM long short-term memory neural network is used for the final prediction, significantly improving the accuracy and robustness of the photovoltaic power generation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings forming a part of this invention are used to provide a further understanding of the invention. The schematic embodiments and descriptions thereof of the invention are used to explain the invention and do not constitute an improper limitation of the invention.

[0039] Figure 1 is a sequence chunking flowchart of the photovoltaic power generation prediction method according to an embodiment of the present invention;

[0040] Figure 2 is a deep feature extraction model structure diagram according to an embodiment of the present invention;

[0041] Figure 3 is a time feature extraction flowchart according to an embodiment of the present invention;

[0042] Figure 4 is a spatial feature extraction flowchart according to an embodiment of the present invention;

[0043] Figure 5 is a deep feature stationarity correction flowchart according to an embodiment of the present invention;

[0044] Figure 6 is a module diagram of the photovoltaic power generation prediction system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0046] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0047] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0048] Embodiment 1

[0049] This embodiment discloses a photovoltaic power generation prediction method based on stationarity correction, and the method includes the following steps:

[0050] S1: Obtain photovoltaic power generation data and perform preprocessing;

[0051] In this embodiment, the photovoltaic power generation data is obtained and defined as , represents the length of the data. To enhance the stability and performance of the prediction method, first, preprocess the photovoltaic power generation data, that is, perform a normalization operation to adjust the data to a more appropriate range. The following formula is used for normalization:

[0052] (1)

[0053] Wherein, represents the mean value of the data , represents the variance of the data .

[0054] S2: Divide the preprocessed photovoltaic power generation data into blocks according to the cycle length to obtain a photovoltaic power generation data block sequence composed of multiple data blocks;

[0055] In this embodiment, as Figure 1 shown, the steps of data sequence block division are specifically as follows:

[0056] Repeat the last data value of the photovoltaic power generation data multiple times and fill it at the end of the data;

[0057] Divide the cycle length as the block length to obtain multiple data blocks;

[0058] Based on multiple data blocks, obtain the statistics of each data block.

[0059] Specifically, after normalization, repeat the last data times, and fill it to the end of the original data sequence before chunking.

[0060] To capture the local features of short cycles, the chunk length is defined as the length of the daily cycle. That is, assuming the time step between two data points is 1 hour, the chunk length is 24.

[0061] The step size between two consecutive data chunks is defined as , and the number of chunks is obtained according to the following formula :

[0062] (2)

[0063] where represents the floor function, represents the length of the input data.

[0064] Therefore, the result after chunking is obtained, that is, multiple data chunks form a data chunk sequence , which is called the photovoltaic power generation data chunk sequence. Since the subsequent operations will cause changes in the data distribution within the data chunks, it is also necessary to record the mean value of , the variance and the autocorrelation matrix (first calculate the covariance based on the data, and then obtain the autocorrelation matrix based on the covariance) for subsequent stationarity correction operations.

[0065] Chunk the preprocessed photovoltaic power generation data to better capture the local time features in the photovoltaic power generation data.

[0066] S3: Input the photovoltaic power generation data chunk sequence into the deep feature extraction model, perform deep time feature extraction and deep space feature extraction respectively to obtain time features and space features, and fuse the time features and space features to obtain deep features;

[0067] In this embodiment, as Figure 3 shown, after the data is chunked according to the cycle, the deep feature extraction model is used to extract the time features and space features of the data respectively. The model structure consists of multiple layers of MLP (Multi-Layer Perceptron) connected in sequence. In this embodiment, as Figure 2 shown, the deep feature extraction model uses 3 layers of MLP (a total of 6 MLPs), which can be flexibly set according to the actual application situation.

[0068] The deep time feature extraction is specifically as follows: First, perform local feature extraction within each data chunk, and then perform global feature extraction between different data chunks to obtain time features.

[0069] Specifically, in the process of deep time feature extraction, local feature extraction is first performed within each data block, and then global feature extraction is performed between different data blocks. Specifically, the data of block data are respectively input into the deep feature extraction model to model the features within each daily cycle, that is, to extract short-term time features from the data values within each data block. The short-term time features are defined as local features. Then, the local features are transposed and input into the deep feature extraction model to model the global features between the blocks, which can capture long-term time features such as monthly cycles and annual cycles. The long-term time features are defined as global features. Finally, the data after modeling the global features is transposed to the original dimension to obtain .

[0070] Specifically, deep space feature extraction is as follows: The meteorological data block sequence and the photovoltaic power generation data block sequence are combined into a multivariate time series, and the multivariate time series is input into the deep feature extraction model for modeling to extract features in the variable dimension, obtaining space features.

[0071] As Figure 4 shown, specifically, for deep space feature extraction, since the prediction of photovoltaic power generation is strongly correlated with meteorological variables such as light intensity, the meteorological data (meteorological data block sequence) and the photovoltaic power generation data (photovoltaic power generation data block sequence) are first combined into multivariate time series data, and then the multivariate time series data is input into the deep feature extraction model for modeling to extract space features. The deep feature extraction model extracts features from the multivariate time series data in the variable dimension to obtain space features. The variable dimension here refers to multiple time series, such as photovoltaic power generation data, light intensity data, temperature data, rainfall data, etc., which are all independent variables (time series). Extracting features in the variable dimension means modeling the associations existing between these variables, and the space features refer to the features after modeling the associations between the photovoltaic power generation data and other meteorological variables.

[0072] Specifically, define the meteorological data , where represents the number of meteorological variables, including light intensity (Light), rainfall (Rain), temperature (Temp), etc., as shown in the following formula:

[0073] . (3)

[0074] The meteorological data is divided into blocks, and the result after division is a data block sequence composed of multiple data blocks , called the meteorological data block sequence. The and Combine them to obtain multivariate time series data , and then transpose to obtain . After the data combination is completed, input into the deep feature extraction model, model the variable dimension of size , and extract the data in the dimension where the photovoltaic power generation is located to complete the process of meteorological data feature fusion.

[0075] Finally, perform weighted fusion on the time feature and the space feature to obtain the final deep feature extraction result , that is, the deep feature.

[0076] S4: Introduce an autocorrelation matrix to correct the stationarity of the deep feature to obtain a corrected feature;

[0077] Since the feature extraction process after partitioning will cause a significant change in the data distribution, the existing methods for restoring the data distribution mainly focus on statistical metrics such as the mean and variance, and do not consider the time dependence within the data. Therefore, the basic features such as trends and seasonality in the original data may be affected. For this reason, the present invention introduces a novel stationarity correction method, which constrains the relationship between the stationarity of the time series before and after the model processing to correct the data distribution while retaining the data dependence.

[0078] The stationarity of a time series is mainly reflected in two aspects: the mean and the covariance. The mean is mainly responsible for constraining the distribution from a statistical perspective, while the covariance constrains the distribution from the perspective of time dependence. The change in the mean of the data distribution is usually achieved through global addition or subtraction operations and does not affect the covariance. However, adjusting the covariance may cause a change in the mean of the data distribution. Therefore, the present invention first adjusts the covariance of the data. The covariance can represent the dependence relationship between the photovoltaic power generation data block sequence and its th lag sequence, but it is not sufficient to fully describe the time dependence of the entire time series. Therefore, as Figure 5 shown, the present invention introduces an autocorrelation matrix to provide a more comprehensive constraint on the time series dependence, and defines as the autocorrelation matrix of , which is expressed as:

[0079] (4)

[0080] (5)

[0081] Among them, represents the th lag sequence and the The correlation coefficient between the lag sequences denotes the th lag sequence of denotes the variance of denotes the th lag sequence of denotes the variance of

[0082] Calculate the matrix (i.e., the stationarity correction coefficient), and achieve the autocorrelation matrix after multiplying it with the deep features after feature extraction to approximate , which is expressed as:

[0083] (6)

[0084] (7)

[0085] where denotes the length of the input data, denotes the th correlation coefficient between the th lag sequence and the

[0086] and The constraint between them is expressed as:

[0087] (8)

[0088] where denotes the constraint between and , denotes the square of the Frobenius norm, which is used to measure the size of the matrix, and the calculation formula is the square root of the sum of the squares of all elements in the matrix.

[0089] is a non-stationary time series. Multiplying by will cause the change between the data points in to be scaled by the coefficient . Therefore is equivalent to the following formula:

[0090] . (9)

[0091] According to the Wiener-Khinchin theorem, the calculation can be accelerated by the fast Fourier transform (FFT), and the calculation process is shown in the following formula: The calculation process is as follows:

[0092] (10)

[0093] (11)

[0094] (12)

[0095] Among them, represents the form after removing the mean, that is, subtracting the mean from all values of to make the mean of be 0 to meet the conditions of the Wiener-Khinchin theorem; represents the form after removing the mean; represents the complex conjugate after performing the Fourier transform on represents the complex conjugate after performing the Fourier transform on is the mean of is the mean of

[0096] Therefore, the deep feature can be updated as:

[0097] (13)

[0098] (14)

[0099] Among them, represents the corrected feature, represents the mean vector of represents the mean vector of represents the mean correction term, which is used to adjust the mean difference between and

[0100] The present invention restores the non-stationary information that may be lost during the normalization process, corrects the data distribution, and at the same time maintains the dependence relationship between data by introducing the autocorrelation matrix.

[0101] S5: Input the corrected feature into the power generation prediction model for prediction to obtain the prediction result.

[0102] ​Finally, the results of the completion time feature modeling, spatial feature modeling, and stationarity correction are flattened into a continuous sequence and input into a long short-term memory neural network (LSTM) to obtain the prediction results. Flattening the data into a continuous sequence is to merge the segmented sequences into a one-dimensional continuous sequence, which is convenient for inputting into the LSTM model to output the prediction results.

[0103] The modeling and output process of the LSTM is shown in Equations (15)-(20):

[0104] (15)

[0105] (16)

[0106] (17)

[0107] (18)

[0108] (19)

[0109] (20)

[0110] where represents the input gate, represents the sigmoid activation function, , , , represent the weight matrices of the input gate, forget gate, candidate memory unit, and output gate respectively, represents the hidden state at the current time, represents the input at the current time, , , , represent the bias terms of the input gate, forget gate, candidate memory unit, and output gate respectively, represents the forget gate, represents the candidate memory unit, represents the output gate, represents the memory unit at the current time.

[0111] The final prediction process is expressed as:

[0112] (21)

[0113] where represents the prediction result.

[0114] Photovoltaic power generation prediction, as a time series prediction task, is a typical regression problem. Therefore, several evaluation metrics for common regression problems are used in this invention to better demonstrate the effectiveness of the proposed method. These metrics include Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). Equations (23)-(26) show the calculation formulas for these evaluation metrics. The smaller the values of these metrics, the better the prediction effect.

[0115] (22)

[0116] (23)

[0117] (24)

[0118] (25)

[0119] Among them, represents the true value of photovoltaic power generation at time , represents the predicted value of photovoltaic power generation by the method of this invention at time , is the prediction time length.

[0120] The method of this invention was compared with four advanced methods: the Transformer model, the Bidirectional Long Short-Term Memory Neural Network (Bi-LSTM), the CNN-LSTM Neural Network, and the Temporal Convolutional Network (TCN). The values of the four evaluation metrics of the method of this invention are the smallest, indicating the effectiveness of the proposed method.

[0121] Example Two

[0122] As Figure 6 shown, this embodiment discloses a photovoltaic power generation prediction system based on stationarity correction, including:

[0123] A data acquisition module, which is configured to: acquire photovoltaic power generation data and perform preprocessing;

[0124] A sequence chunking module, which is configured to: chunk the preprocessed photovoltaic power generation data according to the period length to obtain multiple data chunks;

[0125] Deep feature extraction module, which is configured to: input the multiple data blocks into a deep feature extraction model, perform deep time feature extraction and deep space feature extraction respectively to obtain time features and space features, and fuse the time features and space features to obtain deep features;

[0126] Stationarity correction module, which is configured to: introduce an autocorrelation matrix to perform stationarity correction on the deep features to obtain corrected features;

[0127] Prediction output module, which is configured to: input the corrected features into a power generation prediction model for prediction to obtain a prediction result.

[0128] Embodiment III

[0129] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method in Embodiment I are implemented.

[0130] Embodiment IV

[0131] The purpose of this embodiment is to provide a computer-readable storage medium. A computer-readable storage medium has a computer program stored thereon, and when the program is executed by a processor, the steps of the method in Embodiment I are executed.

[0132] The steps involved in the devices in the above Embodiments III and IV correspond to those in Method Embodiment I. For specific implementation manners, reference may be made to the relevant description part of Embodiment I. The term "computer-readable storage medium" should be understood to include a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0133] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to implement. The present invention is not limited to any specific combination of hardware and software.

[0134] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0135] Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made without creative efforts on the basis of the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A photovoltaic power generation prediction method based on stationarity correction, characterized in that: include: Acquire photovoltaic power generation data and perform preprocessing; The pre-processed photovoltaic power generation data is divided into blocks according to the cycle length to obtain a photovoltaic power generation data block sequence consisting of multiple data blocks; Inputting the photovoltaic power generation data block sequence into a deep feature extraction model, performing deep time feature extraction and deep space feature extraction respectively to obtain time features and space features, and fusing the time features and space features to obtain deep features; An autocorrelation matrix is ​​introduced to perform stationarity correction on the deep features to obtain correction features; obtaining the correction features is specifically as follows: the mean, variance and autocorrelation matrix of the photovoltaic power generation data block sequence are calculated respectively; a stationarity correction coefficient is calculated so that the autocorrelation matrix after the stationarity correction coefficient is multiplied by the deep features is close to the autocorrelation matrix of the photovoltaic power generation data block sequence, and a fast Fourier transform is used to accelerate the calculation of the stationarity correction coefficient; the deep features are affine transformed using the stationarity correction coefficient to obtain the correction features; The stability correction coefficient expression is expressed as: in, represents the stability correction coefficient, Indicates the length of input data. Represents a sequence of photovoltaic power generation data blocks No. The lagged series and The correlation coefficient between the lagged series is Representing deep features No. The lagged series and The correlation coefficient between the lagged series; The correction feature is input into a power generation prediction model to perform prediction and obtain a prediction result.

2. A photovoltaic power generation prediction method based on stationarity correction as claimed in claim 1, characterized in that: The preprocessed photovoltaic power generation data is divided into blocks according to the cycle length. Specifically, the last data value of the photovoltaic power generation data is repeated multiple times and filled to the end of the data, and the cycle length is used as the block length to obtain multiple data blocks.

3. A photovoltaic power generation prediction method based on stationarity correction as claimed in claim 1, characterized in that: The deep feature extraction model is composed of multi-layer perceptrons connected in sequence.

4. A photovoltaic power generation prediction method based on stationarity correction as claimed in claim 1, characterized in that: The specific steps of obtaining the time feature are as follows: firstly, local features are extracted within each data block, and then global features are extracted between different data blocks to obtain the time feature.

5. The photovoltaic power generation prediction method based on stationarity correction according to claim 1, characterized in that: The spatial features are obtained specifically as follows: the meteorological data block sequence and the photovoltaic power generation data block sequence are combined into a multivariate time series, the multivariate time series is input into a deep feature extraction model for modeling, and feature extraction is performed on the variable dimension to obtain the spatial features.

6. A photovoltaic power generation prediction system based on stationarity correction, characterized in that: include: A data acquisition module is configured to: acquire photovoltaic power generation data and perform preprocessing; A data block module is configured to: block the preprocessed photovoltaic power generation data according to the cycle length to obtain a photovoltaic power generation data block sequence consisting of multiple data blocks; A deep feature extraction module is configured to: input the photovoltaic power generation data block sequence into a deep feature extraction model, perform deep time feature extraction and deep space feature extraction respectively to obtain time features and space features, and fuse the time features and space features to obtain deep features; A stationarity correction module is configured to: introduce an autocorrelation matrix to perform stationarity correction on the deep features to obtain correction features; obtaining the correction features is specifically: respectively calculating the mean, variance and autocorrelation matrix of the photovoltaic power generation data block sequence; calculating the stationarity correction coefficient, so that the autocorrelation matrix after multiplying the stationarity correction coefficient and the deep features is close to the autocorrelation matrix of the photovoltaic power generation data block sequence, and using fast Fourier transform to accelerate the calculation of the stationarity correction coefficient; using the stationarity correction coefficient to perform affine transformation on the deep features to obtain the correction features; The stability correction coefficient expression is expressed as: in, represents the stability correction coefficient, Indicates the length of input data. Represents a sequence of photovoltaic power generation data blocks No. The lagged series and The correlation coefficient between the lagged series is Representing deep features No. The lagged series and The correlation coefficient between the lagged series; The prediction output module is configured to: input the correction feature into the power generation prediction model for prediction to obtain a prediction result.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a photovoltaic power generation prediction method based on stationarity correction as described in any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the photovoltaic power generation prediction method based on stationarity correction as described in any one of claims 1 to 5 are implemented.

Images