Typhoon intensity prediction method based on AI initial field driven NWP area mode and fused with vortex power initialization, medium and computer program

Abstract

The invention discloses a typhoon intensity prediction method based on an AI initial field driven NWP area mode and fused with vortex power initialization, a medium and a computer program, belongs to the technical field of artificial intelligence and area numerical simulation, and aims to improve the intensity prediction precision and the structure reduction capability of a fast enhanced typhoon. The method comprises the following steps: firstly, constructing an AI weather model based on deep learning, generating a continuous initial field and a boundary field covering multiple meteorological elements, and driving an area NWP mode; secondly, a vortex power initialization process is introduced before report starting, and the consistency and strength accuracy of a typhoon axial symmetry structure in a simulation initial field are enhanced; then, integral simulation is carried out based on an atmosphere-ocean bidirectional coupling mode, and dynamic optimization of a local structure is realized in combination with high-frequency multi-source observation data assimilation; and finally, an AI-NWP residual feedback mechanism is constructed, and closed-loop correction and model fine tuning of prediction errors are realized. The method is suitable for high-precision prediction of typhoon intensity and has good engineering practicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence and regional numerical simulation, and relates to high-precision modeling and prediction of the typhoon intensity evolution process. Specifically, it is a typhoon intensity prediction method, medium and computer program based on AI initial field driving NWP regional model and integrating vortex dynamic initialization, which can improve the intensity prediction accuracy and structure restoration ability in the rapid intensification stage of strong typhoons. Background Art

[0002] As a typical tropical cyclone system, the intensity change process of typhoons is particularly complex and is affected by the combined action of various atmospheric, oceanic and topographic factors. Especially in the rapid intensification (RI) stage, it shows extremely high nonlinearity and uncertainty. Accurate prediction of typhoon intensity is not only of great significance for disaster prevention and reduction, but also one of the key technical problems faced by current numerical weather prediction (NWP). Traditional global NWP models usually generate initial fields using static data assimilation methods. Limited by resolution, dynamic framework and parameterization schemes, it is difficult to finely depict key features such as typhoon eyewall structure, low-level wind field intensity and mid-level warm core structure, resulting in systematic underestimation in typhoon intensity simulation by traditional NWP, and the problem of weak prediction results is particularly prominent.

[0003] In recent years, the meteorological prediction methods driven by Artificial Intelligence (AI) have developed rapidly, and some AI global weather prediction models have shown excellent performance in the prediction of medium- and short-term weather elements. However, due to the regression-based structure of AI models, there are problems of intensity smoothing and structure diffusion, resulting in significant biases in the intensity prediction of severe weather systems such as typhoons, especially in the 10m near-surface wind speed, minimum central pressure, and vortex axisymmetric structure. In the prior art, Chinese Patent CN119202880A discloses a method for quickly identifying long-distance heavy rain in typhoons based on an AI meteorological model, which constructs a sensitivity experiment by removing the typhoon core vortex to analyze the long-distance precipitation effect, but its technical solution does not solve the systematic underestimation problem of AI models in typhoon intensity prediction. CN108983320A trains a deep neural network through a numerical weather prediction model and measured data, and combines a mesoscale meteorological numerical model for downscaling calculation to predict extreme wind speeds. However, this method focuses on the prediction of extreme wind speeds, and the coupling method of the AI model and the numerical model is relatively simple, making it difficult to fully utilize the advantages of the AI model in large-scale circulation prediction. CN119106249A uses an AI model to generate simulated typhoon samples to evaluate the impact of climate change on typhoon extreme winds, focusing on long-term climate impact assessment rather than short-term or medium-term typhoon intensity prediction, and it is difficult to fully utilize the advantages of the AI model in processing complex meteorological data.

[0004] To make up for the deficiencies of global models in local weather simulation, regional high-resolution numerical weather prediction models such as the WRF model (Weather Research and Forecasting Model) are widely used for high-resolution reproduction and intensity evolution simulation of typhoon processes. By setting nested grids, local physical parameter schemes, and initialization processing of vortex structures, the WRF model can better capture and reproduce the internal dynamic structure of typhoons. However, the simulation results of regional numerical models highly depend on the quality of their initial and boundary conditions. The traditional initial fields provided by global NWP still have deficiencies in simulating typhoon intensity and are difficult to capture the true three-dimensional structure of the atmosphere. In the prior art, although AI weather prediction has advantages in large-scale circulation prediction and the WRF model performs well in fine-scale process simulation, there is currently no mature technical solution on how to effectively integrate the advantages of both to improve typhoon intensity prediction accuracy. In particular, how to use the prediction results of the AI model as the driving conditions for regional models and how to optimize the vortex initialization technology according to the characteristics of AI prediction results are still technical blanks.

[0005] In summary, while existing AI weather models possess strong large-scale circulation forecasting capabilities, they are significantly weak in depicting typhoon intensity. While the traditional WRF model offers high resolution and structural simulation capabilities, its simulation results still exhibit significant errors due to limitations in initial conditions and the quality of boundary fields. Therefore, effectively integrating AI weather forecasting technology with regional numerical weather models, and combining it with vortex dynamics initialization techniques to construct a high-precision typhoon intensity forecast method, is a pressing technical challenge. Summary of the Invention

[0006] (1) Purpose of the invention In response to the above-mentioned defects and deficiencies in the prior art, the present invention aims to provide a typhoon intensity prediction method, medium, and computer program based on AI initial field-driven NWP regional model and integrated vortex dynamic initialization. By using AI meteorological forecast models to generate high-quality initial fields and boundary conditions, driving regional numerical weather forecast models and superimposing vortex dynamic initialization processes, the dynamic consistency and intensity fitting capabilities of the initial vortex structure are enhanced, achieving high-precision simulation and structural reproduction of the intensity evolution process of strong typhoons, especially those in the rapid intensification stage. This method fully integrates the advantages of AI models in capturing mesoscale circulations and the capabilities of high-resolution NWP models in analyzing local structures, significantly improving the regional model's simulation accuracy for key parameters such as typhoon maximum wind speed, central pressure, and eyewall structure, thereby effectively improving the systematic underestimation problem of existing forecast systems in typhoon intensity forecasting.

[0007] (2) Technical solution In order to achieve the purpose of the invention and solve the technical problems, the present invention adopts the following technical solutions: The first object of the present invention is to provide a typhoon intensity prediction method based on an AI initial field-driven NWP regional model and integrated with vortex dynamic initialization, which is used to improve the intensity prediction accuracy and structure restoration capability of tropical cyclone systems, especially typhoons in the rapid intensification stage. The method comprises the following steps: S100. Build an AI weather model and generate initial and boundary fields: An AI weather model is constructed and trained using a deep learning architecture and historical reanalysis data. The AI weather model includes multiple deep learning subnetworks to support different forecast timeframes, uses an Earth spherical sensing structure for input encoding, and outputs multi-timeframe continuous forecast results covering high-altitude meteorological elements, which serve as the initial and boundary fields for subsequent NWP regional models. S200. Constructing the NWP regional model and performing vortex dynamics initialization processing: Construct a multi-layer nested structure NWP regional model computational domain, where the middle domain and the inner domain are set as adaptive tracking moving grids based on the typhoon vortex center position; and in the preprocessing stage before the start of forecasting, introduce a vortex dynamic initialization process, extract the typhoon axisymmetric vortex structure of the AI initial field, perform spectral perturbation enhancement and position repositioning, and construct a dynamic initial field with intensity fitting ability and dynamic consistency; S300. Start the AI-driven atmosphere-ocean two-way coupled model integration simulation: Based on the dynamic initial field after vortex initialization, start the NWP regional atmospheric model, and simultaneously start the ocean model for coupled integration simulation. In each integration time step, the ocean model real-time updates key underlying surface variables including at least sea surface temperature, latent heat flux, and sensible heat flux, and inputs them as feedback into the atmospheric model to dynamically correct the sea surface boundary conditions along the typhoon movement path; S400. Real-time assimilation correction of multi-source observation data to optimize the simulation process: During the execution of the atmosphere-ocean regional coupled model integration process, introduce multi-source near-surface observation data dynamically based on a predetermined time step, and use the data assimilation algorithm to perform local high-frequency numerical assimilation correction on the typhoon eyewall and its adjacent areas, and dynamically update the local wind field and temperature field states; S500. Construct an AI-NWP residual closed-loop feedback correction mechanism: During the execution of the regional coupled model integration process, periodically obtain the error residuals between the simulation results and the typhoon observation data, and input the error information as feedback into the AI weather model through constructing a residual-driven network, as a fine-tuning signal to online correct the initial field output of the AI weather model in the next prediction cycle, and realize the dynamic two-way closed-loop coupling between the AI weather model and the NWP regional model; S600. Output and evaluate the typhoon intensity simulation results: After completing the full-time integration, output the key parameters of the typhoon intensity, and conduct a structural consistency evaluation in combination with the actual typhoon observation data to complete the quantitative prediction analysis of the typhoon intensity.

[0008] The second object of the present invention is to provide a computer program product, including computer instructions, and the computer instructions are used to execute the above-mentioned typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization.

[0009] The third 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 above-mentioned typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization is realized.

[0010] (III) Technical effects Compared with the prior art, the typhoon intensity prediction method, medium and computer program based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization of the present invention have the following remarkable technical effects: (1) By generating high-dimensional meteorological initial fields and boundary fields with continuous time series through the AI weather model and constructing the NWP regional model based on the AI initial field, the present invention significantly improves the refinement and time series continuity of the initial conditions. At the same time, combined with the vortex dynamic initialization strategy, the axisymmetric circulation structure of the typhoon is extracted and reconstructed, enhancing the intensity fitting and dynamic consistency of the typhoon initial field, effectively improving the problem of insufficient early structure characterization of traditional models for rapidly intensifying typhoons, and improving the accuracy and physical consistency of intensity simulation. (2) The present invention constructs an AI-NWP two-way closed-loop coupling system, introduces an AI-NWP residual feedback mechanism in the process of integrating the regional atmosphere-ocean coupling model to realize the dynamic identification and real-time correction of simulation errors. In cooperation with the assimilation process of high-frequency multi-source observation data, the local high-resolution wind field and temperature field of key structural areas such as the typhoon eyewall are updated, strengthening the constraint ability of key physical processes during the typhoon evolution process, and significantly improving the prediction accuracy and time series stability of typhoon track, intensity and structure evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 The figure shows the implementation flowchart of the typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization of the present invention; Figure 2 The figure shows the comparison diagram of the difference distribution of column integral water vapor flux between the initial fields of the AI weather model and the ECMWF model. Among them, (a) is the difference between the AI weather model and ERA5 (AI-ERA5); (b) is the difference between the ECMWF model and ERA5 (EC-ERA5); Figure 3 The figure shows the comparison effect diagram before and after vortex dynamic initialization. In the figure: (a) is the vortex structure predicted by the AI weather model; (b) is the enhanced vortex structure after vortex dynamic initialization; Figure 4 The figure shows the comparison analysis diagram of typhoon track and intensity. In the figure: (a) is the comparison between the track simulation result and the best track; (b) is the comparison of the evolution process of the maximum wind speed of 10 meters at the periphery. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] The present invention aims to provide a typhoon intensity prediction method, medium and computer program based on an AI initial field driving an NWP regional model and integrating vortex dynamic initialization, so as to improve the intensity prediction accuracy and structure restoration ability of tropical cyclone systems, especially typhoons, during the rapid intensification stage. To make the purpose, technical solution and advantages of the implementation of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are part of the embodiments of the present invention, rather than all of the embodiments, and the described embodiments are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0013] Embodiment 1: Typhoon intensity prediction method The typhoon intensity prediction method based on an AI initial field driving an NWP regional model and integrating vortex dynamic initialization provided by the embodiment of the present invention, as Figure 1 shown, mainly includes the following steps during implementation: S100. Construct an AI weather model and generate an initial field and a boundary field: Use a deep learning architecture and construct and train an AI weather model based on historical reanalysis data. The AI weather model includes multiple deep learning sub-networks to support different prediction lead times. Use a spherical earth perception structure for input encoding and output continuous prediction results of multiple lead times covering upper-air meteorological elements as the initial field and boundary field for the subsequent NWP regional model.

[0014] In the embodiment of the present invention, an encoder-decoder structure is adopted during the construction of the AI weather model. The input end uses a spherical earth perception encoding structure to process globally gridded input data, and the longitude and latitude information in the input data is explicitly feature-enhanced through a position embedding mechanism to improve the model's expression ability for the position change and geographical dependence characteristics of tropical cyclones. The AI weather model enhances the cross-temporal dependence modeling ability through a deep residual connection structure, and the output end generates continuous upper-air meteorological variable fields of multiple prediction lead times. Its input variables at least include upper-air variables such as atmospheric temperature, pressure, specific humidity, zonal and meridional wind speeds, and geopotential height. The missing surface variables are supplemented and reconstructed through external reanalysis data or ocean models to ensure that a complete initial field and boundary conditions are provided for the subsequent NWP regional model.

[0015] It should be noted that the AI weather model's training dataset uses ERA5 reanalysis data, covering global atmospheric reanalysis data from 1979 to 2023, with a spatial resolution of 0.25°×0.25° and a temporal resolution of 1 hour. This ensures the model's generalization capabilities across diverse climate regions and seasonal variations worldwide. During model training, the weighting of samples from the typhoon-active region of the northwest Pacific Ocean was specifically enhanced. Data augmentation techniques were used to generate more typhoon-related training samples, improving the AI weather model's ability to identify and forecast typhoon systems. Furthermore, considering the limitations of AI weather models in terms of surface variables, the present invention supplements the model with external ocean model data and surface reanalysis data to ensure a complete three-dimensional atmospheric state field is provided to the regional model. Furthermore, to enhance the AI weather model's adaptability to different typhoon development stages, a multi-source historical typhoon sample set is introduced during model training. Data augmentation strategies are used to expand the sample distribution, enhancing its generalization prediction capabilities for typhoons with rapid intensification, path turning, and multi-vortex structures. A multi-objective loss function is employed during model training to balance prediction errors between different variables and improve the dynamic consistency of the overall initial field.

[0016] S200. Constructing the NWP regional model and performing vortex dynamics initialization processing: A multi-layer nested NWP regional model computational domain is constructed, with the middle and inner domains set as adaptive tracking moving grids based on the typhoon vortex center position. A vortex dynamic initialization process is introduced in the pre-processing stage before the warning is triggered. By extracting the axisymmetric vortex structure of the typhoon in the AI initial field, spectral perturbation enhancement and position relocation are performed to construct a dynamic initial field with intensity fitting capabilities and dynamic consistency. As a preference, the NWP regional model selects the WRF model and adopts a three-layer nested structure, including three layers of grids: outer domain, middle domain and inner domain. The grid spacing of the outer domain is 15-20 kilometers, which is used to provide large-scale circulation background conditions; the grid spacing of the middle domain is 5-8 kilometers, which is used to analyze the mesoscale circulation structure; the grid spacing of the inner domain is 500 meters-2 kilometers, which is used to simulate the typhoon eyewall, rain belt and local severe convective structure with high precision; the middle domain and the inner domain are set as typhoon vortex adaptive tracking moving grids, which automatically adjust the center coordinates of the calculation domain every 1-10 minutes according to the position of the typhoon center, ensuring that the high-resolution grid domain always covers the typhoon main vortex area; a two-way feedback mechanism is adopted between the three nested layers to ensure that the structural information of the fine grid domain is transmitted upward to the coarse-resolution area, enhancing the coordination and dynamic consistency of the entire simulation.

[0017] In the embodiment of the present invention, a vortex power initialization process is introduced in the pre-processing stage before the alarm is triggered, and the process preferably includes the following sub-steps when implemented: S201. Extract the axisymmetric vortex structure of the typhoon: Based on the initial field generated by the AI weather model, extract the key structure variables in the main vortex region of the typhoon, including the geopotential height field, zonal and meridional wind fields, air temperature field, and specific humidity field on the isobaric surface, extract the vortex circulation structure with axisymmetric characteristics, eliminate the asymmetric perturbation components, and construct an idealized typhoon initial structure framework; S202. Perform spectral embedding enhanced simulation: Embed the vortex structure into the initial background field of the NWP regional model, apply constraints to the large-scale background field using spectral embedding technology, and at the same time release the small and medium-scale perturbation energy, and perform a 6-hour short-term pre-integration simulation to enhance the typhoon core intensity, eyewall structure, and symmetry; S203. Implement spatial repositioning of the vortex structure: After the short-term integration is completed, according to the deviation between the actual observed typhoon center and the typhoon center position in the simulation field, use the vortex extraction and displacement function interpolation algorithm to perform spatial repositioning and re-embedding of the simulated vortex structure to ensure the spatial consistency of the dynamic initial field; S204. Construct the final dynamic initial field: Weightedly fuse the vortex simulation results after structure enhancement and position correction with the background field of the regional model to form a dynamic initial field that takes into account the large-scale accuracy and local vortex intensity, and use it as the starting input for the regional model integration simulation to improve the credibility and structure restoration ability of subsequent typhoon intensity simulations.

[0018] It should be noted that the vortex dynamic initialization process is one of the core technical innovations of the present invention. This process effectively solves the problems of weak typhoon intensity and incomplete structure output by the AI weather model. The spectral perturbation enhancement technology not only enhances the axisymmetric circulation intensity of the typhoon through 6 hours of dynamic integration, but more importantly, maintains the dynamic and thermodynamic balance of the vortex structure, avoiding the numerical instability that may be caused by traditional artificial enhancement methods. The vortex repositioning algorithm adopts a dual positioning mechanism based on the vorticity center and the pressure center to ensure that the enhanced typhoon vortex can accurately correspond to the actual observed position, providing a high-quality dynamically consistent initial field for subsequent coupled simulations.

[0019] S300. Start the AI-driven atmospheric-ocean two-way coupled model integration simulation: Based on the dynamic initial field after vortex initialization, start the NWP regional atmospheric model and simultaneously start the ocean model for coupled integration simulation. At each integration time step, the ocean model real-time updates key underlying surface variables including at least sea surface temperature, latent heat flux, and sensible heat flux, and uses them as feedback inputs into the atmospheric model to dynamically correct the sea surface boundary conditions along the typhoon movement path.

[0020] Preferably, the ocean model selects the ROMS or HYCOM model, and the grid resolution is set to match the inner domain of the atmospheric model. Its calculation depth is extended to 1000 meters to fully cover the upper mixed layer and thermocline structure, and the ocean state parameters on the typhoon path are refreshed at high frequency. During the integration process, two-way coupling is performed with the NWP regional atmospheric model through a coupling interface. The coupling variables include at least sea surface temperature, latent heat flux, sensible heat flux, and 10-meter wind stress, and are dynamically fed back to the NWP atmospheric model to update its boundary layer physical parameters and dynamically correct the sea surface thermal boundary conditions along the typhoon movement path.

[0021] In the embodiment of the present invention, the AI-driven two-way coupled atmosphere-ocean model integration simulation is further divided into the following sub-steps during implementation: S301. Synchronous initialization of the coupling system: Based on the initial atmospheric dynamic field after vortex initialization and the initial temperature and salinity field of the ocean model, the initial state matching of the atmosphere-ocean interface is performed through a coupler, and consistency constraints on sea surface temperature, sea surface height, and 10-meter wind field are established to ensure the coordination and unity of the physical fields of the atmospheric model and the ocean model at the interface boundary when the coupling system starts. S302. Real-time calculation of air-sea interface fluxes: In each integration time step, the ocean model calculates the momentum flux, sensible heat flux, latent heat flux, and net radiation flux at the air-sea interface in real time, dynamically determines the flux exchange coefficient according to the difference between the sea surface temperature and the atmospheric temperature, 10-meter wind speed, and relative humidity, and establishes a non-linear flux parameterization scheme. S303. Calculation of ocean dynamic response: The ocean model calculates the Ekman transport, vertical mixing, and upwelling processes in the ocean surface layer in real time according to the wind stress drive provided by the atmospheric model, simulates the phenomena of deepening of the ocean mixed layer, cooling of the sea surface temperature, and upwelling of subsurface water caused by typhoon strong winds, adopts a turbulent mixing scheme to handle the enhanced ocean vertical mixing induced by typhoons, and simulates the characteristics of cold water vortices in the typhoon wake through an ocean vortex parameterization scheme. S304. Dynamic update of the atmospheric boundary layer: The atmospheric model receives the sea surface temperature, sea surface roughness, and air-sea flux information fed back by the ocean model, and updates the thermal structure and dynamic characteristics of the atmospheric boundary layer in real time. The temperature profile and stability stratification of the atmospheric boundary layer are affected by the spatial distribution of the sea surface temperature, the near-surface wind speed shear and turbulence intensity are adjusted by changes in sea surface roughness, and the water vapor content and thermal balance of the atmospheric boundary layer are adjusted by the input of air-sea fluxes, realizing the dynamic response of the underlying surface conditions along the typhoon movement path.

[0022] It should be noted that the atmosphere-ocean coupled simulation is a key technical means to make up for the lack of ocean physical processes in the AI weather model. The introduction of the ocean model not only provides the dynamically changing sea surface temperature boundary condition, but more importantly, simulates the interaction process between typhoons and the ocean, including ocean mixing, upwelling, and sea surface temperature cooling effects caused by typhoon strong winds. This two-way coupling mechanism enables the simulation system to capture the physical mechanisms of typhoon rapid intensification and rapid weakening, especially the process of typhoons obtaining sufficient energy supply in deep warm water areas, and the phenomenon of typhoon intensity being inhibited in shallow or cold water areas.

[0023] S400. Real-time assimilation correction of multi-source observation data to optimize the simulation process: During the execution of the atmosphere-ocean regional coupling model integration process, multi-source near-surface observation data is dynamically introduced based on a predetermined time step, and a data assimilation algorithm is used to perform local high-frequency numerical assimilation correction on the typhoon eyewall and its adjacent areas, dynamically updating the local wind field and temperature field states; Preferably, the introduced multi-source near-surface observation data is synthetic aperture radar (SAR) retrieved wind field data, wind profiler radar data, microwave satellite retrieved data, and / or automatic weather station ground observation data, and an ensemble Kalman filter algorithm is used for numerical assimilation. An adaptive local enhancement coefficient is set in the typhoon eyewall area to improve the ability to capture the high-resolution structure of the wind speed and temperature field in the strong convection core area. An observation data quality control mechanism is established during the assimilation process, and abnormal observation values are eliminated through background field inspection and spatial consistency analysis to ensure the physical rationality and numerical stability of the assimilated data.

[0024] In the embodiment of the present invention, the real-time assimilation correction of multi-source observation data includes the following sub-steps during implementation: S401. Quality control and preprocessing of multi-source observation data: Dynamically introduce multi-source observation data, perform format unification and time series alignment based on a spatio-temporal matching strategy, and perform preprocessing and quality control on all observation data, eliminating outliers, missing measurement points, and low-confidence samples that do not conform to the assimilation window; S402. Perform preprocessing of observation data and error modeling: For the time resolution, spatial distribution characteristics, and system error characteristics of different types of observation data, construct independent observation error covariance estimation models, and optimize and configure them in combination with the background error covariance matrix to meet the requirements of subsequent numerical assimilation algorithms for error coordination and weight allocation. For wind field observation data, use a wind direction inversion correction algorithm and radial filtering technology for direction consistency processing to enhance its physical usability in the typhoon eyewall area; S403. Construct a numerical assimilation framework based on the regional model: Integrate the data assimilation algorithm into the NWP regional model, set the length of the assimilation time window and the time step, and set the assimilation interpolation strategy according to the available frequency of the observed data to achieve multi-time, localized, and dynamic assimilation of different key variables; S404. Implement high-frequency local assimilation in key regions: Set an assimilation sensitive area in the typhoon eyewall area and its adjacent areas, and perform high-frequency local reanalysis and update of the observed data to ensure accurate fitting of the wind speed peak structure and asymmetry characteristics during the rapid intensification process, and improve the ability to depict the eyewall wind speed and radial wind distribution; S405. Feed the assimilation results back to the integration simulation system in real time: After each assimilation cycle is completed, feed the updated local wind field and temperature field states back to the atmosphere-ocean regional coupling model as constraint conditions, replace the original predicted state variables, and restart the NWP integration to continuously optimize the structural evolution path and intensity development trend in typhoon simulation.

[0025] It should be noted that the multi-source observed data used in this embodiment is not a simple superposition, but is fused through a data assimilation algorithm. The data assimilation algorithm can perform weighted averaging on different observed data according to their error characteristics and spatial distributions to obtain a more accurate analysis field. In this embodiment, an adaptive local enhancement coefficient is set in the typhoon eyewall area to improve the ability to capture the high-resolution structures of the wind speed and temperature fields in the strong convection core area, which is crucial for accurately predicting typhoon intensity.

[0026] S500. Construct an AI–NWP residual closed-loop feedback correction mechanism: During the execution of the regional coupling model integration process, periodically obtain the error residuals between the simulation results and the typhoon observed data, and use the constructed residual-driven network to input the error information as feedback into the AI weather model as a fine-tuning signal to online correct the initial field output of the AI weather model in the next prediction cycle, realizing the dynamic two-way closed-loop coupling between the AI weather model and the NWP regional model; Preferably, the residual-driven network adopts a deep neural network architecture. The input layer receives the deviation vector between the NWP simulation result and the observation data. The hidden layer extracts the spatio-temporal feature patterns of the deviation through a multi-layer perceptron. The output layer generates a correction signal corresponding to the weight space of the AI weather model. The network training adopts an online learning method, and the learning rate is set to be adaptively adjusted and dynamically optimized according to the correction effect. The fine-tuning process only adjusts the weights of the output layer and the last two hidden layers of the AI weather model to maintain the stability of the main architecture of the model and avoid the phenomenon of catastrophic forgetting. Moreover, the AI-NWP residual feedback correction mechanism introduces a control threshold strategy. When the error residual between the simulation result of the NWP regional model and the observation data exceeds the set threshold, a full-cycle re-initialization process is started to reconstruct the input of the AI weather model and the initial field of the NWP regional model simultaneously, so as to avoid the expansion of the prediction deviation caused by error accumulation.

[0027] In the embodiment of the present invention, the AI-NWP residual feedback correction mechanism includes the following sub-steps when implemented: S501. Multi-dimensional forecast error analysis and quantification: Compare the forecast results of the NWP coupled model with the typhoon observation data at a preset cycle, calculate the prediction deviations of the typhoon center position, maximum sustained wind speed, minimum central pressure and typhoon movement speed, and use the weighted root mean square error and / or mean absolute error to quantify the magnitude and distribution characteristics of the systematic deviation; S502. Residual feature extraction and pattern recognition: Construct a residual analysis network based on deep learning. The input layer receives the deviation vector between the NWP simulation result and the observation data, extracts the spatial distribution pattern of the error through a convolutional neural network, establishes a non-linear mapping relationship between the error and the environmental factors, and outputs a correction signal corresponding to the weight space of the AI weather model; S503. Online fine-tuning and optimization of the AI weather model weights: Based on the correction signal output by the residual analysis network, online adjust the model weight parameters. The fine-tuning process only performs incremental updates on the weights of the model output layer and the last two hidden layers to adapt to the NWP prediction error performance in the current simulation cycle; S504. Construct an AI-NWP two-way closed-loop coupling process: During the regional simulation process, every time a predetermined integration period (such as 6 hours or 12 hours) passes, repeat the residual extraction and AI correction process to achieve dynamic cyclic feedback between the NWP and the AI weather model, so that the AI weather model has the ability to dynamically respond and adaptively correct with the NWP simulation error.

[0028] It should be noted that the AI-NWP residual feedback correction mechanism is an important innovation of the present invention. This mechanism identifies systematic deviation patterns of the AI model through a deep learning network and converts this deviation information into a correction signal for the model weights, enabling online learning and continuous optimization of the AI model. The fine-tuning process adopts a conservative strategy, only adjusting the weights of the output layer and the last two hidden layers to avoid damaging the core architecture of the AI model. The design of the threshold control mechanism prevents the negative impact of incorrect feedback information on the model performance. When an abnormally large forecast deviation is detected, the system will automatically trigger the re-initialization process to ensure the stability and reliability of the forecast system.

[0029] S600. Output and evaluate the typhoon intensity simulation results: After completing the integration for the entire time period, key parameters of the typhoon intensity are output, and a structural consistency evaluation is carried out in combination with the actual typhoon observation data to complete the quantitative prediction analysis of the typhoon intensity. Preferably, the evaluation indicators of the typhoon intensity simulation results include the maximum 10-meter wind speed, the minimum central pressure, the eye wall radius, the wind field asymmetry index, and the path error. A time evolution curve of the typhoon intensity is generated through time series analysis. The structural consistency evaluation is quantitatively evaluated by comparing the matching degree of the simulated typhoon eye diameter, eye wall thickness, and spiral rainband distribution with the actual typhoon observation data.

[0030] Example 2: Application example Based on the above Example 1, to further verify the actual application effect of the technical solution of the present invention, this example takes Typhoon Doksuri in 2023 as a case to detail the complete implementation process and forecast effect verification of the typhoon intensity prediction method of the present invention based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization.

[0031] According to the technical solution of Example 1, the NWP regional model adopts a three-layer nested structure: the outer domain D01 (18 km resolution, 311×251 grid), the middle domain D02 (6 km resolution, 271×271 grid), and the inner domain D03 (2 km resolution, 211×211 grid). The middle domain and the inner domain are configured as typhoon tracking moving grids, with 50 vertical layers and a model top of 50 hPa. The WSM6 microphysics, Dudhia shortwave, and RRTM longwave radiation, modified MM5 boundary layer, and Kain-Fritsch cumulus parameterization schemes are adopted.

[0032] AI Initial Field Quality Verification. According to step S100, an AI weather model is used to generate the initial field and boundary conditions. Through comparison and verification with the ECMWF model and ERA5 reanalysis data, the AI weather model is in good agreement with the observations in the prediction of the column integrated water vapor flux field, accurately capturing the monsoon transport, subtropical high, and double typhoon circulation structures. Quantitative analysis shows that the root mean square error of the geopotential height field of the AI weather model is reduced by 2 m compared with the ECMWF model 2 s -2 , and the wind field error is reduced by 0.5 m / s -1 , verifying the effectiveness of the AI weather model as the driving field of the regional model. However, as shown in Figures a and b of Figure 2 , both the AI weather model and the ECMWF model have obvious problems of underestimating the intensity in the typhoon core area, further confirming the necessity of introducing the vortex dynamic initialization technology.

[0033] After that, the vortex dynamic initialization is implemented according to step S200. As shown in Figures a and b of Figure 3 , the maximum wind speed of the typhoon predicted by the AI weather model before initialization is only 23.9 m / s -1 , and the eyewall structure is blurred; after vortex separation, 6-hour dynamic enhancement, and repositioning, the maximum wind speed of the typhoon is enhanced to 49.0 m / s -1 , approaching the SAR observation value of 48.3 m / s -1 , the eyewall structure is clear, and the radial wind field distribution is reasonable. The vortex enhancement process strictly follows the dynamic balance, ensuring the physical consistency of the enhanced vortex and the background field.

[0034] Then, the atmosphere-ocean two-way coupled integration is started according to step S300. The ROMS ocean model is used, with a 2-km resolution matching the inner atmospheric domain and a calculation depth of 1000 meters. The simulation results show that the sea surface temperature cooling effect caused by the typhoon reaches 2 - 3 °C, the characteristics of the cold water vortex in the ocean wake are consistent with the observations, and the air-sea interaction process is accurately reproduced, providing an important ocean thermal feedback mechanism for typhoon intensity changes.

[0035] Next, the multi-source observation data assimilation is implemented according to step S400. The ensemble Kalman filter algorithm is used to assimilate the SAR wind field, AMSU-A profile, and surface observation data. After assimilation, the typhoon center position error is reduced by 15%, the maximum wind speed error is reduced by 20%, the eyewall wind field structure is closer to the observations, and the typhoon asymmetry characteristics are accurately described.

[0036] The typhoon track and intensity forecast evaluation results are shown in Figure 4As shown in Figures a and b, the typhoon track simulation results indicate that the AI-driven NWP model (AI_NWP) has a high degree of coincidence with the best track data in the first 24 hours of track forecasting. Compared with the pure AI model forecasting, the track deviation in the later stage is reduced, and the overall track forecasting accuracy is better than that of the traditional ECMWF-driven NWP model (EC_NWP). In terms of typhoon intensity prediction, the AI-driven WRF model has achieved significant improvement compared with the pure AI model. During the 72-hour forecast of Typhoon Doksuri, the root mean square error of the maximum wind speed of the pure AI model was as high as 29.3 m / s -1 , systematically underestimating the typhoon intensity; the root mean square error of the AI-driven NWP model was reduced to 5.1 m / s -1 , and the forecasting accuracy improvement reached 87%. The track forecast in the first 24 hours is highly consistent with the best track, and the overall accuracy is better than that of the traditional EC-driven NWP model.

[0037] In addition, using high-resolution SAR data for independent verification, at 10:00 on July 24, the maximum wind speed of the typhoon observed by SAR was 48.3 m / s -1 , and the AI-driven NWP simulation was 49.0 m / s -1 , with an error of 1.5%; at 10:00 on July 25, the SAR observation was 57.2 m / s -1 , and the simulation was 59.1 m / s -1 , with an error of 3.3%. Compared with the more than 50% underestimation of the pure AI model, this method has significantly improved in both typhoon intensity and eyewall structure forecasting.

[0038] Construct a residual feedback mechanism according to step S500. By comparing the deviation between the NWP simulation results and the observed data, use a deep neural network to extract error features and perform online fine-tuning of the AI model weights. Execute a feedback loop every 12 hours to make the AI model gradually adapt to regional characteristics and further improve the forecasting accuracy and system stability. The results show that this method is significantly better than traditional methods in terms of typhoon intensity forecasting accuracy, structure restoration ability, and rapid change prediction, especially in the forecasting ability during the rapid intensification stage of typhoons. It provides an advanced technical means for typhoon disaster prevention and mitigation and ocean meteorological operational applications, and has important scientific value and broad application prospects.

[0039] Through the above embodiments, the object of the present invention is completely and effectively achieved. Those skilled in the art can understand that the present invention includes but is not limited to the content described in the drawings and the above specific embodiments. Although the present invention has been described with respect to the currently considered most practical and preferred embodiments, 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 typhoon intensity prediction method based on driving the NWP regional model with an AI initial field and integrating vortex dynamic initialization, characterized in that, At least include the following steps: S100. Adopt a deep learning architecture and construct and train an AI weather model based on historical reanalysis data. The model includes multiple deep learning sub-networks to support different prediction lead times, uses a spherical Earth perception structure for input encoding, and outputs multi-lead-time continuous prediction results covering upper-air meteorological elements; S200. Construct a multi-nested structure NWP regional model computational domain, set the middle domain and the inner domain as adaptive tracking moving grids based on the typhoon vortex center position, and introduce vortex dynamic initialization in the preprocessing stage before the start of forecasting, and construct a dynamic initial field with intensity fitting ability and dynamic consistency; S300. Based on the dynamic initial field after vortex initialization, start the synchronous coupled integration simulation of the NWP regional atmospheric model and the ocean model. In each integration time step, the ocean model updates in real time key underlying surface variables including at least sea surface temperature, latent heat flux, and sensible heat flux, and uses them as feedback inputs into the atmospheric model to dynamically correct the sea surface boundary conditions along the typhoon movement path; S400. During the execution of the regional coupled model integration process, introduce multi-source near-surface observation data dynamically based on a predetermined time step, and use a data assimilation algorithm to perform local high-frequency numerical assimilation correction on the typhoon eyewall and its adjacent areas, and dynamically update the local wind field and temperature field states; S500. During the execution of the regional coupled model integration process, periodically obtain the error residuals between the simulation results and the typhoon observation data, and use the error information as feedback and input it into the AI weather model through a constructed residual-driven network, as a fine-tuning signal to online correct the initial field output of the AI weather model in the next prediction cycle; S600. After completing the full-time integration, output the key parameters of the typhoon intensity, and combine the actual typhoon observation data to conduct a structural consistency evaluation to complete the quantitative prediction analysis of the typhoon intensity.

2. The typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to claim 1, wherein In step S100, the AI weather model adopts an encoder-decoder structure. The input end uses a spherical Earth perception encoding structure, and explicitly enhances the longitude and latitude information in the input data through a position embedding mechanism; the AI weather model enhances the cross-temporal dependence modeling ability through a deep residual connection structure, and the output end generates continuous upper-air meteorological variable fields for multiple prediction lead times. The input variables include at least atmospheric temperature, pressure, specific humidity, zonal and meridional wind speeds, and geopotential height, and the missing surface variables are supplemented and reconstructed through external reanalysis data or the ocean model.

3. The typhoon intensity prediction method based on the AI initial field-driven NWP regional model and integrating vortex dynamic initialization according to claim 1, wherein, In step S200, the NWP regional model adopts a three-layer nested structure, including three layers of grids: the outer layer domain, the middle layer domain, and the inner layer domain. The grid spacing of the outer layer domain is 15 - 20 km, the grid spacing of the middle layer domain is 5 - 8 km, and the grid spacing of the inner layer domain is 500 m - 2 km. The middle layer domain and the inner layer domain are set as typhoon vortex adaptive tracking moving grids, and automatically adjust the center coordinates of the computational domain every 1 - 10 minutes according to the typhoon center position to ensure that the high-resolution grid domain always covers the typhoon main vortex area; a two-way feedback mechanism is adopted between the three-layer nesting to ensure that the structural information of the fine grid domain is transmitted upward to the coarse resolution area.

4. The typhoon intensity prediction method based on the AI initial field-driven NWP regional model and integrated with vortex dynamic initialization according to claim 1 or 3, characterized in that, In step S200, during the preprocessing stage before the start of forecasting, introduce a vortex dynamic initialization process, which at least includes the following sub-steps: S201. Extract the axisymmetric vortex structure of the typhoon: Based on the initial field generated by the AI weather model, extract the key structure variables in the main vortex region of the typhoon, extract the vortex circulation structure with axisymmetric characteristics, eliminate the asymmetric perturbation components, and construct an idealized typhoon initial structure framework; S202. Perform spectral embedding enhanced simulation: Embed the vortex structure into the initial background field of the NWP regional model, apply constraints to the large-scale background field using spectral embedding, and at the same time release the small and medium-scale perturbation energy. Perform a 6-hour short-term pre-integration simulation to enhance the typhoon core intensity, eyewall structure, and symmetry; S203. Implement spatial repositioning of the vortex structure: After the short-term integration is completed, according to the deviation between the actual observed typhoon center and the typhoon center position in the simulation field, use the vortex extraction and displacement function interpolation algorithm to perform spatial repositioning and re-embedding of the simulated vortex structure; S204. Construct the final dynamic initial field: Perform weighted fusion of the vortex simulation results with enhanced structure and corrected position and the background field of the regional model to form a dynamic initial field that takes into account the large-scale accuracy and local vortex intensity, and use it as the starting input for the regional model integration simulation.

5. The typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to claim 1, characterized in that, In step S300, the ocean model selects ROMS or HYCOM, the ocean grid resolution is set to match the inner domain of the atmospheric model, the calculation depth of the ocean model is extended to 1000 meters to completely cover the upper mixed layer and thermocline structure, and the ocean state parameters on the typhoon path are refreshed at high frequency. During the integration process, two-way coupling is performed with the atmospheric model through the coupling interface, and the coupling variables include at least sea surface temperature, latent heat flux, sensible heat flux, and 10-meter wind stress, and are dynamically fed back to the NWP atmospheric model to update its boundary layer physical parameters.

6. The typhoon intensity prediction method based on the AI initial field-driven NWP regional model and integrated with vortex dynamic initialization according to claim 1 or 5, characterized in that, In step S300, the AI-driven atmosphere-ocean two-way coupled model integration simulation includes the following sub-steps when implemented: S301. Coupled system synchronous initialization: Based on the atmospheric dynamic initial field after vortex initialization and the initial temperature and salinity field conditions of the ocean model, perform initial state matching at the atmosphere-ocean interface through the coupler to establish consistency constraints on sea surface temperature, sea surface height, and 10-meter wind field; S302. Real-time calculation of air-sea interface fluxes: In each integration time step, the ocean model calculates the momentum flux, sensible heat flux, latent heat flux, and net radiation flux at the air-sea interface in real time, and dynamically determines the flux exchange coefficient according to the difference between the sea surface temperature and the atmospheric temperature, 10-meter wind speed, and relative humidity; S303. Ocean dynamic response calculation: The ocean model calculates the Ekman transport, vertical mixing, and upwelling processes in the ocean surface layer in real time according to the wind stress drive provided by the atmospheric model, uses the turbulent mixing scheme to handle the enhanced ocean vertical mixing induced by the typhoon, and simulates the cold eddy characteristics in the typhoon wake through the ocean vortex parameterization scheme; S304. Dynamic update of the atmospheric boundary layer: The atmospheric model receives the sea surface temperature, sea surface roughness, and air-sea flux information fed back by the ocean model, and updates the thermal structure and dynamic characteristics of the atmospheric boundary layer in real time to achieve dynamic response to the underlying surface conditions along the typhoon movement path.

7. The typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to claim 1, characterized in that, In step S400, the introduced multi-source near-surface observation data are SAR-inverted wind field data, wind profiler radar data, microwave satellite-inverted data, and / or automatic weather station ground observation data. The EnKF algorithm is used for numerical assimilation, and an adaptive local enhancement coefficient is set in the typhoon eyewall area to improve the ability to capture the high-resolution structures of wind speed and temperature fields in the severe convection core area. During the assimilation process, an observation data quality control mechanism is established, and abnormal observation values are removed through background field tests and spatial consistency analyses.

8. The typhoon intensity prediction method based on the AI initial field-driven NWP regional model and integrated with vortex dynamic initialization according to claim 1 or 7, characterized in that, In step S400, the real-time assimilation correction of multi-source observation data to optimize the simulation includes the following sub-steps when implemented: S401. Quality control and preprocessing of multi-source observation data: During the integration process, multi-source observation data are dynamically introduced, format unification and time series alignment are performed based on the spatio-temporal matching strategy, and preprocessing and quality control are performed on all observation data to remove outliers, missing measurement points, and low-confidence samples that do not conform to the assimilation window. S402. Perform preprocessing of observation data and error modeling: For the time resolution, spatial distribution characteristics, and systematic error characteristics of different types of observation data, independent observation error covariance estimation models are constructed and optimized in combination with the background error covariance matrix. For wind field observation data, a wind direction inversion correction algorithm and a radial filtering technique are used for direction consistency processing. S403. Construct a numerical assimilation framework based on the regional model: Integrate the data assimilation algorithm into the NWP regional model, set the assimilation time window length and time step, and set the assimilation interpolation strategy according to the available frequency of the observation data to achieve multi-time, local, and dynamic assimilation of different key variables. S404. Implement high-frequency local assimilation in key areas: Set an assimilation sensitive area in the typhoon eyewall area and its adjacent areas, and perform high-frequency local reanalysis and update of the observation data to ensure accurate fitting of the wind speed peak structure and asymmetry characteristics during the rapid intensification process. S405. Feed the assimilation results back to the coupled model in real time: After each assimilation cycle is completed, the updated local wind field and temperature field states are fed back to the atmosphere-ocean regional coupled model as constraint conditions, replacing the original predicted state variables and restarting the NWP integration.

9. The typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to claim 1, characterized in that, In step S500, the residual-driven network adopts a deep neural network architecture. The input layer receives the deviation vector between the NWP simulation result and the observation data. The hidden layer extracts the spatio-temporal feature patterns of the deviation through a multi-layer perceptron. The output layer generates a correction signal corresponding to the weight space of the AI weather model. The network training adopts an online learning method, and the learning rate is set to be adaptively adjusted and dynamically optimized according to the correction effect. The fine-tuning process only adjusts the weights of the output layer and the last two hidden layers of the AI weather model. Moreover, the AI-NWP residual feedback correction mechanism introduces a control threshold strategy. When the error residual between the simulation result of the NWP regional model and the observation data exceeds the set threshold, a full-cycle re-initialization process is started, and the input of the AI weather model and the initial field of the NWP regional model are reconstructed simultaneously.

10. The typhoon intensity prediction method based on the AI initial field-driven NWP regional model and integrated with vortex dynamic initialization according to claim 1 or 9, characterized in that, In step S500, the AI-NWP residual feedback correction mechanism includes the following sub-steps when implemented: S501. Multi-dimensional forecast error analysis and quantification: Compare the forecast results of the NWP coupled model with the typhoon observation data at a preset period, calculate the prediction deviations of the typhoon center position, maximum sustained wind speed, minimum central pressure, and typhoon movement speed, and quantify the magnitude and distribution characteristics of the systematic deviation using the weighted root mean square error and / or mean absolute error; S502. Residual feature extraction and pattern recognition: Construct a residual analysis network based on deep learning. The input layer receives the deviation vector between the NWP simulation results and the observation data, extracts the spatial distribution pattern of the error through a convolutional neural network, establishes a non-linear mapping relationship between the error and environmental factors, and outputs a correction signal corresponding to the weight space of the AI weather model; S503. Online fine-tuning and optimization of the AI weather model weights: Based on the correction signal output by the residual analysis network, perform online adjustment of the weight parameters of the AI weather model. The fine-tuning process only performs incremental updates on the weights of the model output layer and the last two hidden layers to adapt to the NWP prediction error performance in the current simulation period; S504. Construct an AI-NWP two-way closed-loop coupling process: During the regional simulation process, repeat the residual extraction and AI correction process every predetermined integration period to achieve dynamic cyclic feedback between the NWP and the AI weather model, enabling the AI weather model to have the ability to dynamically respond and adaptively correct with the NWP simulation error.

11. The typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to claim 1, characterized in that, In step S600, the evaluation indicators of the typhoon intensity simulation results include the maximum 10-meter wind speed, minimum central pressure, eyewall radius, wind field asymmetry index, and path error, and a time evolution curve of the typhoon intensity is generated through time series analysis. The structural consistency evaluation is quantitatively evaluated by comparing the matching degree of the simulated typhoon eye diameter, eyewall thickness, and spiral rainband distribution with the actual typhoon observation data.

12. A computer program product, comprising computer instructions, characterized in that, The computer instructions are used to execute the typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to any one of claims 1 to 11.

13. A computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the typhoon intensity prediction method based on the AI initial field driving the NWP regional model and integrating vortex dynamic initialization according to any one of claims 1 to 11.

Images