Photovoltaic ultra-short-term power prediction method, electronic equipment and storage medium

Abstract

The invention discloses a photovoltaic ultra-short-term power prediction method, electronic equipment and a storage medium, and relates to the technical field of photovoltaic power prediction. The method comprises the following steps: constructing photovoltaic power station space mixed data by using remote sensing data, a solar elevation angle and a solar azimuth angle; carrying out one-dimension processing on data of various two-dimensional structures by utilizing linear one-dimension processing and convolution one-dimension processing; position coding is carried out on the one-dimensional data to obtain photovoltaic power station space features, cloud layer space features and cloud layer structure features, the features are input to a double-stage cloud feature dynamic coding module, and cloud layer features interfering photovoltaic power generation are extracted; utilizing a feedforward neural network to convert the cloud layer characteristics of the interference photovoltaic power generation into interference factors; and obtaining an interference factor time sequence at the (t-T)-(t + 1) moment through parallel calculation, forming multivariable time sequence data with the historical power and the timestamp sequence, and sending the multivariable time sequence data into the hybrid neural network to predict the power. According to the method, the problems of excessive dependence on a display formula and insufficient spatial modeling fineness at present can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power prediction, and particularly to a photovoltaic ultra-short-term power prediction method, an electronic device, and a storage medium. Background Art

[0002] Solar photovoltaic power generation occupies an important position in the energy consumption structure. However, affected by various meteorological factors, the photovoltaic power generation will fluctuate violently, which will greatly affect the operation level of the power grid. Therefore, improving the accuracy of photovoltaic power prediction is the premise for ensuring the safe and stable operation of the future power grid.

[0003] In the research on photovoltaic power prediction based on satellite cloud images, capturing the cloud characteristics that interfere with photovoltaic power generation can effectively improve the power prediction accuracy. However, the current methods have the following deficiencies: 1) The non-end-to-end (empirical calculation) method is difficult to solve. Determining the intersection position requires determining the cloud top height at the intersection position, and vice versa. There is a circular dependency between the two, and the solution complexity is high. If empirical values are introduced for solution, biases are likely to occur; 2) The refinement degree of the spatial modeling of the photovoltaic power station is insufficient. In many hilly areas, due to terrain limitations, the distribution of photovoltaic power stations is often irregular and scattered, and the spatial span is large. The occluding cloud information inferred from the fuzzy position of the photovoltaic power station is incomplete; 3) It is difficult to determine the range of interfering clouds. The solar irradiance received by the photovoltaic power station is not only determined by the directly occluding clouds, but also affected by the indirectly occluding clouds around the power station. It is difficult for the preset occluding cloud range to adapt to the dynamic changes of the clouds.

[0004] Therefore, there is an urgent need for a new technical solution to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a photovoltaic ultra-short-term power prediction method, an electronic device, and a storage medium, which can solve the problems that the current methods rely too much on explicit formulas and the lack of refinement in spatial modeling.

[0006] To achieve the above purpose, the present invention provides the following solutions:

[0007] A photovoltaic ultra-short-term power prediction method includes:

[0008] Obtaining target photovoltaic data; the target photovoltaic data includes remote sensing data, solar altitude angle, solar azimuth angle, cloud top height data, and satellite cloud images;

[0009] Constructing a spatial hybrid data of a photovoltaic power station by using the remote sensing data, the solar altitude angle, and the solar azimuth angle;

[0010] Convert the spatial hybrid data and cloud top height data of a photovoltaic power station with a two-dimensional structure into one-dimensional spatial hybrid data and one-dimensional cloud top height data of the photovoltaic power station by linear one-dimensionalization, and convert the satellite cloud image with a two-dimensional structure into satellite cloud Figure 1 one-dimensional data;

[0011] Perform position encoding on the one-dimensionalized spatial hybrid data, cloud top height data, and satellite cloud image data of the photovoltaic power station to obtain the spatial characteristics of the photovoltaic power station, the spatial characteristics of the cloud layer, and the cloud layer structure characteristics;

[0012] Input the spatial characteristics of the photovoltaic power station, the spatial characteristics of the cloud layer, and the cloud layer structure characteristics into the two-stage cloud feature dynamic encoding module to extract the cloud layer characteristics that interfere with photovoltaic power generation;

[0013] Use a feedforward neural network to convert the cloud layer characteristics that interfere with photovoltaic power generation into interference factors;

[0014] Obtain the interference factor time series from t - T to t + 1 through parallel computing, and form multivariate time series data with the historical power and timestamp series; where t is the current time and T is the length of historical data;

[0015] Send the multivariate time series data into a hybrid neural network to predict the photovoltaic power.

[0016] Optionally, the construction of the spatial hybrid data of the photovoltaic power station using remote sensing data, solar altitude angle, and solar azimuth angle includes:

[0017] Determine the spatial distribution map of the photovoltaic power station according to the remote sensing data, map the spatial distribution map into the spatial grid map of the meteorological satellite, and then embed the solar altitude angle and solar azimuth angle into the spatial network map to obtain the spatial hybrid data of the photovoltaic power station.

[0018] Optionally, the steps of the linear one-dimensionalization specifically include:

[0019] First, reconstruct the two-dimensional data into a series of data blocks Then, fold the data blocks through dimension folding into Finally, through the linear layer, map the characteristics of to one-dimensional data where H and W are the height and width of the two-dimensional data respectively, N = HW / B 2 represents the number of data blocks, (B, B) represents the size of the data block, B is the width of the data block, C represents the number of channels, and D represents the number of features.

[0020] Optionally, the steps of the convolutional one-dimensionalization specifically include:

[0021] First, Reconstructed into a series of data blocks Then use a convolutional neural network to process the data blocks to convert the dimension of Then, through dimension folding, the dimension is folded into Finally, through a linear layer, the characteristics are mapped to

[0022] Optionally, the two-stage cloud feature dynamic encoding module is composed of cascaded cross-attention encoders, and the calculation formula of the cross-attention encoder is as follows:

[0023]

[0024] where f MCA represents multi-head cross-attention, f MLP represents a multi-layer perceptron, f Norm represents normalization, F1 and F2 are the two inputs of the cross-attention mechanism respectively, and F1 N-1 and F1 N represent the features of the (N - 1)-th layer and the N-th layer respectively.

[0025] Optionally, the specific processing steps for extracting the cloud features interfering with photovoltaic power generation include:

[0026] Input the photovoltaic power station spatial features and cloud spatial features into the first encoder to extract the spatial coupling features between the spatial positions of the photovoltaic power station and the clouds;

[0027] Input the spatial coupling features and cloud structure features into the second encoder to extract the cloud features interfering with photovoltaic power generation.

[0028] The present invention also provides an electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the photovoltaic ultra-short-term power prediction method according to the above.

[0029] The present invention also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the photovoltaic ultra-short-term power prediction method as described above.

[0030] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0031] The present invention discloses a photovoltaic ultra-short-term power prediction method, an electronic device, and a storage medium. The method includes constructing photovoltaic power station spatial hybrid data by using remote sensing data, solar altitude angle, and solar azimuth angle; converting the two-dimensional photovoltaic power station spatial hybrid data and cloud top height data into one-dimensional photovoltaic power station spatial hybrid data and cloud top height one-dimensional data by using linear one-dimensionalization, and converting the two-dimensional satellite cloud image into satellite cloud one-dimensional data by using convolutional one-dimensionalization. Figure 1 Performing positional encoding on the one-dimensionalized photovoltaic power station spatial hybrid data, cloud top height data, and satellite cloud image data to obtain photovoltaic power station spatial features, cloud layer spatial features, and cloud layer structure features; inputting the photovoltaic power station spatial features, cloud layer spatial features, and cloud layer structure features into a two-stage cloud feature dynamic encoding module to extract cloud layer features that interfere with photovoltaic power generation; converting the cloud layer features that interfere with photovoltaic power generation into interference factors by using a feedforward neural network; obtaining a time series of interference factors at t-T to t+1 moments through parallel computing, and forming multivariate time series data with historical power and timestamp sequences; and sending the multivariate time series data into a hybrid neural network to predict photovoltaic power. The present invention can solve the problems that the current method relies too much on explicit formulas and the spatial modeling fineness is insufficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0033] Figure 1 It is a schematic flow chart of the photovoltaic ultra-short-term power prediction method of the present invention;

[0034] Figure 2 It is a framework diagram of the photovoltaic ultra-short-term power prediction method in this embodiment;

[0035] Figure 3 It is a schematic diagram of the construction method of the photovoltaic power station spatial hybrid data in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0037] The object of the present invention is to provide a photovoltaic ultra-short-term power prediction method, an electronic device and a storage medium, which can solve the problems that the current method relies too much on explicit formulas and the spatial modeling fineness is insufficient.

[0038] To make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0039] As Figure 1 - Figure 2 shown, the present invention provides a photovoltaic ultra-short-term power prediction method, including:

[0040] S1: Construct photovoltaic power station spatial mixed data by using remote sensing data, solar altitude angle and solar azimuth angle; Combine Figure 3 The specific steps of constructing the photovoltaic power station spatial mixed data are described as follows:

[0041] S1.1: Obtain spatial distribution data. Obtain the refined spatial distribution map of the photovoltaic power station through the remote sensing image of the high-resolution remote sensing satellite ( Figure 3 the photovoltaic panel laying area in);

[0042] S1.2: Spatial information mapping. Map the high-resolution spatial distribution map into the spatial grid map of the meteorological satellite ( Figure 3 the grid of the area where the photovoltaic power station is located in);

[0043] S1.3: Construct spatial mixed data. Embed the solar altitude angle and solar azimuth angle into the spatial grid to obtain the photovoltaic power station spatial mixed data ( Figure 3 the right side in);

[0044] The dimension of the constructed photovoltaic power station spatial mixed data is

[0045] Among them, the calculation formulas of the solar altitude angle and solar azimuth angle are:

[0046]

[0047] In the formula: α s , γ s are the solar altitude angle and azimuth angle respectively, ω and δ are the hour angle and declination angle respectively, t s represents the local solar time (24-hour system), d represents the serial number of a certain day in a year, represents the latitude.

[0048] S2: Convert the two-dimensional photovoltaic power station spatial mixed data and cloud top height data into one-dimensional photovoltaic power station spatial mixed data and cloud top height one-dimensional data through linear one-dimensionalization, and convert the two-dimensional satellite cloud image into satellite cloud Figure 1 one-dimensional data through convolutional one-dimensionalization;

[0049] The steps of linear one-dimensionalization include:

[0050] S2.1-a: Reconstruct the two-dimensional data into a series of data blocks

[0051] S2.2-a: Fold the data blocks through dimensional folding into

[0052] S2.3-a: Map the features with dimension to

[0053] The steps of convolutional one-dimensionalization include:

[0054] S2.1-b: Reconstruct the two-dimensional data into a series of data blocks

[0055] S2.2-b: Use a convolutional neural network to convert the dimension of the data blocks to

[0056] This convolutional neural network consists of a convolutional layer, a residual block, and global pooling;

[0057] S2.2-b: Fold the dimension through dimensional folding into

[0058] S2.3-b: Map the features with dimension to

[0059] S3: Perform positional encoding on the one-dimensionalized space hybrid data, cloud top height data, and satellite cloud image data of the photovoltaic power station to obtain the space features of the photovoltaic power station, the cloud space features, and the cloud structure features;

[0060] The calculation formula for positional encoding is:

[0061]

[0062] In the formula: pos is the position index, i is the dimension index, and d is the total dimension of the encoding vector.

[0063] S4: Input the three features into the two-stage cloud feature dynamic encoding module to extract the cloud features that interfere with photovoltaic power generation;

[0064] The two-stage cloud feature dynamic encoding module consists of cascaded cross-attention encoders. Each encoder stacks N groups of encoder layers inside. The definition from the (N - 1)-th layer going forward to the N-th layer is as follows:

[0065]

[0066] In the formula: f MCA represents multi-headed cross-attention (MCA), f MLP represents multi-layer perceptron (MLP), f Norm represents normalization, F1 N-1 and F1 N represent the features of the (N - 1)-th layer and the N-th layer respectively.

[0067] The calculation method of cross-attention is as follows:

[0068]

[0069] Among them:

[0070]

[0071] In the formula: W is the matrix coefficient, d k is the number of columns of matrix K, F A is the output of the attention mechanism.

[0072] The steps to extract cloud layer features are as follows:

[0073] S4.1: Input the spatial features of the photovoltaic power station and the cloud layer spatial features into the first encoder to extract the spatial coupling features between the spatial positions of the photovoltaic power station and the cloud layer;

[0074] S4.2: Input the spatial coupling features in S4.1 and the cloud layer structure features into the second encoder to extract the cloud layer features affecting photovoltaic power generation.

[0075] S5: Use a feedforward neural network to convert the cloud layer features in S5 into interference factors;

[0076] S6: Obtain the interference factor time series from t - T to t + 1 through parallel computing, and form multivariate time series data with the historical power and timestamp series;

[0077] The lengths of the above three types of time series data are all T + 1, and the power value at the t + 1 moment is set to 0 to keep the sequence alignment.

[0078] S7: Feed the multivariate time series data into a hybrid neural network to predict photovoltaic power.

[0079] The hybrid neural network is a combined model of a convolutional neural network and a gated recurrent unit.

[0080] In this embodiment, a very short-term prediction of the power generation of a certain photovoltaic power station in a central province for the next 1 hour is carried out. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ) are used to evaluate the performance of the model:

[0081]

[0082] In the formula: P i and are the actual power and predicted power of the i-th sample respectively, n is the total number of samples, is the sample mean.

[0083] Two groups of comparative analyses are set:

[0084] In the first group, based on the Active Cloud Region Selection (ACRS) rule as the occlusion cloud region benchmark selection method, the required cloud top height is the mean of the cloud top heights around the photovoltaic power station. Since this benchmark method does not require constructing cross-attention, the VIT model is used to encode the local cloud map determined by this method and extract interference factors.

[0085] In the second group of simulations, the backbone networks of different advanced models are used as benchmarks to replace S2 - S4 in the present invention, aiming to prove the effectiveness of the two-stage cloud feature dynamic encoding module. At the same time, as shown in the following formula, X RAA , X CTH and X CI are integrated into X Cat as the input for the benchmark model to process.

[0086]

[0087] In the formula: f cat is the splicing operation.

[0088] Table 1 Overall comparison results

[0089]

[0090] The present invention uses the VIT model to verify the two methods respectively. Except for the prediction 15 minutes in advance, the prediction accuracy based on the ACRS method is slightly lower than that of the end-to-end method based on multi-modal data coupling. This difference mainly stems from the fact that empirical calculations rely on a large number of parameter estimations. Introducing the empirical value of cloud top height is likely to cause a large deviation in the solution result of ACRS. The calculated regional cloud map is not the occlusion area, which affects the prediction accuracy.

[0091] Although the photovoltaic power prediction accuracy can be effectively improved based on spatial hybrid data and cloud top height data, the self-attention mechanism in VIT does not seem suitable for processing multi-modal data. Its prediction accuracy is only superior to the relatively single-function CNN and Resnet-18 as a whole, but inferior to SENet and SKNet with stronger feature extraction capabilities. It cannot fully establish the coupling relationship between the spatial hybrid data, cloud top height data and satellite cloud map of the photovoltaic power station, and extract the cloud features affecting photovoltaic power prediction.

[0092] The dual-stage cloud feature dynamic encoding module proposed in the present invention effectively solves this problem. By means of the cross-attention mechanism, the dependence relationships among the three are established in turn, and the cloud interference variables affecting photovoltaic power generation are adaptively extracted. Compared with the self-attention mechanism in VIT, the MAE and RMSE in multi-step prediction of the method proposed in the present invention are reduced by 13.38% and 15.67% on average, and the R 2 is increased by 5.37% on average.

[0093] To ensure the reliability of the above results, taking the prediction 15 minutes in advance as an example, the present invention shows the correlation coefficient between the interference factors generated by different methods and the actual power. Since the stronger the shielding effect of the cloud layer, the lower the photovoltaic power generation, there is a negative correlation between the interference factor and the actual power. The smaller the value, the higher the degree of correlation. As can be seen from Table 2, the degree of correlation between the interference factors generated by the method of the present invention and the actual power is the highest. This shows that the method of the present invention can process multi-modal data more effectively and mine the dependence relationships between data compared with other advanced models, so as to accurately extract the interference factors affecting photovoltaic power generation. At the same time, this result can be mutually confirmed with the results reported in Table 1, proving the effectiveness of the method proposed in the present invention.

[0094] Table 2 Correlation coefficient between interference factor and actual power

[0095]

[0096] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other.

[0097] In this specification, specific examples are used to illustrate the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A photovoltaic ultra-short-term power prediction method, characterized in that, Including: Obtain target photovoltaic data; the target photovoltaic data includes remote sensing data, solar altitude angle, solar azimuth angle, cloud top height data, and satellite cloud images; Construct spatial hybrid data of a photovoltaic power station using remote sensing data, solar altitude angle, and solar azimuth angle; Use linear one-dimensionalization to convert the two-dimensional spatial hybrid data of the photovoltaic power station and cloud top height data into one-dimensional spatial hybrid data of the photovoltaic power station and one-dimensional cloud top height data, and use convolutional one-dimensionalization to convert the two-dimensional satellite cloud image into one-dimensional satellite cloud image data; Perform positional encoding on the one-dimensionalized spatial hybrid data of the photovoltaic power station, cloud top height data, and satellite cloud image data to obtain spatial characteristics of the photovoltaic power station, cloud layer spatial characteristics, and cloud layer structure characteristics; Input the spatial characteristics of the photovoltaic power station, cloud layer spatial characteristics, and cloud layer structure characteristics into a two-stage cloud feature dynamic encoding module to extract cloud layer features that interfere with photovoltaic power generation; Use a feedforward neural network to convert the cloud layer features that interfere with photovoltaic power generation into interference factors; Obtain a time series of interference factors at times t - T to t + 1 through parallel computing, and form multivariate time series data with historical power and timestamp sequences; where t is the current time and T is the length of historical data; Send the multivariate time series data into a hybrid neural network to predict photovoltaic power.

2. The photovoltaic ultra-short-term power prediction method according to claim 1, wherein The construction of the spatial hybrid data of the photovoltaic power station using remote sensing data, solar altitude angle, and solar azimuth angle includes: Determine the spatial distribution map of the photovoltaic power station according to the remote sensing data, map the spatial distribution map into the spatial grid map of the meteorological satellite, and then embed the solar altitude angle and solar azimuth angle into the spatial network map to obtain the spatial hybrid data of the photovoltaic power station.

3. The photovoltaic ultra-short-term power prediction method according to claim 1, wherein The steps of the linear one-dimensionalization specifically include: First, the two-dimensional data is reconstructed into a series of data blocks Then, the data blocks are folded through dimensional folding into Finally, the features of are mapped to one-dimensional data where H and W are the height and width of the two-dimensional data respectively, N = HW / B 2 represents the number of data blocks, (B, B) represents the size of the data block, B is the width of the data block, C represents the number of channels, and D represents the number of features.

4. The photovoltaic ultra-short-term power prediction method according to claim 1, characterized in that The steps of the convolutional one-dimensionalization specifically include: First, the two-dimensional data is reconstructed into a series of data blocks Then, a convolutional neural network is used to convert the dimension of the data block to Then, the dimension is folded into Finally, the features of are mapped to 5. The photovoltaic ultra-short-term power prediction method according to claim 1, wherein The two-stage cloud feature dynamic encoding module is composed of cascaded cross-attention encoders, and the calculation formula of the cross-attention encoder is: Among them, f MCA represents multi-head cross-attention, f MLP represents a multi-layer perceptron, f Norm represents normalization, F1 and F2 are the two inputs of the cross-attention mechanism, F1 N-1 and F1 N represent the features of the (N - 1)-th layer and the N-th layer respectively.

6. The photovoltaic ultra-short-term power prediction method according to claim 1, wherein The specific processing steps for extracting the cloud layer features that interfere with photovoltaic power generation include: Input the spatial characteristics of the photovoltaic power station and the cloud layer spatial characteristics into the first encoder to extract the spatial coupling characteristics between the spatial position of the photovoltaic power station and the cloud layer position; Input the spatial coupling characteristics and the cloud layer structure characteristics into the second encoder to extract the cloud layer features that interfere with photovoltaic power generation.

7. An electronic device, characterized in that, Including a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the photovoltaic ultra-short-term power prediction method according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by the processor, it implements the photovoltaic ultra-short-term power prediction method according to any one of claims 1-6.

Images