Ultra-short-term photovoltaic power prediction method and system based on sky-ground multimodal data fusion

Abstract

The present invention discloses an ultra-short-term photovoltaic power prediction method and system for fusion of sky-ground multimodal data, including the following steps: S1, acquiring multimodal data; S2, processing and fusing the multimodal data; S3, evaluating the quality of the fused multimodal data, judging whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceeding to the next step; S4, constructing an integrated prediction model, and iteratively training the integrated prediction model using multimodal data; S5, inputting the fused multimodal data into the integrated prediction model to obtain a photovoltaic power prediction trajectory. By fusing sky-ground multimodal data, the present invention can more comprehensively capture the complex changes in the surrounding environment of a photovoltaic power station, thereby improving the accuracy of photovoltaic power prediction. By processing and fusing data from different sources, the advantages of various data can be fully utilized to improve the comprehensiveness and depth of prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy power prediction, and in particular to an ultra-short-term photovoltaic power prediction method based on sky-ground multimodal data fusion. Background Art

[0002] As the global energy structure transforms to renewable energy, photovoltaic power generation has attracted widespread attention due to its clean and efficient characteristics. However, due to the complex influence of factors such as weather and geographical environment, photovoltaic power output has great uncertainty. Among them, random clouds above photovoltaics will affect the real-time power output, causing high instability in the new power system. Due to the unstable shape, thickness and movement speed of clouds, the prediction model needs to process highly random input data, which greatly increases the complexity of the model. At the same time, considering the power fluctuation phenomenon caused by intermittent obstruction of clouds, as well as the local effects caused by cloud obstruction, various factors have brought great challenges to the accurate prediction of photovoltaic power.

[0003] Existing methods for photovoltaic power prediction can be roughly divided into prediction methods based on historical photovoltaic numerical time series data, prediction methods based on numerical weather forecasting (NWP) and multi-physics field simulation, and prediction based on cloud map data. However, such methods are overly dependent on past trends and patterns and cannot accurately predict the impact of sudden weather changes on photovoltaic power. In view of the complex and changeable influence of cloud movement, the conventional use of cloud map data to predict the power output of photovoltaic power stations in a short period of time is not accurate enough. In addition, the multimodal data such as historical power, meteorological text time series, images, videos, etc. contained in satellite cloud maps, drone remote sensing, ground-to-air videos, and ground monitoring have different information characteristics such as data form, spatiotemporal dimension, etc., which makes data processing difficult. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an ultra-short-term photovoltaic power prediction method based on multi-modal data fusion of sky and ground.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] The ultra-short-term photovoltaic power prediction method of the fusion of sky-ground multimodal data includes the following steps: S1, obtaining multimodal data; S2, processing and fusing the multimodal data; S3, evaluating the quality of the fused multimodal data, and judging whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceeding to the next step; S4, constructing an integrated prediction model, and iteratively training the integrated prediction model using multimodal data; S5, inputting the fused multimodal data into the integrated prediction model to obtain the photovoltaic power prediction trajectory.

[0007] Furthermore, step S1 includes: S11, using remote sensing satellites to obtain cloud data, and using drones to obtain cloud data; using ground-to-air cameras to monitor cloud data above the photovoltaic power station in real time; multimodal data includes cloud data obtained by remote sensing satellites, local cloud data obtained by drones, and cloud data obtained by cameras, and the cloud data includes cloud image data; S12, transmitting and sharing the data information obtained by remote sensing satellites, drones, and cameras with a ground control center in real time; S13, time synchronizing the data information obtained by remote sensing satellites, drones, and cameras.

[0008] Furthermore, step S2 includes: S21, extracting multimodal features from each multimodal data; S22, collaboratively fusing cloud image data acquired by remote sensing satellites, drones and cameras.

[0009] Furthermore, S3 specifically includes: evaluating the quality of the fused multimodal data according to the following formula to obtain a peak signal-to-noise ratio: ;

[0010] is the maximum possible noise of the multimodal data signal, and MSE is the mean square error of the signal-to-noise ratio between the fused data and the original data.

[0011] Furthermore, the cloud layer image data is collaboratively fused using the following formula:

[0012] ;

[0013] is the fusion feature result of the i-th layer, N is the number of pyramid layers, is the i-th layer image feature component, is the i+1th layer video feature component, is the upsampling function, Reconstruct the fusion feature results for the front Ni layer, Image fusion data.

[0014] Furthermore, the constructing of the integrated prediction model specifically includes: the integrated prediction model includes a first model, a second model and a multi-layer prediction model.

[0015] Furthermore, step S5 specifically includes: inputting the fused meteorological text time series data into the first model to obtain a first photovoltaic prediction value; inputting the cloud image fusion data into the second model to obtain a second photovoltaic prediction value; obtaining an intermediate prediction value by weighted fusion or stacking generalization of the first photovoltaic prediction value and the second photovoltaic prediction value; inputting the intermediate prediction value into the multi-layer prediction model to obtain the endpoint photovoltaic prediction value; and obtaining a photovoltaic power prediction trajectory based on the time and the corresponding endpoint photovoltaic prediction value.

[0016] Furthermore, step S53 specifically obtains the intermediate prediction value through the following formula: ;in, and are the first photovoltaic prediction value and the second photovoltaic prediction value, respectively. To correct the weight value for the prediction result, P is the bias term, is the intermediate prediction value.

[0017] Furthermore, the preset threshold is 50.

[0018] The present invention also provides an ultra-short-term photovoltaic power prediction system for fusion of sky-ground multimodal data, which is used to implement an ultra-short-term photovoltaic power prediction method for fusion of sky-ground multimodal data, including: a data acquisition module, used to acquire multimodal data; a data processing module, used to process and fuse the multimodal data; a data quality assessment module, used to assess the quality of the fused multimodal data, and determine whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceed to the next step; a prediction model construction module, used to construct an integrated prediction model, and iteratively train the integrated prediction model using multimodal data; a prediction module, used to input the fused multimodal data into the integrated prediction model to obtain a photovoltaic power prediction trajectory.

[0019] The present invention has the following beneficial effects:

[0020] By fusing multimodal data from the sky and the ground, the complex changes in the surrounding environment of the photovoltaic power station can be captured more comprehensively, thereby improving the accuracy of photovoltaic power prediction. Since photovoltaic power output is affected by many uncertain factors, such as random occlusion by clouds, this method can better cope with these uncertainties and reduce prediction errors by analyzing multimodal data in real time. By evaluating the quality of the fused data and judging the reliability of the data based on the peak signal-to-noise ratio, this method can ensure that the data used in the model training and prediction process have high quality and enhance the adaptability of the model to different environmental conditions. By processing and fusing data from different sources (such as satellite cloud images, drone remote sensing, ground monitoring, etc.), this method can make full use of the advantages of various data and improve the comprehensiveness and depth of prediction. By building an integrated prediction model and iteratively training with multimodal data, this method can continuously optimize model performance, adapt to environmental changes, and improve the stability and reliability of prediction. Although historical data is still important in prediction, this method reduces over-reliance on past trends and patterns by analyzing multimodal data in real time, and improves the flexibility and real-time nature of prediction. By providing more accurate photovoltaic power prediction, this method helps the power system better manage the volatility of renewable energy and improve the stability and reliability of the entire system.

[0021] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0023] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0024] Figure 2 It is a schematic flow chart of some steps of the present invention;

[0025] Figure 3 is a photovoltaic power prediction trajectory and an actual power trajectory of an embodiment of the present invention;

[0026] Figure 4 This is a photovoltaic power prediction trajectory and an actual power trajectory according to another embodiment of the present invention. DETAILED DESCRIPTION

[0027] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0030] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0031] Please refer to Figure 1 In a preferred embodiment of the present invention, an ultra-short-term photovoltaic power prediction method based on sky-ground multimodal data fusion includes steps S1, S2, S3, S4 and S5.

[0032] S1, obtain multimodal data; multimodal data includes multi-dimensional cloud images, videos, and meteorological text time series data; meteorological text time series data at least includes ground light intensity, ground temperature, ground humidity, and wind speed, which can be obtained through monitoring sensors.

[0033] S2, process and fuse the multimodal data. Usually, the image and video data provided by different devices are fused, such as satellite cloud images, ground monitoring videos or pictures, and the meteorological text time series data is fused to obtain the fused meteorological text time series data. The fusion of meteorological text time series data can be obtained by a simple weighted accumulation method. The fused meteorological text time series data can be obtained by multiplying each parameter contained in the meteorological text time series data by their respective weight coefficients and then adding them up to obtain the fused meteorological text time series data.

[0034] S3, evaluate the quality of the fused multimodal data, and determine whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceed to the next step, that is, step S4. If not, clean and denoise the fused multimodal data and return to step S3.

[0035] S4. Build an integrated prediction model and use multimodal data to iteratively train the integrated prediction model. During training, the output structure will be verified through historical power data to ensure the effectiveness of the training.

[0036] S5. Input the fused multimodal data into the integrated prediction model to obtain the photovoltaic power prediction trajectory.

[0037] The ultra-short-term photovoltaic power prediction method of sky-ground multimodal data fusion provided by the present invention can more comprehensively capture the complex changes in the surrounding environment of the photovoltaic power station, including the dynamic changes of the cloud layer, the real-time changes of meteorological conditions, etc., by fusing the sky-ground multimodal data, thereby improving the accuracy of photovoltaic power prediction. Since photovoltaic power output is affected by a variety of uncertain factors, such as random occlusion of clouds, this method can better cope with these uncertainties and reduce prediction errors by analyzing multimodal data in real time. By evaluating the quality of the fused data and judging the reliability of the data according to the peak signal-to-noise ratio, the method can ensure that the data used in the model training and prediction process has high quality and enhance the adaptability of the model to different environmental conditions. By processing and fusing data from different sources (such as satellite cloud images, drone remote sensing, ground monitoring, etc.), the method can make full use of the advantages of various data and improve the comprehensiveness and depth of prediction. By constructing an integrated prediction model and using multimodal data for iterative training, the method can continuously optimize the model performance, adapt to environmental changes, and improve the stability and reliability of prediction. Although historical data is still important in prediction, the method reduces excessive reliance on past trends and patterns by analyzing multimodal data in real time, and improves the flexibility and real-time performance of prediction. By providing more accurate PV power forecasts, this method helps power systems better manage the volatility of renewable energy and improve overall system stability and reliability. Figure 3 and Figure 4 It can be seen that the photovoltaic power prediction trajectory of the present invention is almost consistent with the actual photovoltaic power trajectory, the error is small, and the photovoltaic power prediction value is close to the actual value.

[0038] In a specific embodiment of the present invention, step S1 includes steps S11, S12 and S13.

[0039] S11, use remote sensing satellites to obtain cloud data, and use drones to obtain cloud data; use ground-to-air cameras to monitor cloud data above photovoltaic power stations in real time; multimodal data includes cloud data obtained by remote sensing satellites, local cloud data obtained by drones, and cloud data obtained by cameras, and the cloud data includes cloud image data. The cloud image data can be photos or videos, and the video can also be understood as a plurality of pictures played frame by frame. It can be understood that the meteorological text time series data can be obtained through ground monitoring sensor equipment.

[0040] S12. The data information obtained by remote sensing satellites, drones and cameras is transmitted and shared with the ground control center in real time. It is understandable that the meteorological text time series data will also be transmitted and shared with the ground control center in real time.

[0041] S13. Time synchronization is performed on the data information obtained by remote sensing satellites, drones and cameras to ensure the synchronization and accuracy of the data. It is understandable that the meteorological text time series data will also be time synchronized.

[0042] In a specific embodiment of the present invention, step S2 includes steps S21 and S22.

[0043] S21, extract multimodal features from each multimodal data; features can be extracted from satellite cloud images and drone remote sensing images. Use image processing techniques such as edge detection, histogram analysis, texture analysis and other methods to extract image features. Deep learning methods, such as convolutional neural networks (CNN), can be used to automatically extract and learn complex patterns in images. Meteorological text time series data is usually parameterized data and can be directly used for feature extraction. Statistical methods can be used to analyze the time series characteristics of these data. Video data: Ground-to-air video data can provide dynamic cloud coverage. Through video processing techniques such as optical flow and motion detection, the motion characteristics of clouds can be extracted. Sensor data: Sensor data provided by ground monitoring equipment, such as light intensity, temperature, humidity, etc., can be extracted through signal processing techniques, such as fast Fourier transform to analyze frequency domain characteristics.

[0044] S22, collaboratively fuse the cloud image data acquired by the remote sensing satellite, the UAV and the camera. It is understandable that step S22 also includes the meteorological text time series data after the meteorological text time series data is fused. The fusion of the meteorological text time series data can be obtained by a simple weighted accumulation method. The fused meteorological text time series data can be obtained by multiplying each parameter contained in the meteorological text time series data by their respective weight coefficients and then adding them.

[0045] In a specific embodiment of the present invention, step S3 specifically includes: evaluating the quality of the fused multimodal data according to the following formula to obtain the peak signal-to-noise ratio: ; is the maximum possible noise of the multimodal data signal, MSE is the mean square error of the signal-to-noise ratio between the fused data and the original data, is the peak signal-to-noise ratio.

[0046] In a specific embodiment of the present invention, the cloud layer image data is collaboratively fused using the following formula:

[0047] ;

[0048] is the fusion feature result of the i-th layer, N is the number of pyramid layers, is the i-th layer image feature component, which can be sourced from satellite cloud images, pictures taken by drones or cameras; is the i+1th layer video feature component, which can be sourced from videos taken by drones or cameras; is the upsampling function, For Upsampling to improve The purpose of the upsampling function is to increase the resolution. For example, a 2x2 matrix is ​​upsampled to 4x4, and each pixel value of the original feature map is expanded to a 2x2 area. Reconstruct the fusion feature results for the front Ni layer, Image fusion data. Reconstruct the fused feature results for the first N-(i+1) layers. Reconstruct the fused feature results for the first N layers.

[0049] In a specific embodiment of the present invention, the building of an integrated prediction model specifically includes: the integrated prediction model includes a first model, a second model and a multi-layer prediction model.

[0050] Reference Figure 2 In a specific embodiment of the present invention, step S5 specifically includes:

[0051] S51, inputting the fused meteorological text time series data into the first model to obtain a first photovoltaic prediction value;

[0052] S52, inputting the cloud image fusion data into the second model to obtain a second photovoltaic prediction value;

[0053] S53, obtaining an intermediate prediction value by weighted fusion or stacking generalization of the first photovoltaic prediction value and the second photovoltaic prediction value;

[0054] S54, inputting the intermediate prediction value into a multi-layer prediction model to obtain an endpoint photovoltaic prediction value, wherein the multi-layer prediction model may be a multi-layer perceptron MLP model;

[0055] S55, obtaining a photovoltaic power prediction trajectory according to the time and the corresponding endpoint photovoltaic prediction value.

[0056] Step S53 specifically obtains an intermediate prediction value by weighting the first photovoltaic prediction value and the second photovoltaic prediction value through the following formula: ;

[0057] in, and are the first photovoltaic prediction value and the second photovoltaic prediction value, respectively. To correct the weight value for the prediction result, P is the bias term, is the intermediate prediction value.

[0058] During training, the historical data also needs to be processed and judged in steps S2 and S3, and the fused data is passed through S51 to S54 to obtain the terminal photovoltaic prediction value. , and then compare the terminal photovoltaic prediction value with the historical power to obtain the error measurement , If the error metric is greater than the preset threshold, Adjustment is performed, and then training is performed after adjustment, thereby achieving adaptive weighted adjustment feedback.

[0059] It is understandable that during training, it is also necessary to input the historical fused meteorological text time series data into the first model for training, and compare the first photovoltaic prediction value with the historical photovoltaic power to obtain a trained first model to perform step S51; the training of the second model also requires the input of the historical cloud image fusion data into the second model, and the comparison and verification of the second photovoltaic prediction value with the historical photovoltaic power to obtain a trained second model. The multi-layer perceptron MLP model also needs to be trained with historical data. It is understandable that the first model can be an LSTM (Long Short-Term Memory Network) model, and the second model can be a CNN (Convolutional Neural Network) model.

[0060] In a specific embodiment of the present invention, the preset threshold is 50, in db.

[0061] The present invention also provides an ultra-short-term photovoltaic power prediction system for fusion of sky-ground multimodal data, including: a data acquisition module, used to acquire multimodal data; a data processing module, used to process and fuse the multimodal data; a data quality assessment module, used to assess the quality of the fused multimodal data, and determine whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceed to the next step; a prediction model construction module, used to construct an integrated prediction model, and iteratively train the integrated prediction model using multimodal data; a prediction module, used to input the fused multimodal data into the integrated prediction model to obtain a photovoltaic power prediction trajectory.

[0062] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An ultra-short-term photovoltaic power prediction method based on multi-modal data fusion of sky and ground, characterized in that: The steps include: S1, obtain multimodal data; S2, processes and fuses multimodal data; S3, evaluating the quality of the fused multimodal data to determine whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceeding to the next step; S4. Build an integrated prediction model and iteratively train the integrated prediction model using multimodal data; S5, inputting the fused multimodal data into the integrated prediction model to obtain the photovoltaic power prediction trajectory; Step S1 includes: S11, using remote sensing satellites to obtain cloud data, using drones to obtain cloud data; using ground-to-air cameras to monitor cloud data above the photovoltaic power station in real time; the multimodal data includes cloud data obtained by remote sensing satellites, local cloud data obtained by drones, and cloud data obtained by cameras, and the cloud data includes cloud image data; S12, transmit and share data information obtained by remote sensing satellites, drones and cameras with the ground control center in real time; S13, time synchronization of data information acquired by remote sensing satellites, drones and cameras; Step S2 includes: S21, extracting multimodal features from each multimodal data; S22, collaborative fusion of cloud image data acquired by remote sensing satellites, drones and cameras; The cloud layer image data is collaboratively fused using the following formula: ； is the fusion feature result of the i-th layer, N is the number of pyramid layers, is the i-th layer image feature component, is the i+1th layer video feature component, is the upsampling function, Reconstruct the fusion feature results for the front Ni layer, is the image fusion data; The building of the integrated prediction model specifically includes: The integrated prediction model includes a first model, a second model and a multi-layer prediction model; Step S5 specifically includes: S51, inputting the fused meteorological text time series data into the first model to obtain a first photovoltaic prediction value; S52, inputting the cloud image fusion data into the second model to obtain a second photovoltaic prediction value; S53, obtaining an intermediate prediction value by weighted fusion or stacking generalization of the first photovoltaic prediction value and the second photovoltaic prediction value; S54, inputting the intermediate prediction value into the multi-layer prediction model to obtain the endpoint photovoltaic prediction value; S55, obtaining a photovoltaic power prediction trajectory according to the time and the corresponding endpoint photovoltaic prediction value.

2. The ultra-short-term photovoltaic power prediction method based on sky-ground multimodal data fusion according to claim 1 is characterized in that: S3 specifically includes: The quality of the fused multimodal data is evaluated according to the following formula to obtain the peak signal-to-noise ratio: ； is the maximum possible noise of the multimodal data signal, MSE is the mean square error of the signal-to-noise ratio between the fused data and the original data, is the peak signal-to-noise ratio.

3. The ultra-short-term photovoltaic power prediction method based on sky-ground multimodal data fusion according to claim 1 is characterized in that: Step S53 specifically obtains the intermediate prediction value through the following formula: ； in, and are the first photovoltaic prediction value and the second photovoltaic prediction value, respectively. To correct the weight value for the prediction result, P is the bias term, is the intermediate prediction value.

4. The ultra-short-term photovoltaic power prediction method based on sky-ground multimodal data fusion according to claim 1 is characterized in that: The preset threshold is 50.

5. An ultra-short-term photovoltaic power prediction system based on sky-ground multimodal data fusion, used to implement the ultra-short-term photovoltaic power prediction method based on sky-ground multimodal data fusion as claimed in any one of claims 1 to 4, characterized in that: include: A data acquisition module, used for acquiring multimodal data; A data processing module, used to process and fuse multimodal data; The data quality assessment module is used to assess the quality of the fused multimodal data and determine whether the peak signal-to-noise ratio of the fused multimodal data is greater than or equal to a preset threshold: if so, proceed to the next step; The prediction model building module is used to build an integrated prediction model and iteratively train the integrated prediction model using multimodal data; The prediction module is used to input the fused multi-modal data into the integrated prediction model to obtain the photovoltaic power prediction trajectory.

Images