A weather forecast method, medium and program product based on AI and numerical assimilation
By combining deep learning AI model and numerical assimilation technology, using real-time observation data for high-frequency assimilation, the problem of insufficient data assimilation capabilities of AI weather forecast models is solved, and efficient and accurate weather forecasts and extreme meteorological incident responses are achieved worldwide .
Patent Information
- Application Number
- CN202410779909.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-17
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-06-17
AI Technical Summary
The existing AI weather forecast models lack complex data assimilation capabilities, limiting their independent operation and long-term, global response capabilities for extreme climate events, and lack real-time and diversity in processing observed data.
Combining deep learning AI models and numerical assimilation technology, through real-time updates and self-optimization, high-frequency assimilation is used to ensure the accuracy and flexibility of weather forecasts, and is suitable for weather forecasts and extreme meteorological incident responses on a global scale.
It significantly improves the speed and accuracy of weather forecasts, reduces the demand for high-performance computing resources, achieves real-time updates around the world and the rapid response of extreme meteorological events, and enhances the adaptability and stability to complex meteorological conditions.
Smart Images

Figure CN118606629B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of meteorological forecasting technology, and to the design and implementation of a weather forecasting system. Specifically, the present invention relates to a method, medium and program product for realizing real-time and automated global weather forecasting by integrating artificial intelligence (AI) technology with numerical assimilation technology. The technology combines the rapid forecasting capability of AI with numerical assimilation technology, and is suitable for global weather forecasting and rapid response to extreme meteorological events. The technology can be used not only for general weather forecasting and long-term climate pattern analysis, but also for industries such as aviation, ocean and agriculture. Background Art
[0002] The development of weather forecasting technology has evolved from simple empirical models to complex numerical weather prediction (NWP) models. Although existing NWP systems provide accurate weather forecasts worldwide, the efficiency and flexibility of these systems are limited by computing resources and data processing capabilities. In recent years, artificial intelligence (AI), especially deep learning technology, has shown great potential in the field of meteorology and has become a new direction for the development of weather forecasting technology. Emerging AI weather forecast models, such as Pangu-Weather, GraphCast, Fengwu, and Fuxi, not only have advantages in speed, but also in some cases, their prediction accuracy has exceeded that of traditional forecast models.
[0003] However, despite the excellent performance of AI-based weather forecast models in terms of processing speed and short-term prediction accuracy, such models usually rely on initial conditions provided by traditional NWP systems. Traditional NWP systems are supported by a powerful data assimilation (DA) framework, which continuously integrates real-time observation data and significantly improves prediction accuracy. Numerical assimilation technology is defined as existing data assimilation techniques applied to numerical forecast systems, such as three-dimensional variational (3DVAR), four-dimensional variational (4DVAR), hybrid variational (En-VAR), and ensemble data (EDA) assimilation methods. In contrast, most AI-based weather forecast models lack complex data assimilation capabilities, which limits their operational independence and usually only works as auxiliary tools rather than independent forecasting systems. In addition, AI weather forecast models also have deficiencies in processing the real-time and diversity of observational data, resulting in inferior performance to traditional NWP models in long-term forecasts and responses to extreme climate events on a global scale.
[0004] In response to this limitation, recent research has increasingly focused on developing new data assimilation systems tailored for AI weather forecast models. These efforts include adopting techniques such as traditional variational methods to provide AI weather forecast models with a continuous stream of observational data to optimize their initial conditions. In addition, some studies are exploring the development of new data assimilation frameworks that leverage AI's unique capabilities, such as strong adaptability, big data processing, and pattern recognition, to achieve more flexible and efficient assimilation strategies and further improve the performance and accuracy of weather forecast models.
[0005] However, building a data assimilation system for AI forecast models from scratch is a huge challenge, requiring extensive expertise, comprehensive data sets, and sophisticated calibration. Mature systems usually require years or even decades of continuous iteration and optimization to achieve optimal performance. In contrast, the AI data assimilation system is still in the exploratory stage, its theoretical basis and application technology are not yet mature, and there are many technical difficulties. Traditional numerical assimilation technology has good performance in dealing with high-dimensional nonlinear problems and has been widely verified in practical applications. Therefore, how to make full use of the advantages of traditional data assimilation systems and combine them with the rapid forecasting capabilities of AI models to build new weather forecasting methods that can be applied to global weather forecasts and rapid responses to extreme meteorological events is a key direction for improving the performance of weather forecasting systems, and it is also a technical problem that needs to be solved urgently in the current field of meteorological science and technology. Summary of the invention
[0006] (I) Purpose of the invention
[0007] In order to solve at least one of the above-mentioned and other technical problems in the prior art, the present invention provides a real-time updated weather forecast method, medium and program product integrating artificial intelligence (AI) and numerical assimilation technology. By integrating AI deep learning technology and numerical assimilation technology, it aims to provide efficient and accurate weather forecasts on a global scale, and significantly improves the speed and accuracy of weather forecasts, reduces the demand for high-performance computing resources, and is particularly suitable for rapid response to climate change and extreme weather events, providing support for meteorology, aviation, shipping, agriculture and disaster management.
[0008] (II) Technical solution
[0009] In order to achieve the purpose of the invention and solve the technical problems, the present invention adopts the following technical solutions:
[0010] The first invention object of the present invention is to provide a weather forecasting method based on AI and numerical assimilation, which is used to provide real-time updated weather forecasts and rapid responses to extreme meteorological events around the world, and the method comprises at least the following steps:
[0011] SS1. Use AI weather forecast models based on deep learning and continuous learning and self-optimization to collect and analyze global meteorological data and generate preliminary weather forecasts;
[0012] SS2. Perform background field interpolation on the preliminary weather forecast field to match the grid and vertical coordinates supported by the Weather Research and Forecasting model (WRF) to ensure spatial consistency of the forecast data;
[0013] SS3. Use numerical assimilation techniques to assimilate real-time observed meteorological data into interpolated preliminary weather forecast fields, and update the analysis fields to accurately reflect current atmospheric conditions;
[0014] SS4. Re-interpolate the assimilated analysis fields back to the native grid and vertical pressure layer of the AI weather forecast model to ensure that it can use the updated data for future weather forecasts;
[0015] SS5. Design a high-frequency update mechanism and repeat the above steps SS1-SS4 to continuously update the background field and analysis field based on the newly acquired real-time observation data, and automatically optimize and adjust the AI weather forecast model parameters according to the comparison between the latest analysis field and forecast results, so as to achieve real-time update of weather forecasts on a global scale and rapid response to extreme meteorological events.
[0016] Preferably, in the above step SS1, the AI weather forecast model is an AI model based on deep learning, including Pangu-Weather, GraphCast, Fengwu or Fuxi model, and is designed for sustainable learning and self-optimization, so as to adjust network parameters and optimize model structure through continuous training, adapt to the dynamic changes of climate conditions and improve the accuracy of weather forecasts.
[0017] Preferably, in the above step SS2, when the background field interpolation processing is performed on the preliminary weather forecast field generated by the AI weather forecast model, it includes horizontal interpolation to adjust the preliminary forecast field data to match the horizontal grid resolution supported by the WRF model, and vertical interpolation to convert the preliminary forecast field data to the vertical layer coordinate system used by the WRF model, so that the preliminary forecast field data after the background field interpolation is consistent with the WRF model in terms of grid and vertical coordinates, ensuring the spatial consistency of the forecast data.
[0018] Furthermore, when the preliminary forecast field data is adjusted to match the horizontal grid resolution supported by the WRF model, the horizontal interpolation adopts a bilinear interpolation algorithm in two-dimensional space, which includes the following sub-steps when implemented:
[0019] SS21. Determine the approximate location of the target interpolation point (x, y) in the source data grid;
[0020] SS22. Find the four nearest source data grid points surrounding the target interpolation point, whose coordinate values are (x1, y1), (x1, y2), (x2, y1), (x2, y1);
[0021] SS23. According to the distance ratio between the target interpolation point and the four adjacent source data grid points in the X direction and the Y direction, the distance weight coefficients w1, w2, w3, and w4 are calculated using linear interpolation:
[0022]
[0023]
[0024] SS24. Perform weighted summation of the data values of each adjacent source data grid point according to the distance weight coefficients w1, w2, w3, and w4, according to the following calculation formula:
[0025] f(x,y)=w1f(x1,y1)+w2f(x1,y2)+w3f(x2,y1)+w4f(x2,y2)
[0026] Get the bilinear interpolation result f(x,y) of the target interpolation point (x,y) in the source data grid.
[0027] Furthermore, when the preliminary forecast field data is converted to the vertical layer coordinate system used by the WRF model, the vertical interpolation adopts a logarithmic pressure coordinate interpolation algorithm that takes into account the nonlinear change of atmospheric pressure with height, which includes the following sub-steps when implemented:
[0028] SS25. For each horizontal grid point to be interpolated, determine its approximate position in the WRF data grid and the pressure value p0 of its corresponding target pressure layer;
[0029] SS26. Find two source pressure layers p1 and p2 that are closest to the target pressure layer p0 of the horizontal grid point to be interpolated, where the pressure value p0 is between p1 and p2;
[0030] SS27. Calculate the logarithmic pressure value of the horizontal grid point to be interpolated, and calculate the logarithmic pressure interpolation weight coefficients w′1 and w′2 of p0 relative to p1 and p2 according to the following mathematical formula:
[0031]
[0032] SS28. According to the logarithmic pressure interpolation weight coefficients w′1 and w′2, the field value f(p0) at the target pressure layer p0 is calculated to complete the vertical interpolation of the logarithmic pressure coordinates:
[0033] f(p0)=w′1·f(p1)+w′2·f(p2)
[0034] Among them, f(p1) and f(p2) are the field values at the source pressure layers p1 and p2 respectively.
[0035] Preferably, in the above step SS3, when the real-time observed meteorological data is assimilated into the interpolated preliminary weather forecast field using numerical assimilation technology, the following sub-steps are included:
[0036] SS31. Real-time data collection, automatically collecting multiple meteorological data sources around the world, including at least satellite data, radar data and ground observation station data;
[0037] SS32. Preprocess the collected real-time meteorological data, including at least data quality control, formatting and / or bias correction preprocessing, to ensure data accuracy and consistency;
[0038] SS33. Use numerical assimilation algorithms to optimally fuse preprocessed real-time observational meteorological data with interpolated preliminary weather forecast field data, and update the analysis field by minimizing the difference between the observational data and the model background field;
[0039] SS34. The updated analysis fields are verified and calibrated and output as new initial fields to ensure that they accurately reflect the current atmospheric conditions and provide high-quality initial fields for subsequent weather forecasts.
[0040] Preferably, in the above step SS4, when the assimilated analysis field is re-interpolated back to the native grid and vertical pressure layer of the AI weather forecast model, a horizontal interpolation and vertical interpolation algorithm similar to step SS2 is adopted, including the following sub-steps:
[0041] SS41. Determine the position of the assimilated analysis field in the native grid of the AI weather forecast model and the corresponding vertical pressure layer;
[0042] SS42. Use the horizontal interpolation algorithm to adjust the assimilated analysis field data from the WRF model grid to the native horizontal grid resolution used by the AI weather forecast model;
[0043] SS43. Use the vertical interpolation algorithm to convert the assimilated analysis field data from the WRF model vertical level to the native vertical layer coordinate system of the AI weather forecast model;
[0044] SS44. Verify the interpolated data to ensure that it matches the input format and requirements of the AI weather forecast model to provide accurate initial conditions for future weather predictions.
[0045] Preferably, in the above step SS5, the update process is automatically triggered at a preset time interval (such as every 6 hours) to quickly respond to global climate change and extreme weather events, ensure the real-time and accuracy of weather forecasts, and generate detailed weather forecasts including at least temperature, precipitation, wind speed and atmospheric pressure, supporting short-term to medium- and long-term forecast time spans.
[0046] The second invention object of the present invention is to provide a computer program product, including computer instructions, for executing the above-mentioned weather forecasting method based on AI and numerical assimilation.
[0047] The third inventive object of the present invention is to provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the weather forecasting method based on AI and numerical assimilation is implemented.
[0048] (III) Technical Effect
[0049] Compared with the prior art, the weather forecast method, medium and program product based on AI and numerical assimilation of the present invention have the following beneficial and significant technical effects:
[0050] (1) The real-time updated weather forecast method of the present invention combines artificial intelligence (AI) and numerical assimilation technology. The AI weather forecast model uses deep learning technology to quickly analyze and predict a large amount of meteorological data, while numerical technology accurately assimilates these predictions with real-time observation data, thereby optimizing the initial conditions and the prediction process, effectively overcoming their respective limitations and achieving an overall improvement in forecasting capabilities.
[0051] (2) The present invention significantly improves the speed and accuracy of weather forecasts. Traditional numerical weather prediction (NWP) systems usually require large-scale computing resources and long computing processes. The present invention reduces the demand for high-performance computing resources by utilizing the efficient data processing capabilities of AI models and the precise data assimilation functions of numerical systems, making it possible to perform complex weather forecasts on ordinary personal computers. The flexibility and scalability of this technology indicate its potential for widespread application in global meteorological monitoring, extreme climate event response, and long-term climate change analysis, providing individual users and small businesses with the possibility of achieving accurate weather forecasts.
[0052] (3) The present invention is flexible and extensible and is applicable to a variety of application scenarios. Through continuous learning and self-optimization, it can adapt to the dynamic changes of climate conditions and improve the adaptability and stability to complex meteorological conditions. Whether in global meteorological monitoring, extreme climate event response, or long-term climate change analysis, the real-time updated weather forecast method of the present invention can provide reliable forecast support.
[0053] (4) The present invention can autonomously adjust the forecast algorithm and optimize future forecasts by continuously collecting new meteorological data and feeding it back to the AI model, without relying on the traditional numerical forecast model for periodic updates. This completely autonomous AI weather forecast cycle technology makes weather forecasting more flexible and efficient, and further improves the timeliness and accuracy of the forecast.
[0054] (5) The present invention combines AI weather forecast models such as Pangu-Weather and Graphcast with numerical assimilation technology to optimize initial conditions and prediction processes, which not only improves the speed and accuracy of weather forecasts, but also enhances the adaptability and stability when dealing with complex meteorological conditions, providing a more accurate and efficient technical solution for global meteorological monitoring and rapid response to extreme climate events. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0056] Figure 1 It is an implementation flow chart of the weather forecasting method based on AI and numerical assimilation of the present invention;
[0057] Figure 2 It is a principle block diagram of the weather forecasting method based on AI and numerical assimilation of the present invention;
[0058] Figure 3 Shown is a schematic diagram of the comparison of the root mean square error of the background field between the assimilated numerical-AI experiment and the non-assimilated experiment, where (a) is the cross-experimental variation of wind speed, (b) is temperature, and (c) is specific humidity. DETAILED DESCRIPTION
[0059] In order to better understand the present invention, the content of the present invention is further explained in conjunction with the embodiments below. In the accompanying drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be understood as limitations on the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. The structure and technical solution of the present invention are further described in detail below in conjunction with the accompanying drawings, and an embodiment of the present invention is given.
[0060] The present invention provides a real-time updated weather forecast method, medium and program product integrating artificial intelligence (AI) and numerical assimilation technology. By integrating AI deep learning technology and numerical assimilation technology, it aims to provide efficient and accurate weather forecasts on a global scale and significantly improve the speed and accuracy of weather forecasts.
[0061] As a specific example, Figure 1 , 2 As shown, the weather forecasting method based on AI and numerical assimilation of the present invention is used to provide real-time updated weather forecasts and rapid response to extreme meteorological events on a global scale, and the implementation thereof mainly includes the following steps:
[0062] SS1.AI generates preliminary weather forecast fields
[0063] Using an AI weather forecast model based on deep learning and capable of continuous learning and self-optimization, it collects and analyzes global meteorological data to generate preliminary weather forecasts.
[0064] In some preferred instances, in this step, the AI weather forecast model preferably adopts an AI model based on deep learning, such as Pangu-Weather, GraphCast, Fengwu or Fuxi models and other AI weather forecast models, and is designed for sustainable learning and self-optimization, and adjusts network parameters and optimizes model structure through continuous training to adapt to dynamic changes in climate conditions and improve the accuracy of weather forecasts.
[0065] SS2. Background field interpolation processing
[0066] The preliminary weather forecast field generated in step SS1 is subjected to background field interpolation processing to match the grid and vertical coordinates supported by the Weather Research and Forecasting model (WRF) to ensure spatial consistency of the forecast data.
[0067] In some preferred instances, when performing background field interpolation processing on the preliminary weather forecast field generated by the AI weather forecast model, it includes horizontal interpolation to adjust the preliminary forecast field data to match the horizontal grid resolution supported by the WRF model, and vertical interpolation to convert the preliminary forecast field data to the vertical layer coordinate system used by the WRF model, so that the preliminary forecast field data after background field interpolation is consistent with the WRF model in terms of grid and vertical coordinates, ensuring the spatial consistency of the forecast data.
[0068] In a further preferred embodiment, when the preliminary forecast field data is adjusted to match the horizontal grid resolution supported by the WRF model, the horizontal interpolation adopts a bilinear interpolation algorithm performed in a two-dimensional space, which includes the following sub-steps when implemented:
[0069] SS21. Determine the approximate location of the target interpolation point (x, y) in the source data grid;
[0070] SS22. Find the four nearest source data grid points surrounding the target interpolation point, whose coordinate values are (x1, y1), (x1, y2), (x2, y1), (x2, y1);
[0071] SS23. According to the distance ratio between the target interpolation point and the four adjacent source data grid points in the X direction and the Y direction, the distance weight coefficients w1, w2, w3, and w4 are calculated using linear interpolation:
[0072]
[0073]
[0074] SS24. Perform weighted summation of the data values of each adjacent source data grid point according to the distance weight coefficients w1, w2, w3, and w4, according to the following calculation formula:
[0075] f(x,y)=w1f(x1,y1)+w2f(x1,y2)+w3f(x2,y1)+w4f(x2,y2)
[0076] Get the bilinear interpolation result f(x,y) of the target interpolation point (x,y) in the source data grid.
[0077] In a further preferred embodiment, when the preliminary forecast field data is converted to the vertical layer coordinate system used by the WRF model, the vertical interpolation adopts a logarithmic pressure coordinate interpolation algorithm that takes into account the nonlinear change of atmospheric pressure with height, which includes the following sub-steps when implemented:
[0078] SS25. For each horizontal grid point to be interpolated, determine its approximate position in the WRF data grid and the pressure value p0 of its corresponding target pressure layer;
[0079] SS26. Find two source pressure layers p1 and p2 that are closest to the target pressure layer p0 of the horizontal grid point to be interpolated, where the pressure value p0 is between p1 and p2;
[0080] SS27. Calculate the logarithmic pressure value of the horizontal grid point to be interpolated, and calculate the logarithmic pressure interpolation weight coefficients w′1 and w′2 of p0 relative to p1 and p2 according to the following mathematical formula:
[0081]
[0082] SS28. According to the logarithmic pressure interpolation weight coefficients w′1 and w′2, the field value f(p0) at the target pressure layer p0 is calculated to complete the vertical interpolation of the logarithmic pressure coordinates:
[0083] f(p0)=w′1·f(p1)+w′2·f(p2)
[0084] Among them, f(p1) and f(p2) are the field values at the source pressure layers p1 and p2 respectively.
[0085] SS3. Numerical Assimilation
[0086] Numerical assimilation techniques are used to assimilate real-time observed meteorological data into interpolated preliminary weather forecast fields, and the analysis fields are updated to accurately reflect current atmospheric conditions.
[0087] In some preferred embodiments, when using numerical assimilation technology to assimilate real-time observed meteorological data into the interpolated preliminary weather forecast field, the following sub-steps may be included, for example:
[0088] SS31. Real-time data collection, automatically collects a variety of meteorological data sources around the world, including at least satellite data, radar data and ground observation station data, such as satellite remote sensing data, radar wind measurement data, sounding data, aviation report data, ship observation data, ground automatic station data, etc., providing a rich source of information for weather forecast models.
[0089] SS32. Preprocess the collected real-time meteorological data, including at least data quality control, formatting and / or bias correction preprocessing, to ensure data accuracy and consistency. Quality control ensures data accuracy, formatting ensures data format consistency, and bias correction corrects systematic errors in observation data, thereby ensuring data consistency and reliability.
[0090] SS33. Use the numerical assimilation algorithm to optimally fuse the preprocessed real-time observational meteorological data with the preliminary weather forecast field data after interpolation processing, and update the analysis field by minimizing the difference between the observation data and the model background field.
[0091] SS34. The updated analysis fields are verified and calibrated and output as new initial fields to ensure that they accurately reflect current atmospheric conditions. The verification step includes comparison with independent observational data, and the calibration step includes adjusting model parameters to reduce forecast errors. Finally, the verified and calibrated analysis fields are output as new initial fields to provide high-quality initial conditions for subsequent weather forecasts.
[0092] As a more specific example, Figure 3The figure shows the comparison of the root mean square error of the assimilated GSI (Gridpoint Statistical Interpolation, a kind of numerical assimilation technology)-AI (taking Pangu as an example) test and the background field of the different assimilated tests, where (a) is the wind speed (unit: m / s), (b) is the temperature (unit: K) and (c) is the specific humidity (unit: 10 -3 g / kg). Background data are based on 6-hour forecasts from the Pangu model. Statistics are calculated globally and averaged over all vertical levels. Figure 3 The experimental results show that by assimilating GSI data, the background field root mean square error of the assimilation experiment is significantly lower than that of the non-assimilating experiment, indicating that the GSI data assimilation technology significantly improves the accuracy of the analysis field, thereby effectively improving the forecast accuracy of the AI weather forecast model.
[0093] SS4. Analytical field re-interpolation
[0094] The assimilated analysis fields are re-interpolated back into the native grid and vertical pressure layer of the AI weather forecast model to ensure that it can use the updated data for future weather predictions.
[0095] In some preferred embodiments, when the assimilated analysis field is re-interpolated back to the native grid and vertical pressure layer of the AI weather forecast model, a horizontal interpolation and vertical interpolation algorithm similar to step SS2 is adopted, including the following sub-steps:
[0096] SS41. Determine the position of the assimilated analysis field in the native grid of the AI weather forecast model and the corresponding vertical pressure layer;
[0097] SS42. Use the horizontal interpolation algorithm to adjust the assimilated analysis field data from the WRF model grid to the native horizontal grid resolution used by the AI weather forecast model;
[0098] SS43. Use the vertical interpolation algorithm to convert the assimilated analysis field data from the WRF model vertical level to the native vertical layer coordinate system of the AI weather forecast model;
[0099] SS44. Verify the interpolated data to ensure that it matches the input format and requirements of the AI weather forecast model to provide accurate initial conditions for future weather predictions.
[0100] Through this reverse interpolation process, the high-quality analysis field data obtained on the WRF model grid is converted back to the grid and coordinate system supported and applicable by the AI forecast model itself, thereby ensuring that the AI model can use the updated fine atmospheric initial field data for future weather forecasts and improve forecast accuracy.
[0101] SS5. High-frequency update mechanism design
[0102] A high-frequency update mechanism is designed to repeatedly execute steps SS1-SS4 above, continuously update the background field and analysis field based on the newly acquired real-time observation data, and automatically optimize and adjust the AI weather forecast model parameters according to the comparison between the latest analysis field and forecast results, so as to realize real-time update of weather forecasts on a global scale and rapid response to extreme meteorological events.
[0103] In some preferred embodiments, in the above step SS5, the update process is automatically triggered at a preset time interval (such as every 6 hours) to quickly respond to global climate change and extreme weather events, ensure the real-time and accuracy of weather forecasts, and generate detailed weather forecasts including at least temperature, precipitation, wind speed and atmospheric pressure, supporting short-term to medium- and long-term forecast time spans.
[0104] In summary, by integrating AI weather forecast models and numerical assimilation technology, the present invention can not only provide real-time and accurate global weather forecasts, but also its flexibility and scalability make it an important tool for facing global climate challenges. By combining AI weather forecast models such as Pangu-Weather, Graphcast, etc. with numerical assimilation technology to optimize initial conditions and prediction processes, AI weather forecast models use their deep learning capabilities to quickly analyze and predict large amounts of meteorological data, while numerical assimilation is responsible for assimilating these predictions with real-time observation data, improving the accuracy of the predictions, and also enhancing adaptability and stability when dealing with complex meteorological conditions.
[0105] In addition, if Figure 2 As shown, the AI weather forecast model of the present invention is different from the existing AI forecast that relies on NWP and assimilation systems to provide analysis as the initial field. Through the independent cycle of the AI model, it can self-learn and adapt to the changing meteorological conditions. By continuously collecting new data and feeding it back to the AI model, the system autonomously adjusts the forecast algorithm to optimize future predictions.
[0106] At the same time, by integrating AI weather forecasting with numerical assimilation, that is, by combining deep learning technology with meteorological data assimilation technology, the present invention sets a new standard in improving the accuracy and speed of weather forecasting, and effectively improves the overall performance of the forecasting system by utilizing the high-speed data processing capability and precise numerical assimilation function of the AI model. Moreover, the present invention realizes weather forecasting driven entirely by AI, facilitating its autonomous learning and adaptation, without relying on traditional numerical forecasting models for periodic updates.
[0107] In summary, the real-time updated weather forecast technology of the present invention combines artificial intelligence and traditional data assimilation technology, significantly improving the speed and accuracy of weather forecasts, reducing the demand for high-performance computing resources, and making it possible to perform complex weather forecasts on ordinary personal computers. The flexibility and scalability of this technology indicate its potential for widespread application in global meteorological monitoring, extreme climate event response, and long-term climate change analysis, providing individual users and small businesses with the possibility of achieving accurate weather forecasts.
[0108] Through the above embodiments, the purpose of the present invention is fully and effectively achieved. Those skilled in the art can understand that the present invention includes but is not limited to the contents described in the drawings and the above specific embodiments. Although the present invention has been described with respect to the most practical and preferred embodiments currently considered, it should be understood that the present invention is not limited to the disclosed embodiments, and any modification that does not deviate from the functional and structural principles of the present invention will be included in the scope of the claims.
Claims
1. A weather forecasting method based on AI and numerical assimilation, characterized in that: The weather forecast method comprises at least the following steps when implemented: SS1. Use an AI weather forecast model based on deep learning and capable of continuous learning and self-optimization to collect and analyze global meteorological data and generate preliminary weather forecasts; SS2. Perform background field interpolation on the preliminary weather forecast field to match the grid and vertical coordinates supported by the WRF model to ensure spatial consistency of the forecast data; SS3. Use numerical assimilation techniques to assimilate real-time observed meteorological data into interpolated preliminary weather forecast fields and update the analysis fields to accurately reflect current atmospheric conditions; SS4. Re-interpolate the assimilated analysis fields back to the native grid and vertical pressure layer of the AI weather forecast model to ensure that it can use the updated data for future weather forecasts; SS5. Design a high-frequency update mechanism and repeat steps SS1-SS4 above to continuously update the background field and analysis field based on the newly acquired real-time observation data, and automatically optimize and adjust the AI weather forecast model parameters according to the comparison between the latest analysis field and forecast results, so as to achieve real-time update of weather forecasts worldwide and rapid response to extreme meteorological events.
2. The weather forecasting method based on AI and numerical assimilation according to claim 1, characterized in that: In the above step SS1, the AI weather forecast model is an AI model based on deep learning, including Pangu-Weather, GraphCast, Fengwu or Fuxi model, and is designed for sustainable learning and self-optimization, so as to adjust network parameters and optimize model structure through continuous training, adapt to the dynamic changes of climate conditions and improve the accuracy of weather forecasts.
3. The weather forecasting method based on AI and numerical assimilation according to claim 1, characterized in that: In the above step SS2, when the background field interpolation processing is performed on the preliminary weather forecast field generated by the AI weather forecast model, it includes horizontal interpolation to adjust the preliminary forecast field data to match the horizontal grid resolution supported by the WRF model, and vertical interpolation to convert the preliminary forecast field data to the vertical layer coordinate system used by the WRF model, so that the preliminary forecast field data after the background field interpolation is consistent with the WRF model in terms of grid and vertical coordinates, ensuring the spatial consistency of the forecast data.
4. The weather forecasting method based on AI and numerical assimilation according to claim 3, characterized in that: When the preliminary forecast field data is adjusted to match the horizontal grid resolution supported by the WRF model, the horizontal interpolation adopts a bilinear interpolation algorithm in two-dimensional space, including: SS21. Determine the target interpolation point ( x , y )’s approximate location in the source data grid; SS22. Find the four nearest source data grid points surrounding the target interpolation point, whose coordinate values are ( x 1, y 1) ( x 1, y 2) ( x 2, y 1) ( x 2, y 2); SS23. Calculate the distance weight coefficient using linear interpolation based on the distance ratio between the target interpolation point and the four adjacent source data grid points in the X and Y directions w 1. w 2. w 3. w 4: SS24. The data values of each adjacent source data grid point are calculated according to the distance weight coefficient. w 1. w 2. w 3. w 4 Perform weighted summation according to the following calculation formula: Get the target interpolation point ( x , y ) results from bilinear interpolation in the source data grid f ( x , y ).
5. The weather forecasting method based on AI and numerical assimilation according to claim 4, characterized in that: When converting the preliminary forecast field data to the vertical layer coordinate system used by the WRF model, a logarithmic pressure coordinate interpolation algorithm is used that takes into account the nonlinear variation of atmospheric pressure with height, including: SS25. For each horizontal grid point to be interpolated, determine its approximate position in the WRF data grid and the pressure value of its corresponding target pressure layer p 0; SS26. Find the target pressure layer closest to the horizontal grid point to be interpolated p 0 two source pressure layers p 1 and p 2, where the pressure value p 0 is located in p 1 and p 2 between; SS27. Calculate the logarithmic pressure value of the horizontal grid point to be interpolated according to the following mathematical formula p 0 relative to p 1 and p 2 logarithmic pressure interpolation weight coefficient : SS28. Interpolation weight coefficients based on logarithmic pressure , calculate the target pressure layer p Field value at 0 f ( p 0), complete the vertical interpolation of the logarithmic pressure coordinates: in, f ( p 1) f ( p 2) Source pressure layer p 1. p Field value at 2.
6. The weather forecasting method based on AI and numerical assimilation according to claim 1, characterized in that: In the above step SS3, when the real-time observed meteorological data is assimilated into the interpolated preliminary weather forecast field using numerical assimilation technology, the following sub-steps are included: SS31. Real-time data collection, automatically collecting multiple meteorological data sources around the world, including at least satellite data, radar data and ground observation station data; SS32. Preprocess the collected real-time meteorological data, including at least data quality control, formatting and / or bias correction preprocessing, to ensure data accuracy and consistency; SS33. Use numerical assimilation algorithms to optimally fuse preprocessed real-time observational meteorological data with interpolated preliminary weather forecast field data, and update the analysis field by minimizing the difference between the observational data and the model background field; SS34. The updated analysis fields are verified and calibrated and output as new initial fields to ensure that they accurately reflect the current atmospheric conditions and provide high-quality initial fields for subsequent weather forecasts.
7. The weather forecasting method based on AI and numerical assimilation according to claim 1, characterized in that: In the above step SS4, when the assimilated analysis field is re-interpolated back to the native grid and vertical pressure layer of the AI weather forecast model, the following sub-steps are included: SS41. Determine the position of the assimilated analysis field in the native grid of the AI weather forecast model and the corresponding vertical pressure layer; SS42. Use a horizontal interpolation algorithm to adjust the assimilated analysis field data from the WRF model grid to the native horizontal grid resolution used by the AI weather forecast model; SS43. Use the vertical interpolation algorithm to convert the assimilated analysis field data from the WRF model vertical level to the native vertical layer coordinate system of the AI weather forecast model; SS44. Validate the interpolated data to ensure that it matches the input format and requirements of the AI weather forecast model to provide accurate initial conditions for future weather predictions.
8. The weather forecasting method based on AI and numerical assimilation according to claim 1, characterized in that: In the above step SS5, the update process is automatically triggered at preset time intervals to quickly respond to global climate change and extreme weather events, ensure the real-time and accuracy of weather forecasts, and generate detailed weather forecasts including at least temperature, precipitation, wind speed and atmospheric pressure, supporting short-term to medium- and long-term forecast time spans.
9. A computer program product, characterized in that It includes computer instructions for executing the weather forecasting method based on AI and numerical assimilation as described in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the weather forecasting method based on AI and numerical assimilation as described in any one of claims 1 to 8 is implemented.
