Photovoltaic short-term power prediction method and device, server and storage medium

Abstract

The invention provides a photovoltaic short-term power prediction method and device, a server and a storage medium, and the method comprises the steps: generating first solar radiation data based on a meteorological large model and WRF-Solar; performing solar radiation data training reasoning on the meteorological large model according to the historical meteorological reanalysis data set; performing fine adjustment on the meteorological large model after solar radiation data training reasoning according to the actually measured solar radiation data of the cloud picture; generating second solar radiation data based on the fine-tuned meteorological large model; performing fusion processing on the first solar radiation data and the second solar radiation data to generate fusion data; and inputting the fused data into a photovoltaic short-term power prediction model to generate a photovoltaic short-term power prediction result. According to the photovoltaic short-term power prediction method and device, the server and the storage medium provided by the invention, the combination of the meteorological large model, the satellite cloud picture and photovoltaic short-term power prediction can be realized, and the accuracy and effectiveness of photovoltaic short-term power prediction are optimized.

Description

Technical Field

[0001] The invention belongs to the technical field of photovoltaic power prediction, and in particular relates to a photovoltaic short-term power prediction method, device, server and storage medium. Background Art

[0002] Photovoltaic power forecasting is a method of predicting the active power of photovoltaic power stations based on meteorological conditions and statistical laws. In actual work, accurate and efficient photovoltaic power forecasting results can maintain grid stability and improve energy storage efficiency, thereby promoting the healthy development of the power market.

[0003] In the prior art, since satellite cloud images can observe and record the solar radiation conditions in the real atmosphere from top to bottom, have a wide coverage, and can update data efficiently, photovoltaic power prediction work usually uses satellite cloud images to obtain real-time solar irradiation data. However, satellite cloud images have poor prediction capabilities for future solar irradiation data, and the prediction accuracy of future solar irradiation data will directly affect the accuracy of photovoltaic power prediction. Therefore, how to accurately predict solar irradiation data has become a key problem to be solved by technicians in this field.

[0004] In the context of the current development of AI technology, meteorological big models, as an emerging frontier technology, have achieved rapid development with the breakthrough progress of deep learning algorithms and graph neural networks and the significant enhancement of GPU computing power. These models have shown great potential in the traditional meteorological field, especially in wind power energy forecasting. However, the existing meteorological big models cannot directly process and predict solar irradiation data, so how to combine meteorological big models with satellite cloud images and apply them to the field of photovoltaic power forecasting is of vital importance. Summary of the invention

[0005] In view of this, the present invention aims to propose a photovoltaic short-term power prediction method, device, server and storage medium, so as to realize the application of large meteorological models and satellite cloud images in the field of photovoltaic power prediction, thereby improving the accuracy and effectiveness of photovoltaic power prediction.

[0006] To achieve the above object, the technical solution created by the present invention is implemented as follows:

[0007] In a first aspect, an embodiment of the present invention provides a photovoltaic short-term power prediction method, comprising:

[0008] Deploy a large meteorological model and generate the first solar radiation data based on the large meteorological model and WRF-Solar;

[0009] Obtain historical meteorological reanalysis datasets, and use them to train and infer solar radiation data for the large meteorological model;

[0010] Obtain satellite cloud image data, determine the cloud image measured solar radiation data based on the satellite cloud image data, and fine-tune the large meteorological model after solar radiation data training and inference based on the cloud image measured solar radiation data;

[0011] Generate the second solar radiation data based on the fine-tuned meteorological model;

[0012] Performing fusion processing on the first solar radiation data and the second solar radiation data to generate fusion data;

[0013] A photovoltaic short-term power prediction model is built based on Transformer, and the fusion data is input into the photovoltaic short-term power prediction model to generate photovoltaic short-term power prediction results.

[0014] Furthermore, the generating of the first solar radiation data based on the meteorological large model and WRF-Solar includes:

[0015] Obtain a weather forecast data set, pre-train a large weather model using the weather forecast data set, and generate background field data using the pre-trained large weather model;

[0016] The geographic location information of the target site is obtained, and the simulation target area is delineated in WRF-Solar. The geographic location information of the target site is interpolated into the simulation target area, and the background field data is interpolated into the simulation target area to generate the first solar radiation data.

[0017] Furthermore, the weather forecast data set includes weather element forecast data for the ground layer and the pressure layer, and the weather element forecast data includes: 2m temperature, u-component wind speed at 10m, v-component wind speed at 10m, sea level pressure, total precipitation, temperature, u-component wind speed, v-component wind speed, w-component wind speed, potential height and specific humidity.

[0018] Further, the historical meteorological reanalysis data set includes historical data of meteorological elements of the ground layer and the pressure layer, and the historical data of meteorological elements include: 2m temperature, u-component wind speed at 10m, v-component wind speed at 10m, sea level pressure, total precipitation, total irradiance, direct irradiance, top of atmosphere radiation, temperature, u-component wind speed, v-component wind speed, w-component wind speed, geopotential height and specific humidity;

[0019] The solar radiation data training and reasoning of the meteorological model based on the historical meteorological reanalysis data set includes:

[0020] Taking total irradiance and direct irradiance as training targets, top of atmosphere radiation as external forcing term, and historical data of other meteorological elements in historical meteorological reanalysis dataset as input data, the solar radiation data is trained and inferred for the large meteorological model.

[0021] Furthermore, the fine-tuning of the meteorological large model after training and reasoning with solar radiation data according to the solar radiation data measured by the cloud image includes:

[0022] The measured solar radiation data of the cloud image is set as the new training target, and the meteorological model is retrained after the solar radiation data training and inference. The weight parameters of the meteorological model are adjusted, and the MSE function is used as the loss function and the gradient descent algorithm is used as the optimization algorithm to optimize the convergence and stability of the meteorological model.

[0023] Furthermore, the fusing the first solar radiation data and the second solar radiation data to generate fused data includes:

[0024] interpolating the first solar radiation data and the second solar radiation data to the target site to generate first irradiance time series data and second irradiance time series data;

[0025] Obtain the ground-measured irradiance data of the target site, use the ground-measured irradiance data as training labels and the elastic network model as the training algorithm to train a data fusion model;

[0026] The first irradiance time series data and the second irradiance time series data are input into a data fusion model to generate fused data.

[0027] Furthermore, after the photovoltaic short-term power prediction model is constructed based on Transformer, the photovoltaic short-term power prediction method further includes:

[0028] Obtain the historical actual power data and historical measured irradiance data of the target site, and build an abnormal data recognition algorithm based on the historical actual power data and historical measured irradiance data;

[0029] The historical actual power data is cleaned through abnormal data identification algorithm;

[0030] The photovoltaic short-term power prediction model is trained based on the historical measured irradiance data and the historical actual power data after data cleaning.

[0031] In a second aspect, an embodiment of the present invention further provides a photovoltaic short-term power prediction device, comprising:

[0032] The first generation module is used to deploy the meteorological large model and generate the first solar radiation data based on the meteorological large model and WRF-Solar;

[0033] The training module is used to obtain historical meteorological reanalysis data sets and perform solar radiation data training and inference on the meteorological large model based on the historical meteorological reanalysis data sets;

[0034] A fine-tuning module is used to obtain satellite cloud image data, determine the cloud image measured solar radiation data based on the satellite cloud image data, and fine-tune the meteorological large model after solar radiation data training and inference based on the cloud image measured solar radiation data;

[0035] A second generation module is used to generate second solar radiation data based on the fine-tuned meteorological large model;

[0036] A fusion module, used for fusing the first solar radiation data and the second solar radiation data to generate fused data;

[0037] The prediction module is used to build a photovoltaic short-term power prediction model based on Transformer, input the fusion data into the photovoltaic short-term power prediction model, and generate photovoltaic short-term power prediction results.

[0038] In a third aspect, an embodiment of the present invention further provides a server, including:

[0039] one or more processors;

[0040] A storage device for storing one or more programs;

[0041] When the one or more programs are executed by the one or more processors, the one or more processors implement the photovoltaic short-term power prediction method provided in the above embodiment.

[0042] In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer executable instructions, wherein the computer executable instructions, when executed by a computer processor, are used to execute the photovoltaic short-term power prediction method provided in the above embodiment.

[0043] Compared with the prior art, the photovoltaic short-term power prediction method, device, server and storage medium created by the present invention have the following advantages:

[0044] The photovoltaic short-term power prediction method, device, server and storage medium described in the invention can generate the first solar radiation data through the meteorological big model and WRF-Solar, and enable the meteorological big model to obtain the ability to predict solar radiation data through the historical meteorological reanalysis data set and satellite cloud map, thereby generating the second solar radiation data. The first solar radiation data and the second solar radiation data can also be fused to generate fused data, so as to use the fused data and the photovoltaic short-term power prediction model to predict the photovoltaic short-term power. Compared with the prior art, the invention can give full play to the powerful data processing and analysis capabilities of the meteorological big model, and reasonably use the high-precision solar irradiation data of the satellite cloud map, realize the combination of the meteorological big model, the satellite cloud map and the photovoltaic short-term power prediction, and optimize the accuracy and effectiveness of the photovoltaic short-term power prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings constituting part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments and descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation on the present invention. In the drawings:

[0046] Figure 1 A flowchart of the photovoltaic short-term power prediction method according to the first embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the structure of the photovoltaic short-term power prediction device according to the second embodiment of the present invention;

[0048] Figure 3 This is a structural diagram of the server described in Example 3 of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0050] Embodiment 1

[0051] Figure 1 1 is a flow chart of a photovoltaic short-term power prediction method provided by Embodiment 1 of the present invention. As shown in the figure, the photovoltaic short-term power prediction method specifically includes the following steps:

[0052] Step 110: deploy a large meteorological model, and generate first solar radiation data based on the large meteorological model and WRF-Solar.

[0053] The meteorological big model is a high-precision AI prediction model used to predict meteorological elements. It plays an irreplaceable and key role in many aspects, such as conventional weather forecasts, extreme event warnings, climate change scenario analysis, ecological and environmental protection decision support, and agricultural meteorological disaster risk management. There are many types of meteorological big model products in the existing technology, such as the "Fengqing" meteorological big model jointly developed by the China Meteorological Administration and Tsinghua University, the Huawei Pangu meteorological big model, the Fengwu meteorological big model of the Shanghai Artificial Intelligence Laboratory, the Fuxi meteorological big model of Fudan University, the Alibaba SwinRDM meteorological big model, the Microsoft Aurora meteorological big model, the Google NeuralGCM meteorological big model, the Google DeepMind GraphCast meteorological big model, and the NVIDIA FourCastNet big model.

[0054] Since different meteorological big models have their own installation and deployment methods, they can also be deployed by uniformly calling the inference module through the ai-models library. Therefore, in this embodiment, Google DeepMind's GraphCast meteorological big model will be taken as an example to explain the subsequent solutions.

[0055] For example, when deploying the GraphCast meteorological large model, the ai-models library and the GraphCast pre-trained model weight file can be installed through pip in a Linux environment to complete the deployment of the meteorological large model.

[0056] Since existing large meteorological model products do not have the ability to directly predict solar radiation data, in this embodiment, the large meteorological model and WRF-Solar are combined to generate the first solar radiation data representing the irradiance.

[0057] It should be noted that WRF-Solar is a highly advanced meteorological model system developed specifically to improve the accuracy of solar resource assessment and prediction. It can simulate the solar radiation transmission process and its attenuation effect under different atmospheric conditions in a refined manner, thereby more accurately capturing the various factors affecting the amount of solar radiation received by the ground, so as to improve the quality of irradiance prediction.

[0058] In this embodiment, based on the meteorological big model and WRF-Solar, generating the first solar radiation data can be specifically optimized into the following steps:

[0059] Obtain a weather forecast data set, pre-train the meteorological model through the weather forecast data set, and generate background field data through the pre-trained meteorological model; obtain the geographical location information of the target site, delineate the simulation target area in WRF-Solar, interpolate the geographical location information of the target site into the simulation target area, and interpolate the background field data into the simulation target area to generate the first solar radiation data.

[0060] The meteorological forecast dataset can be provided by one or more of the following: China Meteorological Administration CMA (GRAPES), European Medium-Range Weather Forecast ECMWF (HRES), National Center for Environmental Prediction NCEP (GFS), Canadian Meteorological Center CMC (GDPS), German Meteorological Service DWD (ICON), Japan Meteorological Agency JMA (GSM), French Meteorological Service MF (ARPEGE) or Australian Bureau of Meteorology BOM (ACCESS-G). The meteorological forecast dataset should include meteorological element forecast data of ground layer and pressure layer, and meteorological element forecast data should include: 2m temperature, 10m u-component wind speed, 10m v-component wind speed, sea level pressure, total precipitation, temperature, u-component wind speed, v-component wind speed, w-component wind speed, geopotential height and specific humidity.

[0061] Exemplarily, this embodiment can use ECMWF's HRES data accumulated over one year to the 10 days before the day to form a weather forecast data set and input it into the GraphCast meteorological model for pre-training. After completing the pre-training of the meteorological model, the GraphCast meteorological model business version can be selected as the driving data (its spatial resolution is 0.25 degrees, and there are 13 pressure layers vertically, namely 50hPa, 100hPa, 150hPa, 200hPa, 250hPa, 300hPa, 400hPa, 500hPa, 600hPa, 700hPa, 850hPa, 925hPa and 1000hPa), and use ai-models to infer the GraphCast meteorological model to generate background field data. The background field data should include more than one year of meteorological variable data, and the meteorological element information corresponding to the meteorological variable data should be the same as the meteorological element information included in the meteorological element forecast data.

[0062] After the background field data is generated, the geographic location information of the target site can be obtained according to the location distribution of the target site, and the simulation target area can be delineated in WRF-Solar. Then, the geographic location information of the target site is interpolated into the simulation target area, and the background field data is interpolated into the simulation target area accordingly, so as to generate the first solar radiation data with the help of the large meteorological model and WRF-Solar.

[0063] Step 120: Obtain a historical meteorological reanalysis data set, and perform solar radiation data training and inference on a large meteorological model based on the historical meteorological reanalysis data set.

[0064] Since the traditional meteorological big model does not have the ability to predict and infer solar radiation related data, in this embodiment, the meteorological big model will be trained and inferred through the historical meteorological reanalysis data set so that the meteorological big model can predict the solar radiation data.

[0065] Optionally, the historical meteorological reanalysis data set can be provided by one or more of ECMWF's ERA5 reanalysis data, NOAA's FNL reanalysis data, or JMA's JRA reanalysis data. The historical meteorological reanalysis data set should include historical data of meteorological elements of the ground layer and the pressure layer, and the historical data of meteorological elements should include: 2m temperature, 10m u-component wind speed, 10m v-component wind speed, sea level pressure, total precipitation, total irradiance, direct irradiance, top of atmosphere radiation, temperature, u-component wind speed, v-component wind speed, w-component wind speed, geopotential height, specific humidity and other meteorological element information. Among them, total irradiance, direct irradiance and top of atmosphere radiation are meteorological element information added specifically for solar radiation data prediction.

[0066] Exemplarily, in this embodiment, the ERA5 reanalysis data from 1979 to 2017 can be used to form a historical meteorological reanalysis data set. Accordingly, when the meteorological big model is trained and inferred on solar radiation data based on the historical meteorological reanalysis data set, the historical meteorological reanalysis data set can be used under Linux to implement the GraphCast meteorological big model training and inference process. At this time, the total irradiance and direct irradiance should be used as training targets, the top of the atmosphere radiation as the external forcing item, and the historical data of other meteorological elements in the historical meteorological reanalysis data set as input data to train and infer solar radiation data on the meteorological big model, so that the meteorological big model can increase the ability to predict total irradiance and direct irradiance on the basis of the pre-trained model.

[0067] Step 130: Obtain satellite cloud image data, determine the cloud image measured solar radiation data based on the satellite cloud image data, and fine-tune the large meteorological model after solar radiation data training and inference based on the cloud image measured solar radiation data.

[0068] Satellite cloud images are images of cloud cover and surface features on the earth observed from top to bottom by meteorological satellites. Satellite cloud images can be used to identify different weather systems and the corresponding solar radiation conditions, determine their locations, estimate their intensity and development trends, and thus provide a basis for weather analysis and weather forecasting. In this embodiment, satellite cloud image data can be obtained through Japan's Himawari satellite cloud image or China's Fengyun-4 satellite cloud image. After obtaining the satellite cloud image data, the satellite cloud image data can determine the solar radiation data measured by the cloud image, so as to fine-tune the meteorological large model according to the solar radiation data measured by the cloud image, and enhance the solar radiation data prediction capability of the meteorological large model in the cloud cover change scenario.

[0069] Optionally, in this embodiment, the meteorological large model after training and reasoning the solar radiation data is fine-tuned according to the cloud image measured solar radiation data, which can be specifically optimized into the following steps:

[0070] The measured solar radiation data of the cloud image is set as the new training target, and the meteorological model is retrained after the solar radiation data training and inference. The weight parameters of the meteorological model are adjusted, and the MSE function is used as the loss function and the gradient descent algorithm is used as the optimization algorithm to optimize the convergence and stability of the meteorological model.

[0071] When retraining the meteorological big model, the pre-trained weight parameters of the GraphCast meteorological big model should be loaded first. Since the pre-trained meteorological big model already has basic prediction capabilities for most meteorological element information, the pre-trained weight parameters can be used as the basis for subsequent adjustments. Then, the cloud map measured solar radiation data is set as the new training target of the meteorological big model, and the weight parameters of the meteorological big model are adjusted to minimize the difference between the model prediction results and the cloud map measured solar radiation data. Accordingly, in order to optimize the convergence and stability of the meteorological big model when making predictions, this embodiment will also use the MSE function as the loss function of the meteorological big model, and use the gradient algorithm as the optimization algorithm, and select the Adam optimizer to optimize the meteorological big model, so as to quantify the difference between the prediction results of the meteorological big model and the real data, and improve the convergence and stability of the meteorological big model.

[0072] In addition, when fine-tuning the large meteorological model, early stopping settings can be set and the learning rate can be reduced to avoid overfitting problems.

[0073] When the meteorological model is fine-tuned, the batch size can be adjusted to balance the training speed and memory usage. When iterative processing is performed, each iteration should process the complete data set, and multiple iterations should be repeated until the early stopping condition or the predetermined maximum number of iterations is reached, thereby updating the parameters of the meteorological model through multiple iterations. During the training process, an independent validation data set can also be used regularly to evaluate the performance of the meteorological model to ensure that the optimization has effectively improved the prediction ability of the meteorological model.

[0074] Step 140: Generate second solar radiation data based on the fine-tuned meteorological model.

[0075] After the meteorological big model completes the training, inference and fine-tuning of solar radiation data, the meteorological big model will have the ability to predict irradiance-related data. Therefore, the solar radiation data can be directly inferred and predicted through the meteorological big model to generate second solar radiation data.

[0076] Step 150: Fusing the first solar radiation data and the second solar radiation data to generate fused data.

[0077] After obtaining the first solar radiation data generated based on the meteorological big model and WRF-Solar and the second solar radiation data generated based on the inference prediction of the meteorological big model, this embodiment will fuse them to generate fused data, so as to facilitate photovoltaic short-term power prediction through the fused data.

[0078] In this embodiment, the first solar radiation data and the second solar radiation data are fused to generate fused data, which can be specifically optimized as follows:

[0079] The first solar radiation data and the second solar radiation data are interpolated to the target site to generate the first irradiance time series data and the second irradiance time series data; the ground measured irradiance data of the target site is obtained, and the ground measured irradiance data is used as the training label and the elastic network model is used as the training algorithm to train the data fusion model; the first irradiance time series data and the second irradiance time series data are input into the data fusion model to generate fused data.

[0080] Step 160: construct a photovoltaic short-term power prediction model based on Transformer, input the fused data into the photovoltaic short-term power prediction model, and generate a photovoltaic short-term power prediction result.

[0081] In order to realize the prediction of photovoltaic short-term power, this embodiment will select the Transformer architecture as the basic algorithm to build a photovoltaic short-term power prediction model. It should be noted that Transformer is a deep learning framework based on the self-attention mechanism, which consists of an encoder and a decoder. Each encoder consists of a multi-head self-attention mechanism and a feedforward network. Each part is followed by a residual connection and a normalization layer, and its decoder is also a similar structure.

[0082] After the construction of the photovoltaic short-term power prediction model is completed, the fused data can be input into the photovoltaic short-term power prediction model to generate the photovoltaic short-term power prediction results.

[0083] Optionally, in order to improve the prediction accuracy of the photovoltaic short-term power prediction model, after building the photovoltaic short-term power prediction model based on Transformer, the photovoltaic short-term power prediction method can also add the following steps:

[0084] The historical actual power data and historical measured irradiance data of the target site are obtained, and an abnormal data recognition algorithm is constructed based on the historical actual power data and the historical measured irradiance data; the historical actual power data is cleaned by the abnormal data recognition algorithm; and the photovoltaic short-term power prediction model is trained based on the historical measured irradiance data and the historical actual power data after data cleaning.

[0085] It should be noted that the historical measured irradiance data described in the above steps may include the historical cloud map measured solar radiation data of the target site obtained through satellite cloud map data, and may also include the historical ground measured irradiance data of the target site observed by a ground-based irradiator. The abnormal data recognition algorithm can remove the abnormal data in the historical actual power data, thereby preventing the prediction result accuracy of the photovoltaic short-term power prediction model from being affected by the abnormal data.

[0086] This embodiment generates the first solar radiation data through the meteorological big model and WRF-Solar, and enables the meteorological big model to obtain the ability to predict solar radiation data through the historical meteorological reanalysis data set and satellite cloud map, thereby generating the second solar radiation data. The first solar radiation data and the second solar radiation data can also be fused to generate fused data, so as to use the fused data and the photovoltaic short-term power prediction model to predict photovoltaic short-term power. In this way, the powerful data processing and analysis capabilities of the meteorological big model are fully utilized, and the high-precision solar irradiation data of the satellite cloud map is reasonably used to realize the combination of the meteorological big model, satellite cloud map and photovoltaic short-term power prediction, and optimize the accuracy and effectiveness of photovoltaic short-term power prediction.

[0087] Embodiment 2

[0088] Figure 2 Schematic diagram of the structure of the photovoltaic short-term power prediction device provided by the second embodiment of the present invention. Figure 2 As shown, the photovoltaic short-term power prediction device includes:

[0089] A first generating module 210 is used to deploy a large meteorological model and generate first solar radiation data based on the large meteorological model and WRF-Solar;

[0090] The training module 220 is used to obtain a historical meteorological reanalysis data set, and perform solar radiation data training and inference on the meteorological large model according to the historical meteorological reanalysis data set;

[0091] The fine-tuning module 230 is used to obtain satellite cloud image data, determine the cloud image measured solar radiation data according to the satellite cloud image data, and fine-tune the meteorological large model after the solar radiation data training and inference is performed according to the cloud image measured solar radiation data;

[0092] A second generating module 240 is used to generate second solar radiation data based on the fine-tuned meteorological large model;

[0093] A fusion module 250 is used to fuse the first solar radiation data and the second solar radiation data to generate fused data;

[0094] The prediction module 260 is used to construct a photovoltaic short-term power prediction model based on Transformer, input the fusion data into the photovoltaic short-term power prediction model, and generate a photovoltaic short-term power prediction result.

[0095] The photovoltaic short-term power prediction device provided in this embodiment can generate the first solar radiation data based on the meteorological large model and WRF-Solar through the first generation module, and can generate the second solar radiation data predicted by the meteorological large model through the cooperation of the training module, the fine-tuning module and the second generation module. It can also fuse the first solar radiation data and the second solar radiation data through the fusion module, and realize the prediction of photovoltaic short-term power through the prediction module, thereby realizing the combination of the meteorological large model, satellite cloud map and photovoltaic short-term power prediction, and optimizing the accuracy and effectiveness of photovoltaic short-term power prediction.

[0096] Based on the above embodiment, the first generating module includes:

[0097] A background field data generation unit is used to obtain a weather forecast data set, pre-train a large weather model using the weather forecast data set, and generate background field data using the pre-trained large weather model;

[0098] The first solar radiation data generating unit is used to obtain the geographical location information of the target site, define the simulation target area in WRF-Solar, interpolate the geographical location information of the target site into the simulation target area, and interpolate the background field data into the simulation target area to generate the first solar radiation data.

[0099] Based on the above embodiment, the fusion module includes:

[0100] An interpolation processing unit, used for interpolating the first solar radiation data and the second solar radiation data to the target site to generate the first irradiance time series data and the second irradiance time series data;

[0101] A model training unit is used to obtain the ground-measured irradiance data of the target site, use the ground-measured irradiance data as training labels and the elastic network model as a training algorithm to train a data fusion model;

[0102] The fusion unit is used to input the first irradiance time series data and the second irradiance time series data into a data fusion model to generate fused data.

[0103] Based on the above embodiment, the photovoltaic short-term power prediction device further includes:

[0104] The data cleaning module is used to obtain the historical actual power data and historical measured irradiance data of the target site, build an abnormal data recognition algorithm based on the historical actual power data and historical measured irradiance data; clean the historical actual power data through the abnormal data recognition algorithm; and train the photovoltaic short-term power prediction model based on the historical measured irradiance data and the historical actual power data after data cleaning.

[0105] The photovoltaic short-term power prediction device provided in the embodiment of the present invention can execute the photovoltaic short-term power prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0106] Embodiment 3

[0107] Figure 3 A schematic diagram of the structure of a server provided in Embodiment 3 of the present invention. Figure 3 A block diagram of an exemplary server 12 suitable for use in implementing embodiments of the present invention is shown. Figure 3 The server 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0108] like Figure 3 As shown, the server 12 is in the form of a general purpose computing device. The components of the server 12 may include, but are not limited to: one or more processors or processing units 16, server memory 28, and a bus 18 connecting different server components (including server memory 28 and processing unit 16).

[0109] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0110] The server 12 typically includes a variety of computer server readable media. These media can be any available media that can be accessed by the server 12, including volatile and non-volatile media, removable and non-removable media.

[0111] Server memory 28 may include computer server readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache 32. Server 12 may further include other removable / non-removable, volatile / non-volatile computer server storage media. By way of example only, storage server 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 3 not shown, usually called a "hard drive"). Although Figure 3 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The server memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0112] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the server memory 28, such program modules 42 including, but not limited to, operating a server, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0113] The server 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the server 12, and / or may communicate with any device that enables the server 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. In addition, the server 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the server 12 via a bus 18. It should be understood that, although not shown in the figure, the server 12 may be assembled to use other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID servers, tape drives, and data backup storage servers, etc.

[0114] The processing unit 16 executes various functional applications and data processing by running the programs stored in the server memory 28 , for example, implementing the photovoltaic short-term power prediction method provided in the embodiment of the present invention.

[0115] Embodiment 4

[0116] Embodiment 4 of the present invention further provides a storage medium comprising computer executable instructions, wherein the computer executable instructions, when executed by a computer processor, are used to execute any of the photovoltaic short-term power prediction methods provided in the above embodiments.

[0117] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0118] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0119] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0120] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or device. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider). 。

[0121] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A photovoltaic short-term power prediction method, characterized in that include: Deploy a large meteorological model and generate the first solar radiation data based on the large meteorological model and WRF-Solar; Obtain historical meteorological reanalysis datasets, and use them to train and infer solar radiation data for the meteorological large model; Obtain satellite cloud image data, determine the cloud image measured solar radiation data based on the satellite cloud image data, and fine-tune the large meteorological model after solar radiation data training and inference based on the cloud image measured solar radiation data; Generate the second solar radiation data based on the fine-tuned meteorological model; Performing fusion processing on the first solar radiation data and the second solar radiation data to generate fusion data; A photovoltaic short-term power prediction model is built based on Transformer, and the fusion data is input into the photovoltaic short-term power prediction model to generate photovoltaic short-term power prediction results.

2. The photovoltaic short-term power prediction method according to claim 1, characterized in that: The first solar radiation data is generated based on the meteorological large model and WRF-Solar, including: Obtain a weather forecast data set, pre-train a large weather model using the weather forecast data set, and generate background field data using the pre-trained large weather model; The geographic location information of the target site is obtained, and the simulation target area is delineated in WRF-Solar. The geographic location information of the target site is interpolated into the simulation target area, and the background field data is interpolated into the simulation target area to generate the first solar radiation data.

3. The photovoltaic short-term power prediction method according to claim 2, characterized in that: The meteorological forecast data set includes meteorological element forecast data of the ground layer and the pressure layer, and the meteorological element forecast data includes: 2m temperature, 10m u-component wind speed, 10m v-component wind speed, sea level pressure, total precipitation, temperature, u-component wind speed, v-component wind speed, w-component wind speed, potential height and specific humidity.

4. The photovoltaic short-term power prediction method according to claim 1, characterized in that: The historical meteorological reanalysis data set includes historical data of meteorological elements of the ground layer and the pressure layer, and the historical data of meteorological elements include: 2m temperature, u-component wind speed at 10m, v-component wind speed at 10m, sea level pressure, total precipitation, total irradiance, direct irradiance, top of atmosphere radiation, temperature, u-component wind speed, v-component wind speed, w-component wind speed, geopotential height and specific humidity; The solar radiation data training and reasoning of the meteorological model based on the historical meteorological reanalysis data set includes: Taking total irradiance and direct irradiance as training targets, top of atmosphere radiation as external forcing term, and historical data of other meteorological elements in historical meteorological reanalysis dataset as input data, the solar radiation data is trained and inferred for the large meteorological model.

5. The photovoltaic short-term power prediction method according to claim 1, characterized in that: The method of fine-tuning the meteorological model after training and reasoning with solar radiation data according to the solar radiation data measured by the cloud image includes: The measured solar radiation data of the cloud image is set as the new training target, and the meteorological model is retrained after the solar radiation data training and inference. The weight parameters of the meteorological model are adjusted, and the MSE function is used as the loss function and the gradient descent algorithm is used as the optimization algorithm to optimize the convergence and stability of the meteorological model.

6. The photovoltaic short-term power prediction method according to claim 1, characterized in that: The fusing the first solar radiation data and the second solar radiation data to generate fused data includes: interpolating the first solar radiation data and the second solar radiation data to the target site to generate first irradiance time series data and second irradiance time series data; Obtain the ground-measured irradiance data of the target site, use the ground-measured irradiance data as training labels and the elastic network model as the training algorithm to train a data fusion model; The first irradiance time series data and the second irradiance time series data are input into a data fusion model to generate fused data.

7. The photovoltaic short-term power prediction method according to claim 1, characterized in that: After the photovoltaic short-term power prediction model is constructed based on Transformer, the photovoltaic short-term power prediction method further includes: Obtain the historical actual power data and historical measured irradiance data of the target site, and build an abnormal data recognition algorithm based on the historical actual power data and historical measured irradiance data; The historical actual power data is cleaned through abnormal data identification algorithm; The photovoltaic short-term power prediction model is trained based on the historical measured irradiance data and the historical actual power data after data cleaning.

8. A photovoltaic short-term power prediction device, characterized in that: include: The first generation module is used to deploy the meteorological large model and generate the first solar radiation data based on the meteorological large model and WRF-Solar; The training module is used to obtain historical meteorological reanalysis data sets and perform solar radiation data training and inference on the meteorological large model based on the historical meteorological reanalysis data sets; A fine-tuning module is used to obtain satellite cloud image data, determine the cloud image measured solar radiation data based on the satellite cloud image data, and fine-tune the meteorological large model after solar radiation data training and inference based on the cloud image measured solar radiation data; A second generation module is used to generate second solar radiation data based on the fine-tuned meteorological large model; A fusion module, used for fusing the first solar radiation data and the second solar radiation data to generate fused data; The prediction module is used to build a photovoltaic short-term power prediction model based on Transformer, input the fusion data into the photovoltaic short-term power prediction model, and generate photovoltaic short-term power prediction results.

9. A server, characterized in that: The server comprises: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the photovoltaic short-term power prediction method as described in any one of claims 1-7.

10. A storage medium comprising computer executable instructions, wherein the computer executable instructions are used to execute the photovoltaic short-term power prediction method according to any one of claims 1 to 7 when executed by a computer processor.

Images