Typhoon path prediction method, system, terminal and storage medium based on multi-modal data and archaeus model of sounding equipment

Abstract

The invention relates to the technical field of typhoon path prediction, and discloses a typhoon path prediction method, system, terminal and storage medium based on multi-modal data of sounding equipment and a sky big model, and the method comprises the steps: obtaining multi-source heterogeneous data collected by detection equipment related to typhoon prediction, carrying out fusion and adaptive enhancement on the multi-source heterogeneous data on the basis of a big Pantou meteorological model, and generating a high-resolution three-dimensional field; constructing a cascaded AI downscaling network, introducing a physical constraint layer, carrying out local refined modeling on a typhoon eye region according to the high-resolution three-dimensional field, and outputting a high-resolution three-dimensional meteorological field; and designing a typhoon-environment field interaction model based on a graph neural network, performing typhoon prediction according to the high-resolution three-dimensional meteorological field, and outputting a typhoon path ensemble forecast containing a confidence interval and a typhoon thermodynamic structure analysis report. The typhoon path prediction speed and precision are improved, the prediction time is shortened, the prediction error is reduced, and calculation resources are saved.

Description

Technical Field

[0001] This invention relates to the field of typhoon path prediction technology, and in particular to a method, system, terminal, and computer-readable storage medium for typhoon path prediction based on multimodal data from radiosonde equipment and the Pangoal Injection Model. Background Technology

[0002] Significant bottlenecks exist in typhoon track prediction. Its core reliance on supercomputers to solve complex atmospheric dynamic equations results in single predictions taking hours or even days, failing to meet the demands for real-time typhoon monitoring and rapid response. Furthermore, numerical models, due to simplified physical processes (such as ignoring small-scale ocean turbulence and local wind field variations), lack the ability to model complex meteorological phenomena like abrupt typhoon track changes and multiple typhoon interactions, leading to significant prediction errors. More critically, data assimilation mechanisms (the temporal and spatial unification of data from aerostats (irregular trajectories), satellites (gridized), and radars (polar coordinates)) are inefficient, primarily relying on satellite and ground station data. They fail to effectively integrate upper-air three-dimensional meteorological data (such as vertical wind speed profiles and temperature and humidity gradients) collected by aerostats and drop equipment, limiting the accuracy of the initial field and further impacting prediction reliability.

[0003] While deep learning-based prediction methods have achieved breakthroughs in computational speed, their limitations are equally prominent. Existing methods largely rely on two-dimensional neural networks to process historical meteorological data, lacking the ability to model three-dimensional atmospheric structures (such as vertical wind shear in the typhoon eye) and failing to incorporate physical constraints (such as energy conservation equations). This results in poor stability of predictions under extreme weather scenarios (such as typhoon reversals). Furthermore, these models fail to fully integrate multimodal data; satellite remote sensing, real-time observations from aerostats, and high-altitude data from drop equipment are isolated, resulting in insufficient ability to capture detailed local features. In addition, model training depends on historical datasets, leading to a lag in response to sudden meteorological events. Moreover, the "black box" nature of these models (poor interpretability of AI algorithms) makes the prediction results uninterpretable, hindering the reliability of disaster prevention decisions.

[0004] While existing large-scale meteorological model forecasting methods perform well in global weather forecasting, they still have shortcomings in specific applications for typhoon tracks. Despite employing a 3D neural network architecture, actual data input still primarily relies on satellite and ground-based data; real-time 3D observation data from aerostat-deployed equipment is not deeply integrated, resulting in insufficient capture of upper-level dynamic characteristics in the typhoon core area. Furthermore, large-scale models require cloud deployment, making it difficult to adapt to the real-time inference needs of edge devices such as aerostats, and hindering data closed-loop optimization of the typhoon core area. In complex scenarios (such as multiple typhoon interactions and coastal topographic effects), the lack of dynamic input from multimodal data such as ocean thermal and topographic disturbances significantly reduces generalization ability and increases prediction error fluctuations. These shortcomings collectively limit the practical effectiveness of large-scale meteorological models in accurate typhoon track forecasting.

[0005] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0006] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for typhoon path prediction based on multimodal data from radiosonde equipment and the Pangoal Meteorological Model, aiming to solve the problems of large prediction errors and insufficient reliability of existing meteorological models in typhoon path prediction.

[0007] To achieve the above objectives, this invention provides a typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Ray model. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Ray model includes the following steps:

[0008] Acquire multi-source heterogeneous data collected by detection equipment related to typhoon forecasting, and fuse and adaptively enhance the multi-source heterogeneous data based on the Pango meteorological big model to generate a high-resolution three-dimensional field;

[0009] A cascaded AI downscaling network is constructed, a physical constraint layer is introduced, and the eye of the typhoon is locally refined based on the high-resolution three-dimensional field to output a high-resolution three-dimensional meteorological field.

[0010] Design a typhoon-environment field interaction model based on graph neural networks, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report.

[0011] Optionally, the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Meteorological Model, wherein acquiring multi-source heterogeneous data collected by the sounding equipment related to typhoon prediction, and fusing and adaptively enhancing the multi-source heterogeneous data based on the Pangoal Meteorological Model to generate a high-resolution three-dimensional field, specifically includes:

[0012] By accessing the global forecast data stream of the Pangu meteorological big data model, low-resolution grid data of the typhoon core area is extracted. The low-resolution grid data includes pressure field, three-dimensional wind field, humidity field and sea surface temperature field.

[0013] A typhoon feature space is constructed based on a historical typhoon database. The current typhoon type is identified through clustering algorithms, and key prediction areas are dynamically delineated.

[0014] The observation data from the detection equipment are synchronized. The observation data from the detection equipment includes: temperature profile data of the typhoon eye area collected by the satellite microwave imager, three-dimensional wind speed field collected by the UAV-borne millimeter-wave radar, sea temperature and air pressure data collected by the ocean buoy, and upper-air three-dimensional meteorological data collected by the airship.

[0015] Spatiotemporal kriging interpolation was used to map point observation data onto the Pango meteorological model, and nested high-resolution subgrids were established for the typhoon eye region.

[0016] A typhoon data generator based on a diffusion model is constructed. Low-resolution grid data and observation data from detection equipment are input into the typhoon data generator for fusion processing, and a complete high-resolution three-dimensional field of temperature, pressure and humidity in the typhoon eye region is output.

[0017] Deploy an anomaly detection module that uses an autoencoder to identify and remove noisy data observed by the detection device.

[0018] Optionally, the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Meteorological Model, wherein the construction of a cascaded AI downscaling network, the introduction of a physical constraint layer, and the local fine-scale modeling of the typhoon eye region based on the high-resolution three-dimensional field to output a high-resolution three-dimensional meteorological field specifically include:

[0019] A cascaded AI downscaling network is constructed, which includes a computer vision model and a three-dimensional convolutional neural network. The computer vision model is used to extract global meteorological features output by the Pangu meteorological model and input them into the three-dimensional convolutional neural network. The three-dimensional convolutional neural network is used to perform local fine modeling of the typhoon eye area and output a three-dimensional meteorological field based on the input satellite infrared cloud image and a database of historical similar typhoon cases.

[0020] A lightweight LSTM-ENKF hybrid assimilation algorithm was developed. The algorithm predicts the systematic bias distribution of the Pangoal Meteorological Model using LSTM. An ensemble Kalman filter is used to dynamically inject UAV observation data into the cascaded AI downscaling network, and a high-resolution three-dimensional meteorological field is output after data assimilation and physical constraint correction.

[0021] A physical constraint layer is introduced, and a partial differential equation verification module is added to the output of the cascaded AI downscaling network to ensure that the predicted field conforms to the fluid dynamics equations.

[0022] Optionally, the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Instantaneous Model, wherein the design of a typhoon-environmental field interaction model based on a graph neural network, and the typhoon prediction based on the high-resolution three-dimensional meteorological field, outputs a typhoon path ensemble forecast including confidence intervals and a typhoon thermodynamic structure analysis report, specifically includes:

[0023] Design a typhoon-environmental field interaction model based on graph neural networks, abstract the typhoon structure into a dynamic graph, input environmental field data, and predict the probability distribution of the typhoon's movement direction;

[0024] A deep reinforcement learning policy optimizer is constructed, which uses the historical typhoon track fit and physical rationality score as the reward function to output a set of typhoon track forecasts containing confidence intervals within a preset time period.

[0025] Develop a multi-task learning architecture to predict the maximum wind speed and minimum air pressure at the center of a typhoon and identify signs of sudden intensity changes.

[0026] Integrating interpretable AI components, the Spatial SHAP algorithm is used to visualize key areas affecting intensity and generate typhoon thermodynamic structure analysis reports.

[0027] Optionally, in the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Instantaneous Model, the typhoon thermodynamic structure analysis report includes information on warm core height, outflow layer intensity, vertical temperature lapse rate, specific humidity, tangential wind, and radial wind.

[0028] Optionally, the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Ray model further includes:

[0029] The TinyML architecture is deployed on the detection equipment to perform knowledge distillation on the cascaded AI downscaling network. The preprocessing process of meteorological data is standardized and modularized through hardware optimization.

[0030] Key structural indicators of the eye of the storm are uploaded to the cloud server at set intervals to support dynamic model optimization and prediction result generation in the cloud, and the prediction results are displayed on the visualization platform.

[0031] Optionally, the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Ray model further includes:

[0032] The prediction error of the typhoon forecast is obtained. When the prediction error exceeds a threshold, the Pangu meteorological model is triggered to perform model fine-tuning.

[0033] Furthermore, to achieve the above objectives, the present invention also provides a typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal-Ray model, wherein the typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal-Ray model includes:

[0034] The multi-source heterogeneous data fusion and adaptive enhancement module is used to acquire multi-source heterogeneous data collected by detection equipment related to typhoon prediction, and to fuse and adaptively enhance the multi-source heterogeneous data based on the Pangu meteorological big model to generate a high-resolution three-dimensional field.

[0035] The dynamic downscaling and physical constraint correction module is used to construct a cascaded AI downscaling network, introduce a physical constraint layer, perform local fine modeling of the typhoon eye region based on the high-resolution three-dimensional field, and output a high-resolution three-dimensional meteorological field.

[0036] The multimodal typhoon evolution prediction and report generation module is used to design a typhoon-environmental field interaction model based on graph neural networks, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report.

[0037] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model stored in the memory and executable on the processor. When the typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model is executed by the processor, it implements the steps of the typhoon path prediction method based on multimodal data from a radiosonde and the Pangoal model as described above.

[0038] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model, and when the typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model is executed by a processor, it implements the steps of the typhoon path prediction method based on multimodal data from a radiosonde and the Pangoal model as described above.

[0039] This invention acquires multi-source heterogeneous data collected by detection equipment related to typhoon forecasting. Based on the Pangoal Meteorological Model, this multi-source heterogeneous data is fused and adaptively enhanced to generate a high-resolution three-dimensional field. A cascaded AI downscaling network is constructed, introducing a physical constraint layer. Based on the high-resolution three-dimensional field, a localized refined model of the typhoon eye region is performed, outputting a high-resolution three-dimensional meteorological field. A typhoon-environmental field interaction model based on a graph neural network is designed. Typhoon forecasting is performed based on the high-resolution three-dimensional meteorological field, outputting a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report. This invention combines the Pangoal Meteorological Model, artificial intelligence technology, and eye detection equipment to achieve accurate typhoon forecasting, improving the speed and accuracy of typhoon path prediction, significantly reducing prediction time and error, and saving computational resources. Attached Figure Description

[0040] Figure 1 This is a flowchart of a preferred embodiment of the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Rayet model of the present invention;

[0041] Figure 2This is a schematic diagram illustrating the principle of typhoon prediction in a preferred embodiment of the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Infinite Model of the present invention.

[0042] Figure 3 This is a structural diagram of a preferred embodiment of the typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal-Rayet model of the present invention.

[0043] Figure 4 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0045] This invention acquires low-resolution data of the typhoon core area through the Pangu meteorological model, integrates real-time observation data from multiple sources such as satellites and drones, and generates a high-resolution field through spatiotemporal interpolation and Diffusion Model. Then, it uses a cascaded AI downscaling network combined with physical constraint correction for fine modeling. Finally, it uses graph neural networks and deep reinforcement learning to quantify the interaction between the typhoon and the environment and outputs a path forecast with confidence intervals and a thermodynamic structure analysis report, thus achieving accurate typhoon prediction.

[0046] The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Rayet model, as described in the preferred embodiment of the present invention, is as follows: Figure 1 and Figure 2 As shown, the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Institutional Array (PANGU) model includes the following steps:

[0047] Step S10: Acquire multi-source heterogeneous data collected by detection equipment related to typhoon forecasting, and fuse and adaptively enhance the multi-source heterogeneous data based on the Pangu meteorological big model to generate a high-resolution three-dimensional field.

[0048] Specifically, the global forecast data stream of the Pangu meteorological model is accessed, and low-resolution grid data (0.25° accuracy) of the typhoon core area is extracted. The low-resolution grid data includes pressure field, three-dimensional wind field, humidity field and sea surface temperature field.

[0049] A typhoon feature space is constructed based on a historical typhoon database (which consists of low-resolution grid data of the core area of ​​historical typhoons). The typhoon feature space is a high-dimensional data space constructed by extracting key meteorological parameters (such as air pressure, wind speed, temperature, and typhoon structure characteristics) from historical typhoon data. The current typhoon type (such as super typhoon or double-walled typhoon) is identified by clustering algorithms, and key prediction areas are dynamically delineated (e.g., adjustable radius of 300-800 kilometers).

[0050] The observation data from the detection equipment will be synchronized. The observation data from the detection equipment includes: temperature profile data of the typhoon eye area collected by the Advanced Microwave Scanning Radiometer 2 (AMSR-2), three-dimensional wind speed field collected by UAV-borne millimeter-wave radar (e.g., 10-meter resolution three-dimensional wind speed field of UAV-borne millimeter-wave radar), sea surface temperature and air pressure data collected by ocean buoys (SST), and upper-air three-dimensional meteorological data (e.g., vertical wind speed profile, temperature, humidity and pressure) collected by airships.

[0051] Design a spatiotemporal alignment algorithm: Use spatiotemporal kriging interpolation to map point observation data (referring to local meteorological data collected by the above detection equipment, which is called point observation data because the spatial and temporal ranges are not very continuous) onto the Pangu meteorological model. Establish nested high-resolution subgrids (0.01° accuracy) for the typhoon eye area (e.g., radius 50 km) to improve the modeling accuracy of the typhoon core area (0.25°->0.01°, approximately 25km->1km).

[0052] A typhoon data generator based on a diffusion model is constructed. Low-resolution grid data and observation data from detection equipment (i.e., sparse observation data) are input into the typhoon data generator for fusion processing, and a complete high-resolution (0.05°) three-dimensional field of temperature, pressure, and humidity in the typhoon eye region is output. The loss function combines Wasserstein distance and physical conservation constraints (such as the mass conservation equation). In the typhoon data generator, the loss function is the core mechanism driving model learning and optimization.

[0053] Deploy an anomaly detection module that uses an autoencoder to identify and remove noise data observed by detection devices (including satellite microwave imagers, drones, ocean buoys, and airships).

[0054] The adaptive enhancement of this invention refers to the intelligent repair, supplementation and optimization of multi-source meteorological data through AI algorithms to improve the integrity, resolution and reliability of the data.

[0055] Step S20: Construct a cascaded AI downscaling network, introduce a physical constraint layer, perform local fine modeling of the typhoon eye region based on the high-resolution three-dimensional field, and output a high-resolution three-dimensional meteorological field.

[0056] Specifically, a cascaded AI downscaling network is constructed, which includes a computer vision model (Vision Transformer) and a three-dimensional convolutional neural network (3D-CNN). The computer vision model is used to extract global meteorological features (such as cyclonic vorticity fields) output by the Pangoal Meteorological Model and input them into the three-dimensional convolutional neural network. The three-dimensional convolutional neural network is used to perform local fine-grained modeling of the typhoon eye area. Based on the input satellite infrared cloud image and a database of historical similar typhoon cases, it outputs a three-dimensional meteorological field (e.g., a three-dimensional meteorological field with a resolution of 0.05°) covering the typhoon core area (e.g., a radius of 200 kilometers).

[0057] A lightweight LSTM-ENKF hybrid assimilation algorithm was developed. LSTM (Long Short-Term Memory) was used to predict the systematic bias distribution of the Pangu meteorological model, quantifying the spatial distribution of model errors and providing a clear correction direction for data assimilation. This enhanced the downscaling model's ability to capture complex meteorological characteristics of the typhoon core area, supporting accurate prediction. An ensemble Kalman filter (ENKF) was used to dynamically inject UAV observation data into the cascaded AI downscaling network, outputting a high-resolution three-dimensional meteorological field after data assimilation and physical constraint correction.

[0058] A physical constraint layer is introduced, and a partial differential equation (PDE) verification module is added to the output of the cascaded AI downscaling network to ensure that the predicted field conforms to the fluid dynamics equations.

[0059] Step S30: Design a typhoon-environmental field interaction model based on graph neural network, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast containing confidence intervals and a typhoon thermodynamic structure analysis report.

[0060] Specifically, a typhoon-environmental field interaction model based on graph neural networks is designed. The typhoon structure (the physical components and dynamic relationships within a typhoon, represented by nodes and edges) is abstracted into a dynamic graph. Environmental field data is input to predict the probability distribution of the typhoon's movement direction.

[0061] A deep reinforcement learning policy optimizer is constructed, which uses the historical typhoon track fit and physical rationality score as the reward function to output a set of typhoon track forecasts containing confidence intervals within a preset time (e.g., 72 hours).

[0062] A multi-task learning architecture was developed, with the main task being to predict the maximum sustained wind (MSW) and the minimum central pressure (CP) at the typhoon center, and the auxiliary task being to identify signs of abrupt changes in intensity (such as eyewall replacement events).

[0063] It integrates interpretable AI components and uses the Spatial SHAP algorithm (Spatial Shapley AdditiveExplanations, an extended method based on SHAP values ​​specifically used to interpret spatial data or models with spatial structures (such as remote sensing images, geographic information systems / GIS, weather forecasts, urban planning, etc.)) to visualize key areas of influence intensity, generating a typhoon thermodynamic structure analysis report (PDF format) with annotations of physical parameters such as warm core height, outflow layer intensity, vertical temperature lapse rate, specific humidity, tangential wind, and radial wind.

[0064] Furthermore, the TinyML architecture (Tiny Machine Learning, a machine learning technology for ultra-low power, micro-embedded devices, designed to deploy AI models on resource-constrained edge devices (such as MCUs, sensors, and IoT devices) is deployed on the detection equipment to achieve "edge intelligence." Knowledge distillation is performed on the cascaded AI downscaling network (teacher network: large cloud model; student network: binarized CNN). Through hardware optimization, the meteorological data preprocessing process is standardized and modularized, enabling it to automatically execute multiple preprocessing steps in a fixed sequence, like an industrial assembly line, to improve data processing efficiency and reduce power consumption. For example, using FPGA to implement the meteorological data preprocessing pipeline reduces power consumption by 42% (actual measured data).

[0065] Key indicators of the eye structure (such as the eyewall gradient wind speed ratio) are uploaded to the cloud server at set intervals (e.g., every 5 minutes) to support dynamic model optimization and prediction result generation in the cloud, and the prediction results are displayed on the visualization platform.

[0066] Construct an online incremental learning framework: Data layer: Streaming real-time observation data streams (Kafka pipeline) to perform continuous, unbounded data processing on meteorological data (such as wind speed, temperature, humidity, radar echoes, etc.) collected in real time by typhoon monitoring equipment (such as drones, radiosondes); Model layer: Employing the Elastic Weight Consolidation (EWC) algorithm to prevent catastrophic forgetting, mainly through a "parameter importance weighted constraint" mechanism to constrain the update magnitude of model parameters and prevent the forgetting of key features of historical typhoon samples during new data training.

[0067] Update strategy: Obtain the prediction error of the typhoon forecast. When the prediction error exceeds a threshold (e.g., path deviation > 50 km / 6 hours), trigger the Pangu meteorological model to perform model fine-tuning.

[0068] The technical effects that this invention can bring are as follows:

[0069] (1) Speed ​​improvement: The typhoon prediction time has been shortened from 5 hours in the traditional numerical model to 10 seconds.

[0070] (2) Improved accuracy: Typhoon path prediction error is reduced by 20%-30%, supporting complex abrupt path modeling.

[0071] (3) Resource saving: The computing power requirement is only 1 / 1000 of that of traditional methods, which is suitable for the deployment of edge computing devices.

[0072] Furthermore, such as Figure 3 As shown, based on the above-mentioned typhoon path prediction method using multimodal data from radiosonde equipment and the Pangoal-King model, this invention also provides a typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal-King model. The typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal-King model includes:

[0073] The multi-source heterogeneous data fusion and adaptive enhancement module 51 is used to acquire multi-source heterogeneous data collected by detection equipment related to typhoon prediction, and to fuse and adaptively enhance the multi-source heterogeneous data based on the Pangu meteorological big model to generate a high-resolution three-dimensional field.

[0074] The dynamic downscaling and physical constraint correction module 52 is used to construct a cascaded AI downscaling network, introduce a physical constraint layer, perform local fine modeling of the typhoon eye area based on the high-resolution three-dimensional field, and output a high-resolution three-dimensional meteorological field.

[0075] The multimodal typhoon evolution prediction and report generation module 53 is used to design a typhoon-environmental field interaction model based on graph neural networks, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast containing confidence intervals and a typhoon thermodynamic structure analysis report.

[0076] Furthermore, such as Figure 4 As shown, based on the above-mentioned typhoon path prediction method and system based on multimodal data from radiosonde equipment and the Pangoal Institutional Array (PANGU) model, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 4 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0077] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a typhoon path prediction program 40 based on multimodal data from a radiosonde and the Pangoal Mechanism model. This typhoon path prediction program 40 based on multimodal data from a radiosonde and the Pangoal Mechanism model can be executed by the processor 10, thereby implementing the typhoon path prediction method based on multimodal data from a radiosonde and the Pangoal Mechanism model in this application.

[0078] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangu model.

[0079] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The terminal's processor 10, memory 20, and display 30 communicate with each other via a system bus.

[0080] In one embodiment, when the processor 10 executes the typhoon path prediction program 40 based on multimodal data from radiosonde devices and the Pangoal Mechanism model stored in the memory 20, the following steps are performed:

[0081] Acquire multi-source heterogeneous data collected by detection equipment related to typhoon forecasting, and fuse and adaptively enhance the multi-source heterogeneous data based on the Pango meteorological big model to generate a high-resolution three-dimensional field;

[0082] A cascaded AI downscaling network is constructed, a physical constraint layer is introduced, and the eye of the typhoon is locally refined based on the high-resolution three-dimensional field to output a high-resolution three-dimensional meteorological field.

[0083] Design a typhoon-environment field interaction model based on graph neural networks, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report.

[0084] The acquisition of multi-source heterogeneous data collected by detection equipment related to typhoon forecasting, and the fusion and adaptive enhancement of the multi-source heterogeneous data based on the Pangu meteorological big data model to generate a high-resolution three-dimensional field, specifically includes:

[0085] By accessing the global forecast data stream of the Pangu meteorological big data model, low-resolution grid data of the typhoon core area is extracted. The low-resolution grid data includes pressure field, three-dimensional wind field, humidity field and sea surface temperature field.

[0086] A typhoon feature space is constructed based on a historical typhoon database. The current typhoon type is identified through clustering algorithms, and key prediction areas are dynamically delineated.

[0087] The observation data from the detection equipment are synchronized. The observation data from the detection equipment includes: temperature profile data of the typhoon eye area collected by the satellite microwave imager, three-dimensional wind speed field collected by the UAV-borne millimeter-wave radar, sea temperature and air pressure data collected by the ocean buoy, and upper-air three-dimensional meteorological data collected by the airship.

[0088] Spatiotemporal kriging interpolation was used to map point observation data onto the Pango meteorological model, and nested high-resolution subgrids were established for the typhoon eye region.

[0089] A typhoon data generator based on a diffusion model is constructed. Low-resolution grid data and observation data from detection equipment are input into the typhoon data generator for fusion processing, and a complete high-resolution three-dimensional field of temperature, pressure and humidity in the typhoon eye region is output.

[0090] Deploy an anomaly detection module that uses an autoencoder to identify and remove noisy data observed by the detection device.

[0091] The construction of a cascaded AI downscaling network, introducing a physical constraint layer, and performing localized refined modeling of the typhoon eye region based on the high-resolution 3D field to output a high-resolution 3D meteorological field specifically includes:

[0092] A cascaded AI downscaling network is constructed, which includes a computer vision model and a three-dimensional convolutional neural network. The computer vision model is used to extract global meteorological features output by the Pangu meteorological model and input them into the three-dimensional convolutional neural network. The three-dimensional convolutional neural network is used to perform local fine modeling of the typhoon eye area and output a three-dimensional meteorological field based on the input satellite infrared cloud image and a database of historical similar typhoon cases.

[0093] A lightweight LSTM-ENKF hybrid assimilation algorithm was developed. The algorithm predicts the systematic bias distribution of the Pangoal Meteorological Model using LSTM. An ensemble Kalman filter is used to dynamically inject UAV observation data into the cascaded AI downscaling network, and a high-resolution three-dimensional meteorological field is output after data assimilation and physical constraint correction.

[0094] A physical constraint layer is introduced, and a partial differential equation verification module is added to the output of the cascaded AI downscaling network to ensure that the predicted field conforms to the fluid dynamics equations.

[0095] The design is based on a graph neural network-based typhoon-environmental field interaction model. Typhoon prediction is performed using the high-resolution three-dimensional meteorological field, outputting a typhoon path ensemble forecast including confidence intervals and a typhoon thermodynamic structure analysis report, specifically including:

[0096] Design a typhoon-environmental field interaction model based on graph neural networks, abstract the typhoon structure into a dynamic graph, input environmental field data, and predict the probability distribution of the typhoon's movement direction;

[0097] A deep reinforcement learning policy optimizer is constructed, which uses the historical typhoon track fit and physical rationality score as the reward function to output a set of typhoon track forecasts containing confidence intervals within a preset time period.

[0098] Develop a multi-task learning architecture to predict the maximum wind speed and minimum air pressure at the center of a typhoon and identify signs of sudden intensity changes.

[0099] Integrating interpretable AI components, the Spatial SHAP algorithm is used to visualize key areas affecting intensity and generate typhoon thermodynamic structure analysis reports.

[0100] The typhoon thermodynamic structure analysis report includes information such as warm core height, outflow layer intensity, vertical temperature lapse rate, specific humidity, tangential wind, and radial wind.

[0101] The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Rayet model further includes:

[0102] The TinyML architecture is deployed on the detection equipment to perform knowledge distillation on the cascaded AI downscaling network. The preprocessing process of meteorological data is standardized and modularized through hardware optimization.

[0103] Key structural indicators of the eye of the storm are uploaded to the cloud server at set intervals to support dynamic model optimization and prediction result generation in the cloud, and the prediction results are displayed on the visualization platform.

[0104] The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal-Rayet model further includes:

[0105] The prediction error of the typhoon forecast is obtained. When the prediction error exceeds a threshold, the Pangu meteorological model is triggered to perform model fine-tuning.

[0106] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model, and when the typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model is executed by a processor, it implements the steps of the typhoon path prediction method based on multimodal data from a radiosonde and the Pangoal model as described above.

[0107] In summary, this invention provides a method, system, terminal, and computer-readable storage medium for typhoon path prediction based on multimodal data from radiosonde equipment and the Pangoal Meteorological Model. The method includes: acquiring multi-source heterogeneous data collected by radiosonde equipment related to typhoon prediction; fusing and adaptively enhancing the multi-source heterogeneous data based on the Pangoal Meteorological Model to generate a high-resolution three-dimensional field; constructing a cascaded AI downscaling network, introducing a physical constraint layer, and performing localized refined modeling of the typhoon eye region based on the high-resolution three-dimensional field to output a high-resolution three-dimensional meteorological field; designing a typhoon-environmental field interaction model based on a graph neural network; performing typhoon prediction based on the high-resolution three-dimensional meteorological field; and outputting a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report. This invention combines the Pangoal Meteorological Model, artificial intelligence technology, and wind eye detection equipment to achieve accurate typhoon prediction, improving the speed and accuracy of typhoon path prediction, significantly reducing prediction time and error, and saving computational resources.

[0108] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0109] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0110] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for predicting typhoon paths based on multimodal data from radiosonde equipment and the Pangoal-Rayet model, characterized in that, The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Institutional Array (PANGU) model includes: Acquire multi-source heterogeneous data collected by detection equipment related to typhoon forecasting, and fuse and adaptively enhance the multi-source heterogeneous data based on the Pango meteorological big model to generate a high-resolution three-dimensional field; A cascaded AI downscaling network is constructed, a physical constraint layer is introduced, and the eye of the typhoon is locally refined based on the high-resolution three-dimensional field to output a high-resolution three-dimensional meteorological field. Design a typhoon-environment field interaction model based on graph neural networks, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report.

2. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Big Data Model as described in claim 1, characterized in that, The acquisition of multi-source heterogeneous data collected by detection equipment related to typhoon forecasting, and the fusion and adaptive enhancement of the multi-source heterogeneous data based on the Pango meteorological big data model to generate a high-resolution three-dimensional field, specifically includes: By accessing the global forecast data stream of the Pangu meteorological big data model, low-resolution grid data of the typhoon core area is extracted. The low-resolution grid data includes pressure field, three-dimensional wind field, humidity field and sea surface temperature field. A typhoon feature space is constructed based on a historical typhoon database. The current typhoon type is identified through clustering algorithms, and key prediction areas are dynamically delineated. The observation data from the detection equipment are synchronized. The observation data from the detection equipment includes: temperature profile data of the typhoon eye area collected by the satellite microwave imager, three-dimensional wind speed field collected by the UAV-borne millimeter-wave radar, sea temperature and air pressure data collected by the ocean buoy, and upper-air three-dimensional meteorological data collected by the airship. Spatiotemporal kriging interpolation was used to map point observation data onto the Pango meteorological model, and nested high-resolution subgrids were established for the typhoon eye region. A typhoon data generator based on a diffusion model is constructed. Low-resolution grid data and observation data from detection equipment are input into the typhoon data generator for fusion processing, and a complete high-resolution three-dimensional field of temperature, pressure and humidity in the typhoon eye region is output. Deploy an anomaly detection module that uses an autoencoder to identify and remove noisy data observed by the detection device.

3. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Big Data Model as described in claim 1, characterized in that, The construction of a cascaded AI downscaling network, introducing a physical constraint layer, and performing localized refined modeling of the typhoon eye region based on the high-resolution 3D field, outputting a high-resolution 3D meteorological field, specifically includes: A cascaded AI downscaling network is constructed, which includes a computer vision model and a three-dimensional convolutional neural network. The computer vision model is used to extract global meteorological features output by the Pangu meteorological model and input them into the three-dimensional convolutional neural network. The three-dimensional convolutional neural network is used to perform local fine modeling of the typhoon eye area and output a three-dimensional meteorological field based on the input satellite infrared cloud image and a database of historical similar typhoon cases. A lightweight LSTM-ENKF hybrid assimilation algorithm was developed. The algorithm predicts the systematic bias distribution of the Pangoal Meteorological Model using LSTM. An ensemble Kalman filter is used to dynamically inject UAV observation data into the cascaded AI downscaling network, and a high-resolution three-dimensional meteorological field is output after data assimilation and physical constraint correction. A physical constraint layer is introduced, and a partial differential equation verification module is added to the output of the cascaded AI downscaling network to ensure that the predicted field conforms to the fluid dynamics equations.

4. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Big Data Model as described in claim 1, characterized in that, The design is based on a graph neural network-based typhoon-environmental field interaction model. Typhoon prediction is performed using the high-resolution three-dimensional meteorological field, outputting a typhoon track ensemble forecast including confidence intervals and a typhoon thermodynamic structure analysis report, specifically including: Design a typhoon-environmental field interaction model based on graph neural networks, abstract the typhoon structure into a dynamic graph, input environmental field data, and predict the probability distribution of the typhoon's movement direction; A deep reinforcement learning policy optimizer is constructed, which uses the historical typhoon track fit and physical rationality score as the reward function to output a set of typhoon track forecasts containing confidence intervals within a preset time period. Develop a multi-task learning architecture to predict the maximum wind speed and minimum air pressure at the center of a typhoon and identify signs of sudden intensity changes. Integrating interpretable AI components, the Spatial SHAP algorithm is used to visualize key areas affecting intensity and generate typhoon thermodynamic structure analysis reports.

5. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Big Data Model according to claim 4, characterized in that, The typhoon thermodynamic structure analysis report includes information on warm core height, outflow layer intensity, vertical temperature lapse rate, specific humidity, tangential wind, and radial wind.

6. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Big Data Model according to claim 3, characterized in that, The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Institutional Array (PIA) model also includes: The TinyML architecture is deployed on the detection equipment to perform knowledge distillation on the cascaded AI downscaling network. The preprocessing process of meteorological data is standardized and modularized through hardware optimization. Key structural indicators of the eye of the storm are uploaded to the cloud server at set intervals to support dynamic model optimization and prediction result generation in the cloud, and the prediction results are displayed on the visualization platform.

7. The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Big Data Model according to claim 6, characterized in that, The typhoon path prediction method based on multimodal data from radiosonde equipment and the Pangoal Institutional Array (PIA) model also includes: The prediction error of the typhoon forecast is obtained. When the prediction error exceeds a threshold, the Pangu meteorological model is triggered to perform model fine-tuning.

8. A typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal-Rayet model, characterized in that, The typhoon path prediction system based on multimodal data from radiosonde equipment and the Pangoal Institutional Array (PANGU) model includes: The multi-source heterogeneous data fusion and adaptive enhancement module is used to acquire multi-source heterogeneous data collected by detection equipment related to typhoon prediction, and to fuse and adaptively enhance the multi-source heterogeneous data based on the Pangu meteorological big model to generate a high-resolution three-dimensional field. The dynamic downscaling and physical constraint correction module is used to construct a cascaded AI downscaling network, introduce a physical constraint layer, perform local fine modeling of the typhoon eye region based on the high-resolution three-dimensional field, and output a high-resolution three-dimensional meteorological field. The multimodal typhoon evolution prediction and report generation module is used to design a typhoon-environmental field interaction model based on graph neural networks, perform typhoon prediction based on the high-resolution three-dimensional meteorological field, and output a typhoon path set forecast including confidence intervals and a typhoon thermodynamic structure analysis report.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model, stored in the memory and executable on the processor. When the processor executes the typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model, it implements the steps of the typhoon path prediction method based on multimodal data from a radiosonde and the Pangoal model as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model. When the typhoon path prediction program based on multimodal data from a radiosonde and the Pangoal model is executed by a processor, it implements the steps of the typhoon path prediction method based on multimodal data from a radiosonde and the Pangoal model as described in any one of claims 1-7.

Images