Multi-modal fusion photovoltaic power prediction method for all-sky imager data

Abstract

The invention discloses a multi-mode fusion photovoltaic power prediction method for all-sky imager data. The method comprises the following steps: obtaining time sequence sky image data and photovoltaic array historical photovoltaic power data in an all-sky imager; extracting features of the time sequence sky image data and the historical photovoltaic power data to obtain modal feature information of the time sequence sky image data and modal feature information of the historical photovoltaic power data; fusing the modal feature information of the time sequence sky image data and the modal feature information of the historical photovoltaic power data to obtain a feature matrix; obtaining a final potential feature according to the feature matrix; training by utilizing the final potential features to obtain a photovoltaic power prediction model; and predicting the photovoltaic power by using the photovoltaic power prediction model. According to the method, feature extraction processing is carried out on the time sequence sky image data and the historical photovoltaic power data, and prediction errors caused by intermittency and uncertainty of solar power generation are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power prediction, and more specifically, to a photovoltaic power prediction method using multi-modal fusion of all-sky imager data. Background Art

[0002] The widespread use of solar energy, the rapid development of renewable energy power generation represented by photovoltaics, and especially the popularization of rooftop photovoltaic panels, have become an important solution to address destructive climate change and promote sustainable energy development. As this trend continues to deepen, the penetration rate of solar energy in the power system is expected to increase steadily. China's cumulative photovoltaic installed capacity has increased from 43.18GW in 2015 to 609.49GW in 2023, with an average annual compound growth rate of about 39%. The annual increase in photovoltaic installed capacity and the cumulative photovoltaic installed capacity are both ranked first in the world. Promoting the development of photovoltaic power generation has become one of the important ways to promote the transformation of my country's power system.

[0003] For photovoltaic power prediction, there is a photovoltaic power multi-step prediction method based on multi-view feature extraction and multi-task learning in the prior art. First, in order to obtain rich and comprehensive feature information, LSTM, MLP, and CNN are used to extract the time series features, global features, and local features in the original photovoltaic power data respectively; secondly, the multi-step photovoltaic power prediction task is converted into multiple single-step photovoltaic power prediction subtasks, and a multi-task learning model based on a lightweight attention mechanism and an expert network (LAMoE) is used for multi-step prediction to fully utilize the correlation of multi-step prediction values; finally, an improved dynamic weighted averaging method is used to adaptively optimize and adjust the loss weights to predict photovoltaics. However, due to the limitation of the feature information of the original photovoltaic power data, the consideration of spatial and cloud feature information factors is lacking, and there is a disadvantage of photovoltaic prediction accuracy deviation. Summary of the invention

[0004] The present invention provides a photovoltaic power prediction method based on multimodal fusion of all-sky imager data, which solves the technical problem in the prior art that the photovoltaic prediction accuracy is biased due to the limitation of characteristic information of original photovoltaic power data and the lack of consideration of spatial and cloud characteristic information factors.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows: A photovoltaic power prediction method based on multi-modal fusion of all-sky imager data comprises the following steps: Obtain time-series sky image data from the all-sky imager and historical photovoltaic power data of the photovoltaic array; Extracting features from the time-series sky image data and the historical photovoltaic power data respectively to obtain modal feature information of the time-series sky image data and modal feature information of the historical photovoltaic power data; fusing the modal feature information of the time-series sky image data and the modal feature information of the historical photovoltaic power data to obtain a feature matrix; According to the feature matrix, a final potential feature is obtained; Using the final potential features to train a photovoltaic power prediction model; The photovoltaic power prediction model is used to predict photovoltaic power.

[0006] In the above-mentioned technical means, feature extraction and processing are performed on the time-series sky image data collected by the all-sky imager and the historical photovoltaic power data of the photovoltaic array, which fully considers the temporal, spatial and cloud dynamic characteristic information in different modes, effectively reduces the prediction error caused by the intermittent and uncertain nature of solar power generation, and improves the accuracy of multi-step prediction of photovoltaic power of solar power stations. It solves the technical problem of photovoltaic prediction accuracy deviation in the existing technology due to the limitations of the characteristic information of the original photovoltaic power data and the lack of consideration of spatial and cloud characteristic information factors.

[0007] Further, extracting features from the time series sky image data to obtain modal feature information of the time series sky image data includes: Reduce the dimension of the time series sky image data to obtain a grayscale image, and convert the grayscale image into a grayscale pixel grid image; Define a circle with the center of the grayscale pixel grid as the center and a radius of wherein the circular area does not exceed the range of the grayscale pixel grid image, is an integer and , ; For a radius of The grayscale values ​​of the pixels in the grayscale pixel grid map located in the circular area are averaged to obtain a radius of The local mean corresponding to the circular area when ; When it is greater than 1, for a radius of For the circular area of ​​grayscale pixels in the grid, the distance from the center of the circle is greater than and less than The average grayscale value of the pixels is taken to get the radius The local mean corresponding to the circular area when ; When the circular area touches the edge of the grayscale pixel grid map, the local means are aggregated to obtain the modal feature information of the time series sky image data.

[0008] Further, extracting features from the time series sky image data to obtain modal feature information of the time series sky image data also includes: The median filtering method is used to calculate the median value of a 3 x 3 fixed window for the feature value points located near the sun in the modal feature information of the time series sky image data, and the median value is used to replace the original feature points.

[0009] Furthermore, features are extracted from the historical photovoltaic power data to obtain modal feature information of the historical photovoltaic power data, including: Calculate the mean of the historical photovoltaic power data and standard deviation ; Calculate abnormal data points in the historical photovoltaic power data according to a Z-score method; Interpolating and resampling missing points and abnormal data points in the historical photovoltaic power data to obtain a photovoltaic power series; After the photovoltaic power sequence is processed by BiLSTM, the forward and reverse outputs are spliced ​​to obtain the modal characteristic information of the historical photovoltaic power data.

[0010] Further, calculating abnormal data points in the historical photovoltaic power data according to the Z-score method includes:

[0011] In the formula, represents a photovoltaic power data point. When Z is greater than a preset threshold, the photovoltaic power data point is determined to be an abnormal data point.

[0012] Furthermore, the modal feature information of the time series sky image data and the modal feature information of the historical photovoltaic power data are fused to obtain a feature matrix, including: The modal feature information of the time series sky image data and the modal feature information of the historical photovoltaic power data are aligned with the timestamps and then superimposed to obtain a feature matrix .

[0013] Furthermore, according to the feature matrix, the final potential features are obtained, including: The feature matrix Expanding to obtain a feature vector, and filtering from the feature vector to obtain regularized feature information; The regularized feature information is processed through a secondary feature extraction module and a pooling layer with L1 norm penalty to obtain the final potential feature.

[0014] Furthermore, the regularized feature information is obtained by screening the feature vector, including: A feature selection module based on L1 norm regularization is constructed, and a group of mean-effective features are selected from the high-dimensional input space by the feature selection module for the feature vector to obtain the regularized feature information.

[0015] Furthermore, the secondary feature extraction module is a multi-layer stacked encoder and decoder structure.

[0016] Furthermore, the photovoltaic power prediction model is a linear regression prediction model.

[0017] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: The present invention proposes a multi-modal fusion photovoltaic power prediction method for sky imager data. First, the time series sky image data of the sky imager is obtained, and the sky image is reduced to a grayscale image. A new method is created, in which the center coordinates of the grayscale pixel matrix are taken as the circle point, the radius parameter is R, and R is gradually accumulated to extract the average grayscale pixel value in the circular neighborhood. This process can effectively extract the brightness information, cloud features and other visual features in the image, and provide rich basic data for subsequent analysis. In order to further improve the quality of the data, the median filter method is used to process the image to remove the noise caused by the mirror reflection effect of sunlight, so as to obtain more accurate image features. Secondly, the historical photovoltaic power time series data of the photovoltaic array of the solar panel is collected, and the BiLSTM can be used to simultaneously capture the forward and backward time dependencies in the data and mine its inherent nonlinear characteristics. By aligning the timestamps, the time series features of different modes are fused into a feature matrix, and then the useful feature information is screened by L1 norm regularization regression. Finally, the final potential feature vector matrix output by the secondary feature processing module is input into the linear regression prediction model to improve the prediction accuracy of photovoltaic power based on the sky image. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic flow chart of a photovoltaic power prediction method using multi-modal fusion of all-sky imager data provided by an embodiment of the present invention; Figure 2 A schematic diagram of the principle of a photovoltaic power prediction method using multi-modal fusion of all-sky imager data provided by an embodiment of the present invention; Figure 3 An embodiment of the present invention provides a photovoltaic power prediction effect diagram for a certain day obtained by using a multi-modal fusion photovoltaic power prediction method of all-sky imager data proposed by the present invention. DETAILED DESCRIPTION

[0019] The drawings are for illustrative purposes only and should not be construed as limiting the present patent; In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0020] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0021] Example 1 This embodiment provides a photovoltaic power prediction method based on multi-modal fusion of all-sky imager data, such as Figure 1 As shown, the following steps are included: Obtain time-series sky image data from the all-sky imager and historical photovoltaic power data of the photovoltaic array; Extracting features from the time-series sky image data and the historical photovoltaic power data respectively to obtain modal feature information of the time-series sky image data and modal feature information of the historical photovoltaic power data; fusing the modal feature information of the time-series sky image data and the modal feature information of the historical photovoltaic power data to obtain a feature matrix; According to the feature matrix, a final potential feature is obtained; Using the final potential features to train a photovoltaic power prediction model; The photovoltaic power prediction model is used to predict photovoltaic power.

[0022] like Figure 2 FIG. 1 is a schematic diagram showing the principle of a photovoltaic power prediction method for multi-modal fusion of all-sky imager data provided in this embodiment, which is specifically as follows: The time-series sky image data from the All Sky Imager includes sky image data with a resolution of one minute; The photovoltaic array historical photovoltaic power data includes historical photovoltaic power data with a size of 2048x2048 and a sampling frequency of one minute.

[0023] In a further embodiment, extracting features from the time series sky image data to obtain modal feature information of the time series sky image data includes: Reduce the dimension of the time series sky image data to obtain a grayscale image, and convert the grayscale image into a grayscale pixel grid image; In this embodiment, the grayscale level range of the grayscale pixel grid image is 0-255; by extracting the texture and edge features of the time-series grayscale image according to the dimensionality reduction law of the time-series sky image data, the time-series image can also obtain the dynamic change trend of the cloud layer; Define a circle with the center of the grayscale pixel grid as the center and a radius of wherein the circular area does not exceed the range of the grayscale pixel grid image, is an integer and , ; In this embodiment, the radius is The pixel points within the circular area are determined by the following formula:

[0024] In the formula, are the pixel coordinates within the circular neighborhood, are the coordinates of the circle center; For a radius of The grayscale values ​​of the pixels in the grayscale pixel grid map located in the circular area are averaged to obtain a radius of The local mean corresponding to the circular area when ; When it is greater than 1, for a radius of For the circular area of ​​grayscale pixels in the grid, the distance from the center of the circle is greater than and less than The average grayscale value of the pixels is taken to get the radius The local mean corresponding to the circular area when ; In this embodiment, the circular domain mean formula is:

[0025] In the formula, For point Accumulate the local mean within the kth radius; The center pixel is the center and the radius is The set of all pixels used to calculate the mean in the circular neighborhood of ; is the number of pixels in the neighborhood; When the circular area touches the edge of the grayscale pixel grid map, the local means are aggregated to obtain the modal feature information of the time series sky image data.

[0026] In a further embodiment, extracting features from the time series sky image data to obtain modal feature information of the time series sky image data further includes: The median filtering method is used to calculate the median value of a 3 x 3 fixed window for the feature value points located near the sun in the modal feature information of the time series sky image data, and the median value is used to replace the original feature points.

[0027] In this embodiment, a median filtering method is used to reduce sunlight reflection.

[0028] In a further embodiment, extracting features from the historical photovoltaic power data to obtain modal feature information of the historical photovoltaic power data includes: Due to uncontrollable factors, historical power data often contains missing values ​​and outliers. It is necessary to process the missing values ​​and outliers. First, the mean of the historical photovoltaic power data is calculated. and standard deviation ; Calculate abnormal data points in the historical photovoltaic power data according to a Z-score method; On a straight line, every position between two points can be predicted. The goal of linear interpolation is to find the value of x between and The value of a point (x, y) between , so the missing points and abnormal data points in the historical photovoltaic power data can be interpolated and resampled to obtain the photovoltaic power series; After the photovoltaic power sequence is processed by BiLSTM, the forward and reverse outputs are spliced ​​to obtain the modal characteristic information of the historical photovoltaic power data.

[0029] In a specific embodiment, the BiLSTM is composed of two independent LSTM units, which are responsible for processing the input sequence from two directions (forward and reverse), wherein the main formula of the LSTM unit is:

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] In the formula, , , , , , , and is the weight matrix and bias of the corresponding input of the network activation function; , , , , They are forget gate, input gate, cell state, output gate and hidden state; represents the sigmoid activation function, is the hyperbolic tangent function.

[0036] After the historical photovoltaic power data is processed by BiLSTM, the forward and reverse outputs are spliced ​​to capture the long-term time dependency and nonlinear feature information in the sequence. After the feature information is complementary and redundant information is eliminated, new time series modal features are generated.

[0037] In a further embodiment, calculating abnormal data points in the historical photovoltaic power data according to a Z-score method includes:

[0038] In the formula, represents a photovoltaic power data point. When Z is greater than a preset threshold, the photovoltaic power data point is determined to be an abnormal data point.

[0039] In a specific embodiment, the threshold is set to 3, and if Z is greater than 3, it is identified as a value deviating from the normal range, and then the abnormal value is marked.

[0040] In a further embodiment, the modal feature information of the time series sky image data and the modal feature information of the historical photovoltaic power data are fused to obtain a feature matrix, including: The modal feature information of the time series sky image data and the modal feature information of the historical photovoltaic power data are aligned with the timestamps and then superimposed to obtain a feature matrix .

[0041] In a further embodiment, the final potential features are obtained according to the feature matrix, including: The feature matrix Expanding to obtain a feature vector, and filtering from the feature vector to obtain regularized feature information; The regularized feature information is processed through a secondary feature extraction module and a pooling layer with L1 norm penalty to obtain the final potential feature.

[0042] In a further embodiment, the regularized feature information is obtained by screening the feature vector, including: A feature selection module based on L1 norm regularization is constructed, and a group of mean-effective features are selected from the high-dimensional input space by the feature selection module for the feature vector to obtain the regularized feature information.

[0043] In a further embodiment, the secondary feature extraction module is a multi-layer stacked encoder and decoder structure.

[0044] In this embodiment, the secondary feature extraction module based on the Transformer architecture can capture long-range temporal dependencies through the self-attention mechanism in the Transformer, effectively model the global information of time series image features, and gradually extract abstract features of different time scales and higher levels through a multi-layer stacked encoder and decoder structure.

[0045] In a further embodiment, a pooling operation based on L1 norm penalty is used to reduce the dimension to obtain the final latent features.

[0046] In a further embodiment, it is characterized in that the photovoltaic power prediction model is a linear regression prediction model.

[0047] Example 2 In order to verify the effectiveness of the multi-modal fusion photovoltaic power prediction method of the all-sky imager data proposed in the present invention, the sky image and historical photovoltaic power generation data of a certain place from 6:00 to 20:00 in one month are first obtained and the two different modal data are preprocessed. Then, the feature matrix is ​​selected according to the feature fusion of the preprocessed data. Finally, the final potential features are output by the secondary feature processing module based on the transformer architecture, and the linear regression model is input to predict the data of a certain day. Figure 3 The predicted effect diagram of photovoltaic power generation is shown in Figure 3 The solid trend line in the center of the circle represents the actual value of photovoltaic power generation, and the solid trend line in the center of the triangle represents the predicted value of photovoltaic power generation. Figure 3 It can be seen from the curves corresponding to the two values ​​shown in that the error between the actual value of photovoltaic power generation and the predicted value of photovoltaic power generation is very small. The method proposed in the present invention can effectively improve the accuracy of photovoltaic power generation prediction.

[0048] Example 3 This embodiment proposes a photovoltaic power prediction system for multi-modal fusion of all-sky imager data, the system comprising: A sky image and historical photovoltaic power data acquisition and processing unit is used to acquire the image mode of the all-sky imager and the historical photovoltaic power of the photovoltaic array, and to perform preliminary processing on the relevant data; The time series feature processing unit is used to extract features and smooth the sky image data, and filter out the feature information of brightness, cloud layer and vision through the feature module; the time series photovoltaic power data is used in the BiLSTM network to obtain the time series feature information.

[0049] The modality feature fusion unit aligns the timestamps of the image modality and the time series modality to form a new feature, flattens it into a long vector, and inputs it into the L1 norm regularized regression feature selection module to filter out useful feature information; The model building unit is used to build a secondary feature processing model and a linear regression prediction model. The multimodal fusion data is processed by the secondary feature processing model to extract abstract features of different time scales and higher levels. Finally, the input features are pooled through the L1 norm penalty pooling layer to form the potential features of the final input prediction model. The prediction unit is used to input the final potential features obtained by the deep learning network feature processing model into the final linear regression prediction model to predict the photovoltaic output power of the power station.

[0050] The same or similar reference numerals correspond to the same or similar components; The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent; Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A photovoltaic power prediction method based on multi-modal fusion of all-sky imager data, characterized in that: The following steps are involved: Obtain time-series sky image data from the all-sky imager and historical photovoltaic power data of the photovoltaic array; Extracting features from the time-series sky image data and the historical photovoltaic power data respectively to obtain modal feature information of the time-series sky image data and modal feature information of the historical photovoltaic power data; fusing the modal feature information of the time-series sky image data and the modal feature information of the historical photovoltaic power data to obtain a feature matrix; According to the feature matrix, a final potential feature is obtained; Using the final potential features to train a photovoltaic power prediction model; The photovoltaic power prediction model is used to predict photovoltaic power.

2. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 1 is characterized in that: Extracting features from the time series sky image data to obtain modal feature information of the time series sky image data includes: Reduce the dimension of the time series sky image data to obtain a grayscale image, and convert the grayscale image into a grayscale pixel grid image; Define a circle with the center of the grayscale pixel grid as the center and a radius of wherein the circular area does not exceed the range of the grayscale pixel grid image, is an integer and , ; For a radius of The grayscale values ​​of the pixels in the grayscale pixel grid map located in the circular area are averaged to obtain a radius of The local mean corresponding to the circular area when ; When it is greater than 1, for a radius of For the circular area of ​​grayscale pixels in the grid, the distance from the center of the circle is greater than and less than The average grayscale value of the pixels is taken to get the radius The local mean corresponding to the circular area when ; When the circular area touches the edge of the grayscale pixel grid map, the local means are aggregated to obtain the modal feature information of the time series sky image data.

3. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 2 is characterized in that: Extracting features from the time series sky image data to obtain modal feature information of the time series sky image data also includes: The median filtering method is used to calculate the median value of a 3 x 3 fixed window for the feature value points located near the sun in the modal feature information of the time series sky image data, and the median value is used to replace the original feature points.

4. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 1 is characterized in that: Extracting features from the historical photovoltaic power data to obtain modal feature information of the historical photovoltaic power data includes: Calculate the mean of the historical photovoltaic power data and standard deviation ; Calculate abnormal data points in the historical photovoltaic power data according to a Z-score method; Interpolating and resampling missing points and abnormal data points in the historical photovoltaic power data to obtain a photovoltaic power series; After the photovoltaic power sequence is processed by BiLSTM, the forward and reverse outputs are spliced ​​to obtain the modal characteristic information of the historical photovoltaic power data.

5. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 4 is characterized in that: Calculating abnormal data points in the historical photovoltaic power data according to the Z-score method includes: In the formula, represents a photovoltaic power data point. When Z is greater than a preset threshold, the photovoltaic power data point is determined to be an abnormal data point.

6. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 1 is characterized in that: The modal feature information of the time series sky image data and the modal feature information of the historical photovoltaic power data are fused to obtain a feature matrix, including: The modal feature information of the time series sky image data and the modal feature information of the historical photovoltaic power data are aligned with the timestamps and then superimposed to obtain a feature matrix .

7. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 6 is characterized in that: According to the feature matrix, the final potential features are obtained, including: The feature matrix Expanding to obtain a feature vector, and filtering from the feature vector to obtain regularized feature information; The regularized feature information is processed through a secondary feature extraction module and a pooling layer with L1 norm penalty to obtain the final potential feature.

8. The multi-modal fusion photovoltaic power prediction method of all-sky imager data according to claim 7 is characterized in that: Regularized feature information is obtained by screening the feature vector, including: A feature selection module based on L1 norm regularization is constructed, and a group of mean-effective features are selected from the high-dimensional input space by the feature selection module for the feature vector to obtain the regularized feature information.

9. The method for predicting photovoltaic power using multi-modal fusion of all-sky imager data according to claim 7, characterized in that: The secondary feature extraction module is a multi-layer stacked encoder and decoder structure.

10. The photovoltaic power prediction method based on multi-modal fusion of all-sky imager data according to any one of claims 1 to 9, characterized in that: The photovoltaic power prediction model is a linear regression prediction model.

Images