Regional numerical mode forecasting system utilizing AI meteorological large model observation guidance

Abstract

The invention discloses a regional numerical model forecasting system utilizing observation guidance of an AI meteorological big model, and relates to the field of data processing systems.An AI forecasting module in the system inputs global weather forecasting mode initial field data into the AI meteorological big model and outputs AI weather forecasting mode data; and then the observation data fusion module performs grid fusion on the global weather forecast mode initial field data and the AI weather forecast mode data by using an observation guidance technology to obtain fused observation data, and then the fused observation data is used for driving a regional numerical forecast model to obtain a regional numerical mode forecast result. Under the limitation that an output result of an AI meteorological large model cannot be directly utilized to drive a regional numerical forecasting model, effective information in the AI meteorological large model is introduced through an observation guiding technology, so that better initial field and boundary conditions of the regional numerical forecasting model can be obtained, and a regional numerical mode forecasting result with higher forecasting quality can be obtained.

Description

Technical Field

[0001] The present application relates to the field of data processing systems, and in particular to a regional numerical model forecasting system guided by observations of a large AI meteorological model. Background Art

[0002] Numerical model forecasting is an important means of conducting climate forecasting and an effective tool for forecasting severe weather. At present, global numerical prediction models and regional numerical prediction models are mainly used for numerical model forecasting. These two types of numerical prediction models have their own characteristics: the global numerical prediction model has a wider range of application, and common global numerical prediction models include ECMWF, GFS, etc. The regional numerical prediction model is a meteorological forecast model designed for a specific region. It has certain regional limitations, but compared with the global numerical prediction model, it has a higher spatial resolution and can provide more detailed terrain and surface feature information, thereby improving the prediction ability of local meteorological elements. Common regional numerical prediction models include WRF, COSMO, RAMS, etc.

[0003] Most of the existing traditional numerical model forecasts use the output of the global numerical prediction model as the initial field and boundary conditions of the regional numerical prediction model, and use ensemble forecasts through initial field perturbations and other methods to improve the accuracy and reliability of the forecast. However, due to the low accuracy and precision of the data of the global numerical prediction model, the accuracy and precision of the results obtained by this numerical model forecast are still not ideal, especially in complex weather environments and extreme weather event scenarios. The forecast performance cannot meet the requirements. Summary of the invention

[0004] In response to the above problems and technical requirements, this application proposes a regional numerical model forecasting system guided by AI meteorological large model observations. The technical solution of this application is as follows:

[0005] A regional numerical model forecasting system using AI meteorological large model observation guidance, the regional numerical model forecasting system comprising:

[0006] A global data acquisition module is used to obtain the initial field data of the global weather forecast model output by the global numerical forecast model;

[0007] AI forecast module, used to input the acquired global weather forecast model initial field data into the AI ​​meteorological big model and output AI weather forecast model data;

[0008] The observation data fusion module is used to grid-fuse the initial field data of the global weather forecast model and the AI ​​weather forecast model data using the observation guidance technology to obtain fused observation data;

[0009] The regional forecast module is used to drive the regional numerical forecast model using fused observation data and obtain regional numerical model forecast results.

[0010] Its further technical solution is that the AI ​​forecast module is used to input the acquired global weather forecast model initial field data into the AI ​​meteorological large model, and then pre-process the three-dimensional high-altitude data output by the AI ​​meteorological large model to obtain and output AI weather forecast model data in the target data format that conforms to the observation guidance technology.

[0011] Its further technical solution is an AI forecast module, which is used to treat each grid point of the three-dimensional high-altitude data output by the AI ​​meteorological model as an observation profile, and obtain AI weather forecast model data containing one-dimensional observation profiles of Lat*Lon by data preprocessing; wherein Lat is the number of latitude lines, Lon is the number of longitude lines, and the intersection of each latitude line and each latitude line is a grid point.

[0012] Its further technical solution is that the regional numerical model forecasting system includes:

[0013] The global data preprocessing module is used to convert the data format of the initial field data of the global weather forecast model according to the data processing format of the regional numerical forecast model, obtain the initial field data of the global weather forecast model whose data format matches the regional numerical forecast model and output it to the observation data fusion module.

[0014] Its further technical solution is that the AI ​​forecast module is used to input the acquired global weather forecast model initial field data into multiple different AI meteorological models respectively, and output multiple groups of different AI weather forecast model data to the observation data fusion module.

[0015] Its further technical solution is that the observation data fusion module is used to use the observation guidance technology to grid-fuse the initial field data of the global weather forecast model and each group of AI weather forecast model data, obtain multiple groups of fused observation data and output them to the regional forecast module;

[0016] The regional forecast module is used to drive the regional numerical forecast model using multiple sets of fused observation data and obtain regional numerical model forecast results.

[0017] Its further technical solution is a regional forecast module, which is used to use each set of fused observation data to drive the regional numerical forecast model and obtain corresponding sub-forecast results, and integrate and fuse multiple sub-forecast results to obtain regional numerical model forecast results.

[0018] A further technical solution is that the method for integrating and fusing multiple sub-forecast results by the regional forecast module includes at least one of simple averaging, weighted averaging, Bayesian model averaging, ensemble Kalman filtering and machine learning methods.

[0019] Its further technical solution is an observation data fusion module, which is used to use observation guidance technology to grid fuse the initial field data of the global weather forecast model and all multiple groups of AI weather forecast model data to obtain fused observation data and output it to the regional forecast module.

[0020] A further technical solution is that the AI ​​meteorological big model used includes at least one of Pangu, Fuxi, Fengwu and graphcast.

[0021] The beneficial technical effects of this application are:

[0022] The present application discloses a regional numerical model forecasting system guided by observation of an AI meteorological large model. The AI ​​forecasting module in the system uses the AI ​​meteorological large model to output AI weather forecasting model data according to the initial field data of the global weather forecasting model. Then, the observation data fusion module uses the observation guidance technology to grid-fuse the initial field data of the global weather forecasting model and the AI ​​weather forecasting model data to obtain fused observation data, which is used to drive the regional numerical forecasting model to obtain the regional numerical model forecasting results. Under the limitation that the output results of the AI ​​meteorological large model cannot be directly used to drive the regional numerical forecasting model, the output results of the AI ​​meteorological large model are introduced on the basis of the traditional method of driving the regional numerical forecasting model based on the initial field data of the global weather forecasting model through the observation guidance technology, so that the regional numerical forecasting model can make full use of the effective information in the output results of the AI ​​meteorological large model. Thanks to the high accuracy of the output results of the AI ​​meteorological large model, better initial fields and boundary conditions of the regional numerical forecasting model can be obtained, thereby improving the forecast quality and obtaining regional numerical model forecasting results with higher forecast quality.

[0023] The AI ​​forecasting module in this system also outputs multiple sets of AI weather forecast model data through multiple different AI meteorological models, and then obtains multiple sets of fused observation data to drive the regional numerical forecasting models respectively, and then obtains the regional numerical model forecast results based on the idea of ​​integrated forecasting. This approach can give full play to the respective advantages of different AI meteorological models to further improve the forecast quality, especially in complex weather and extreme environments, and can provide more robust results. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the system structure of a regional numerical model forecasting system according to an embodiment of the present application.

[0025] Figure 2It is a schematic diagram of the system structure of a regional numerical model forecasting system according to another embodiment of the present application. DETAILED DESCRIPTION

[0026] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.

[0027] With the development of deep learning and machine learning algorithms, various AI meteorological models have gradually been developed and put into use. AI meteorological models are meteorological prediction models developed using artificial intelligence technology, which are conducive to improving and enhancing the capabilities of traditional meteorological prediction models. These AI meteorological models can provide more accurate and detailed prediction results than global numerical forecast models by processing and analyzing large amounts of meteorological data. In addition, by combining artificial intelligence technology and meteorological science, AI meteorological models not only improve forecast accuracy and efficiency, but also provide important technical support for responding to climate change and extreme weather events.

[0028] Since the AI ​​meteorological big model can provide more accurate forecast results compared to the global numerical forecast model, introducing the AI ​​meteorological big model into the regional numerical model forecast process is of great significance for improving forecast accuracy and efficiency. However, due to some of the characteristics of the AI ​​meteorological big model itself, the forecast results output by the AI ​​meteorological big model often lack some key elements, including vertical wind, ground humidity, radiation-related variables, etc. This inherent defect makes it impossible to directly use the forecast results of the AI ​​meteorological big model to drive the regional numerical forecast model. Therefore, this application discloses a regional numerical model forecast system guided by observations of the AI ​​meteorological big model, which uses observation guidance technology to introduce the data of the AI ​​meteorological big model into the regional numerical forecast model. Please refer to Figure 1 As shown in the system structure diagram, the regional numerical model forecasting system includes the following modules:

[0029] (1) A global data acquisition module, which is used to obtain the initial field data of the global weather forecast model output by the global numerical forecast model. In actual implementation, the global data acquisition module directly obtains the initial field data of the global weather forecast model from the public platform, or obtains the initial field data of the global weather forecast model by running and driving the global numerical forecast model by itself. The global numerical forecast model referred to in this application includes existing global numerical forecast models such as ECMWF and GFS.

[0030] (2) an AI forecast module, which is used to input the acquired global weather forecast model initial field data into the AI ​​meteorological big model and output the AI ​​weather forecast model data. The AI ​​meteorological big model used by the AI ​​forecast module includes various meteorological prediction models developed using artificial intelligence technology. In one embodiment, the existing AI meteorological big model can be directly used, including at least one of Pangu, Fuxi, Fengwu, and graphcast.

[0031] The AI ​​weather forecast model data output by the AI ​​forecast module needs to be provided to the observation data fusion module for observation guidance, so the output AI weather forecast model data conforms to the target data format of the observation guidance technology, so that the observation guidance technology can directly read and use the AI ​​weather forecast model data. However, the results directly output by the general AI meteorological large model may not conform to the target data format of the observation guidance technology. Therefore, after the AI ​​forecast module inputs the acquired global weather forecast model initial field data into the AI ​​meteorological large model, the results output by the AI ​​meteorological large model are preprocessed to obtain AI weather forecast model data in the target data format of the observation guidance technology and output.

[0032] The results output by the AI ​​meteorological large model usually include two-dimensional ground data (including ground temperature, wind speed and direction, precipitation, etc.) and three-dimensional high-altitude data (including temperature, humidity, meridional wind, zonal wind, potential height, etc. at each altitude layer in the high altitude). Since the two-dimensional ground data is relatively smooth, in order to have a better observation and guidance effect, the three-dimensional high-altitude data output by the AI ​​meteorological large model is used in this embodiment, and the three-dimensional high-altitude data is preprocessed to obtain AI weather forecast model data. The method of data preprocessing includes: taking each grid point of the three-dimensional high-altitude data output by the AI ​​meteorological large model as an observation profile, and obtaining AI weather forecast model data containing Lat*Lon one-dimensional observation profiles by data preprocessing. Among them, Lat is the number of latitude lines, Lon is the number of longitude lines, and the intersection of each latitude line and each latitude line is a grid point.

[0033] (3) Observation data fusion module, which is used to grid-fuse the initial field data of the global weather forecast model and the AI ​​weather forecast model data using observation guidance technology to obtain fused observation data. Observation guidance technology (Nudging) is a kind of data assimilation, which can be achieved by running software such as OBSGRID.

[0034] Since the fused observation data is to be output to the regional forecast module to drive the regional numerical forecast model, in order to match the data processing format of the regional numerical forecast model so that the regional numerical forecast model can read and write, the regional numerical model forecast system also includes a global data preprocessing module, please refer to Figure 2The global weather forecast model initial field data output by the global data acquisition module is provided to the global data preprocessing module. The global data preprocessing module is used to convert the data format of the global weather forecast model initial field data according to the data processing format of the regional numerical forecast model, obtain the global weather forecast model initial field data whose data format matches the regional numerical forecast model and output it to the observation data fusion module. The data processing formats of different regional numerical forecast models are different, and the data format conversion is performed according to the regional numerical forecast model actually used.

[0035] (4) Regional forecast module, which is used to drive the regional numerical forecast model using fused observation data and obtain regional numerical model forecast results.

[0036] Since the fused observation data is obtained by grid fusion of the initial field data of the global weather forecast model and the AI ​​weather forecast model data using observation-guided technology, the method of the present application introduces the three-dimensional high-altitude data information of the AI ​​meteorological large model on the basis of the traditional method of driving the regional numerical prediction model based on the initial field data of the global weather forecast model. Under the limitation that the output results of the AI ​​meteorological large model cannot be directly used to drive the regional numerical prediction model, the regional numerical prediction model can still make full use of the effective information in the output results of the AI ​​meteorological large model. Compared with the traditional method of only using the output results of the initial field data of the global weather forecast model, better initial fields and boundary conditions of the regional numerical prediction model can be obtained, thereby improving the forecast quality and obtaining regional numerical model forecast results with higher forecast quality.

[0037] In addition, considering that different AI meteorological models often have different advantages in different meteorological elements, in another embodiment, the output results of multiple AI meteorological models are introduced for integrated forecasting to further improve the accuracy and reliability of the forecast. Figure 2 The system information flow diagram is shown.

[0038] In this embodiment, the AI ​​forecast module has multiple different AI meteorological models built in. The AI ​​forecast module will input the acquired global weather forecast model initial field data into multiple different AI meteorological models respectively, and output multiple groups of different AI weather forecast model data to the observation data fusion module. Figure 2 Taking the AI ​​forecast module with N built-in AI meteorological models as AI meteorological big model 1 to AI meteorological big model N as an example, after data preprocessing the output results of each AI meteorological big model, a corresponding set of AI weather forecast model data is obtained, and AI weather forecast model data 1 to AI weather forecast model data N are output respectively, and the integer parameter N≥2.

[0039] One approach is that the observation data fusion module uses observation guidance technology to grid fuse the initial field data of the global weather forecast model and all multiple groups of AI weather forecast model data to obtain a set of fused observation data and output it to the regional forecast module.

[0040] However, due to the different characteristics of the AI ​​weather forecast model data output by different AI meteorological models, the above-mentioned method of uniformly observing and guiding all multiple groups of AI weather forecast model data to achieve grid fusion is not ideal. Therefore, in order to improve the forecast quality, in another embodiment, the observation data fusion module uses observation guidance technology to grid fuse the global weather forecast model initial field data and each group of AI weather forecast model data, obtain multiple groups of fused observation data and output them to the regional forecast module. For example, Figure 2 In the example, the observation data fusion module uses the observation guidance technology to grid fuse the initial field data of the global weather forecast model and the AI ​​weather forecast model data 1 and outputs the fused observation data 1. The observation data fusion module uses the observation guidance technology to grid fuse the initial field data of the global weather forecast model and the AI ​​weather forecast model data 2 and outputs the fused observation data 2. The observation data fusion module uses the observation guidance technology to grid fuse the initial field data of the global weather forecast model and the AI ​​weather forecast model data N and outputs the fused observation data N, thereby outputting a total of N groups of fused observation data to the regional forecast module.

[0041] The regional forecast module uses multiple groups of fused observation data to drive the regional numerical forecast model and obtain the regional numerical model forecast results, including: using each group of fused observation data to drive the regional numerical forecast model and obtain the corresponding sub-forecast results, and then integrating and fusing multiple sub-forecast results to obtain the regional numerical model forecast results. It should be noted that Figure 2 The multiple regional numerical forecast models in the regional forecast module are the same regional numerical forecast model. The method for integrating and fusing the multiple sub-forecast results in the regional forecast module includes at least one of simple averaging, weighted averaging, Bayesian model averaging, ensemble Kalman filtering and machine learning methods.

[0042] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.

Claims

1. A regional numerical model forecasting system guided by AI meteorological large model observation, characterized in that: The regional numerical model forecasting system comprises: A global data acquisition module is used to obtain the initial field data of the global weather forecast model output by the global numerical forecast model; An AI forecast module, used to input the acquired initial field data of the global weather forecast model into the AI ​​meteorological big model and output AI weather forecast model data; The observation data fusion module is used to grid-fuse the initial field data of the global weather forecast model and the AI ​​weather forecast model data using the observation guidance technology to obtain fused observation data; The regional forecast module is used to drive the regional numerical forecast model using fused observation data and obtain regional numerical model forecast results.

2. The regional numerical model forecasting system according to claim 1, characterized in that: The AI ​​forecast module is used to input the acquired initial field data of the global weather forecast model into the AI ​​meteorological large model, and then preprocess the three-dimensional high-altitude data output by the AI ​​meteorological large model to obtain and output AI weather forecast model data in a target data format that conforms to the observation guidance technology.

3. The regional numerical model forecasting system according to claim 2, characterized in that: The AI ​​forecast module is used to treat each grid point of the three-dimensional high-altitude data output by the AI ​​meteorological large model as an observation profile, and obtain AI weather forecast model data containing Lat*Lon one-dimensional observation profiles through data preprocessing; wherein Lat is the number of latitude lines, Lon is the number of longitude lines, and the intersection of each latitude line and each latitude line is a grid point.

4. The regional numerical model forecasting system according to claim 1, characterized in that: The regional numerical model forecasting system comprises: The global data preprocessing module is used to convert the data format of the global weather forecast model initial field data according to the data processing format of the regional numerical forecast model, obtain the global weather forecast model initial field data whose data format matches the regional numerical forecast model and output it to the observation data fusion module.

5. The regional numerical model forecasting system according to claim 1, characterized in that: The AI ​​forecast module is used to input the acquired initial field data of the global weather forecast model into multiple different AI meteorological models respectively, and output multiple groups of different AI weather forecast model data to the observation data fusion module.

6. The regional numerical model forecasting system according to claim 5, characterized in that: The observation data fusion module is used to perform grid fusion on the initial field data of the global weather forecast model and each group of AI weather forecast model data by using observation guidance technology to obtain multiple groups of fused observation data and output them to the regional forecast module; The regional forecast module is used to drive the regional numerical forecast model using multiple groups of fused observation data and obtain regional numerical model forecast results.

7. The regional numerical model forecasting system according to claim 6, characterized in that: The regional forecast module is used to use each group of fused observation data to drive the regional numerical forecast model and obtain corresponding sub-forecast results, and integrate and fuse multiple sub-forecast results to obtain the regional numerical model forecast result.

8. The regional numerical model forecasting system according to claim 7, characterized in that: The method for the regional forecast module to integrate and fuse multiple sub-forecast results includes: at least one of simple averaging, weighted averaging, Bayesian model averaging, ensemble Kalman filtering and machine learning methods.

9. The regional numerical model forecasting system according to claim 5, characterized in that: The observation data fusion module is used to use observation guidance technology to grid fuse the initial field data of the global weather forecast model and all multiple groups of AI weather forecast model data to obtain fused observation data and output it to the regional forecast module.

10. The regional numerical model forecasting system according to claim 1, characterized in that: The AI ​​meteorological models used include at least one of Pangu, Fuxi, Fengwu, and graphcast.

Images