Systematic tropical cyclone and high-resolution wind field modeling method, system and equipment based on environment driving and medium

Abstract

The invention belongs to the technical field of meteorological simulation, discloses a systematic tropical cyclone and high-resolution wind field modeling method, system and device based on environment driving and a medium, and aims to solve the problem of wind field simulation distortion in complex terrains. The method comprises the following steps: acquiring and unifying environment field data; calculating a dynamic generation potential index field based on the base month average climate field, defining a space-time generation probability, and randomly sampling to generate an initial event set of a plurality of synthetic tropical cyclone events; performing iterative simulation on each synthetic tropical cyclone event, and outputting full life cycle information; aiming at the full life cycle information, based on a gradient wind balance physical model, calculating each position point in the moving track information and a corresponding central air pressure difference value, and obtaining a near-surface wind field under an open terrain assumption as a first near-surface wind field; and acquiring high-resolution terrain data, inputting the high-resolution terrain data and the first near-surface wind field into the terrain adaptive correction model, and outputting the corrected near-surface wind field as a second near-surface wind field.

Description

Technical Field

[0001] This invention belongs to the field of meteorological simulation technology, specifically relating to methods, systems, equipment, and media for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches. Background Technology

[0002] Tropical cyclones (TCs) are among the major natural disasters affecting coastal areas, posing a serious threat to infrastructure, socio-economic development, and public safety. Conducting accurate tropical cyclone disaster risk assessments is crucial for disaster prevention and mitigation, emergency management, and long-term planning.

[0003] Currently, the industry mainly relies on three types of models for tropical cyclone risk analysis. The first type is purely statistical models, such as statistical-dynamic combined models. These models primarily use historical observation data for regression analysis to establish statistical relationships to predict tropical cyclone activity. However, these models lack a detailed characterization of specific physical processes, and their intrinsic relationships may change under the context of climate change, limiting their long-term applicability and extrapolation capabilities. The second type is parametric wind models (PWM). These models are based on gradient wind balance and boundary layer theory, offering high computational efficiency and are widely used in operational risk assessment. However, their core drawback is that they typically assume a homogeneous, open terrain (such as an open sea), failing to accurately simulate the local effects of complex terrain such as mountains, hills, and urban complexes on wind fields, including acceleration, bypassing, and obstruction. Therefore, in areas with complex terrain, the wind field simulation results of this type of model may have significant biases. The third type is high-precision numerical models, such as Weather Research and Forecasting Models (WRF). These models can simulate the interaction between wind fields and topography with high accuracy by directly solving atmospheric dynamic and thermodynamic equations. However, their computational cost is extremely high; simulating a single tropical cyclone event typically requires hundreds or even thousands of CPU hours of computing resources. Such a high computational cost makes these models unsuitable for generating event sets containing tens of thousands of synthetic tropical cyclone events, which are essential for reliable probabilistic risk assessments.

[0004] In summary, existing technologies suffer from a fundamental deficiency: the lack of an end-to-end tropical cyclone simulation framework that can maintain consistency in physical mechanisms, operate efficiently, and accurately characterize the fine-grained structure of wind fields under complex terrain. Furthermore, many existing models are primarily developed for the Atlantic basin and lack sufficient localization for the Western North Pacific (WNP), one of the most active regions for tropical cyclone activity globally; the applicability of their parameters and mechanisms in this region needs further optimization. Summary of the Invention

[0005] Based on the aforementioned shortcomings and deficiencies in the existing technology, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the existing technology. In other words, one of the objectives of this invention is to provide an environment-driven systematic tropical cyclone and high-resolution wind field modeling method, system, device and medium that meets one or more of the aforementioned requirements, aiming to achieve integrated simulation of the entire life cycle of tropical cyclones from generation, movement, intensity evolution to near-surface wind fields, and to solve the problem of wind field simulation distortion under complex terrain.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches, comprising the following steps: S1. Acquire and unify the spatiotemporal resolution of environmental field data to form a basic dataset. The environmental field data includes wind components, vertical velocity, relative vorticity, sea surface temperature, and relative humidity of the multi-layer atmosphere. S2. Based on the monthly average climate field in the basic dataset, calculate the dynamic generation potential index field and define the spatiotemporal generation probability distribution of tropical cyclones based on this. Generate an initial event set of multiple synthetic tropical cyclone events by random sampling. The initial record of each synthetic tropical cyclone event includes the initial location, initial time and initial intensity. S3. For each synthetic tropical cyclone event in the initial event set, perform an iterative simulation process until a preset termination condition is met. The iterative simulation process includes: S31. At each simulation time step, based on the monthly average wind field in the basic dataset and the preset drift parameterization scheme, calculate the moving speed of the event at the current center position and update its center position to generate the moving trajectory information of the event. S32. Along the path determined by the movement trajectory information, at the preset intensity prediction time interval, environmental variables are extracted from the basic dataset by interpolation, and the predicted intensity value is calculated by the intensity prediction model that includes physical factors and random disturbances, and then converted into the corresponding center pressure difference value. Through the iterative simulation process, the full life cycle information of each synthetic tropical cyclone event is output. The full life cycle information includes at least the movement trajectory information determined by the path simulation, the intensity sequence determined by the intensity simulation, and the central pressure difference sequence. S4. For the full life cycle information of each synthetic tropical cyclone event, based on the gradient wind balance physical model, calculate the location point in the movement trajectory information and its corresponding central pressure difference to obtain the near-surface wind field under the open terrain assumption, which is used as the first near-surface wind field. S5. Acquire high-resolution terrain data, input the first near-surface wind field and the high-resolution terrain data into a pre-trained terrain adaptive correction model, the deep learning model outputs a near-surface wind field corrected by terrain effect as the second near-surface wind field, and the spatial resolution of the second near-surface wind field is higher than that of the first near-surface wind field.

[0007] As a preferred embodiment, the formula for calculating the dynamic generation potential index of a grid point in the dynamic generation potential index field is as follows: , In the formula, Indicates the first Month in OK Dynamic generation potential index on column grid points The vertical wind shear at this grid point This represents the meridional gradient of the 500 hPa zonal wind at this grid point. The vertical velocity at this grid point is 500 hPa. The absolute vorticity at this grid point is 850 hPa.

[0008] As a preferred option, in step S31: The preset drift parameterization scheme is the Beta drift scheme; The moving speed is a composite of the large-scale environmental guiding airflow and the Beta drift wind speed term; The large-scale environmental guiding airflow is obtained by weighted synthesis of wind vectors at 850 hPa and 250 hPa and superimposed with random perturbations generated based on the statistical characteristics of the monthly average wind field.

[0009] As a preferred option, in step S32: The environmental variables include potential intensity, deep vertical wind shear, mid-level relative humidity, tropical cyclone movement speed, and the change in the land area ratio between the current point and the previous time point. The intensity prediction model is an autoregressive model. Its deterministic linear prediction term is a multiple linear regression model for current intensity, intensity change, movement speed, difference between current and potential intensity, vertical wind shear, mid-level relative humidity, and change in land area. Different regression coefficients are used for sea and land.

[0010] As a preferred approach, the calculation of the first near-surface wind field based on the gradient wind balance physical model in step S4 is specifically as follows: Using the Holland model with the center of the tropical cyclone as the origin, the gradient wind field is obtained by iteratively solving the gradient wind equation based on the central pressure difference, the maximum wind speed radius, the Holland B parameter, the Coriolis parameter, and the moving speed. Then, the gradient wind field is reduced to the near-surface wind field at a height of 10 meters through the boundary layer model.

[0011] As a preferred option, in step S5: The terrain adaptive correction model is a Transformer neural network based on a U-shaped encoder-decoder architecture; The inputs to the terrain adaptive correction model include wind speed scalar and U / V wind speed components derived from the first near-surface wind field, as well as elevation and land use type information derived from the high-resolution terrain data. The terrain adaptive correction model is obtained by training with high-precision numerical model simulation results as ground truth.

[0012] As a preferred option: The basic feature transformation unit of the Transformer neural network is a local enhancement window transformation block; The local enhancement window transformation block sequentially includes layer normalization, a multi-head self-attention mechanism based on local windows, and a feedforward network containing depthwise convolution operations.

[0013] Secondly, the present invention provides an environment-driven system for modeling tropical cyclones and high-resolution wind fields, for implementing the modeling method described in the first aspect.

[0014] Thirdly, the present invention provides an electronic device, the computer device including a memory, a processor and a computer program, wherein the computer program, when executed by the processor, implements the modeling method as described in the first aspect.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the modeling method as described in the first aspect.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention creates an end-to-end tropical cyclone (TC) simulation system that is entirely driven by environmental data and combines physical processes with statistical models. It achieves seamless connection and simulation from large-scale climate fields to local refined wind disaster risks, overcoming the single shortcomings of traditional methods such as lack of physical mechanisms, excessively high computational costs, or inability to adapt to complex terrain.

[0017] 2. In the intensity prediction model, key physical factors such as potential intensity (PI), vertical wind shear (SHRD), and mid-level humidity (midRH) are innovatively introduced. By adopting land-sea regression coefficients and random disturbance terms that conform to a normal distribution, the intensity simulation is significantly improved, especially the simulation capability of rapid intensification (RI) processes.

[0018] 3. Through the proposed Terrain Adaptive Wind Field Correction Module (TTWM), an innovative approach is taken to apply the advanced U-former deep learning architecture to downscaling and bias correction of meteorological wind fields. This module can effectively learn and reproduce the acceleration, bypass, and blocking effects of complex terrain (such as mountains, hills, and urban clusters) on wind fields, reducing the mean absolute error (MAE) of parameterized wind fields in complex terrain areas by approximately 30%-50%, with a particularly significant improvement in the accuracy of wind direction simulation.

[0019] 4. While maintaining the core physical mechanisms and high simulation accuracy, the computational efficiency of this invention is far superior to that of pure numerical simulations (such as the WRF model). This makes it possible to generate synthetic tropical cyclone event sets with timescales of tens of thousands of years for reliable probabilistic risk assessment, solving the problem that high-precision models cannot be used for large-scale risk analysis.

[0020] 5. The system adopts a modular design, with each module (generation, path, intensity, wind field, and terrain correction) being relatively independent and capable of being improved or replaced individually (e.g., by integrating more advanced path models). It can also easily integrate climate model data from different sources (such as CMIP6) and be directly used for risk analysis under future climate scenarios, thus possessing future-oriented adaptability.

[0021] 6. From the Generation Index (DGPI) and Beta drift parameterization to the intensity model coefficients, all parameters have been specifically optimized and calibrated for the Northwest Pacific (WNP), the most active region for tropical cyclones globally, significantly improving the accuracy and reliability of simulations in this region. Furthermore, by recalibrating key parameters, this framework can be extended to other active tropical cyclone sea areas worldwide.

[0022] 7. The entire simulation framework is highly automated, with controllable computational costs, and can support operational use. Its high-resolution, terrain-adaptive wind field output can be directly used as reliable input for infrastructure failure analysis, vulnerability assessment, and disaster risk pricing in industries such as power (transmission towers), new energy (wind turbines), building structures, transportation, and insurance. It provides high-quality technical tools and disaster scenario libraries for disaster prevention and mitigation, emergency management, and long-term planning.

[0023] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the modeling method described in Embodiment 1 of the present invention.

[0026] Figure 2 This is a schematic diagram of the terrain adaptive correction model TTWM described in Embodiment 1 of the present invention.

[0027] Figure 3 This is a schematic diagram of the structure of the local enhancement window transformation block described in Embodiment 1 of the present invention.

[0028] Figure 4 This is a structural diagram of the electronic device provided in the embodiment of the present invention.

[0029] Icon labels: 400. Electronic devices; 401. Processor; 402. Communication bus; 403. User interface; 404. Network interface; 405. Memory. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0031] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0032] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0033] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0034] The modeling method described in this specification is applied to scenarios requiring refined and quantitative risk assessment of wind disasters caused by tropical cyclones (typhoons). In these scenarios, the application of this modeling method aims to provide a set of efficient, accurate, and operationally feasible technical tools. Its core objective is to overcome the single deficiency of traditional methods in terms of efficiency, accuracy, or physical consistency. Through full-chain, environment-driven simulation, it generates large-scale, high-resolution, and terrain-adaptive synthetic wind field datasets, thereby supporting decision-making needs at all levels, from macro-risk mapping to micro-engineering-level load analysis, ultimately enhancing societal resilience in the face of tropical cyclone disasters.

[0035] The following is a brief explanation of the environmental field data, Dynamic Generation Potential Index (DGPI), Parametric Wind Field Model (PWM), Terrain Adaptive Wind Field Correction Module (TTWM), U-former, and Local Enhancement Window Transformation Block involved in several embodiments of this specification: Environmental field data: refers to the large-scale meteorological and underlying surface data set used to drive and constrain tropical cyclone simulations. In this invention, it specifically refers to data preprocessed with uniform spatiotemporal resolution from reanalysis data (such as ERA5) or climate model outputs (such as CMIP), mainly including variables such as wind speed / direction, temperature, humidity, vertical velocity, vorticity, and sea surface temperature (SST) at multiple atmospheric levels (such as 850hPa, 500hPa, 250hPa, etc.).

[0036] The Dynamic Generation Potential Index (DGPI) is a dimensionless index that integrates multiple large-scale environmental factors (vertical wind shear, zonal and meridional wind gradients, vertical velocity, and low-level absolute vorticity) to quantitatively characterize the climate probability of tropical cyclone formation at a specific spatiotemporal point (grid point and month). A higher DGPI value indicates more favorable formation conditions.

[0037] Parametric wind field model (PWM): This refers to a simplified wind field model built based on physical principles such as gradient wind balance and boundary layer theory. It can quickly calculate the near-surface wind field around a tropical cyclone, which is approximately circularly symmetrical and does not consider the effects of complex terrain, based on the center location, intensity (central pressure difference), and scale parameters of the tropical cyclone. It has high computational efficiency but insufficient ability to simulate the influence of terrain.

[0038] Terrain Adaptive Wind Field Correction Module (TTWM): This is one of the core innovative modules of this invention. It is a deep learning-based model (specifically a U-shaped converter architecture) whose function is to combine the low-resolution, terrain-ignoring "coarse" wind field output by PWM with high-precision terrain data to correct and downscale it into a high-resolution, "refined" wind field that can accurately reflect the influence of terrain such as mountains and buildings.

[0039] The U-former is a deep learning network architecture used for image processing or similar grid-like data processing. It is named for its encoder-decoder structure, which resembles the letter "U". It excels at capturing multi-scale features of data and fuses shallow details with deep semantic information through skip connections. In this invention, it is used to learn the complex nonlinear mapping relationship between wind fields and terrain.

[0040] The Local Enhancement Window Transformer Block (LEB) is the basic feature processing unit that constitutes the U-shaped transformer network described above. Building upon the standard transformer's self-attention mechanism, it introduces local window partitioning and offset window mechanisms to reduce computational complexity. Furthermore, it incorporates deep convolution operations into the feedforward network to enhance the model's ability to capture local spatial features (such as the local interaction between terrain and wind).

[0041] Example 1: This embodiment provides a systemic tropical cyclone and high-resolution wind field modeling method based on environment-driven approaches. For example... Figure 1 As shown, this method is a serial, automated simulation chain, and its overall workflow includes the following steps: Step 1, Data Preparation and Preprocessing: All environmental field data used to drive the model must undergo uniform preprocessing before the simulation begins to eliminate differences in spatiotemporal resolution of the original data and ensure consistency of inputs to subsequent modules.

[0042] The data sources used in this step are ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) or output data from the Global Climate Model (GCM) in the Coupled Model Intercomparison Programme (CMIP). Specifically, this includes the data described in Table 1.

[0043] Table 1:

[0044] Preprocessing specifically includes two aspects: time processing and space processing. 1. Time processing: The time resolution of most variables is unified to monthly averages. However, the calculation of steering airflow used for path simulation requires additional daily average wind field data.

[0045] 2. Spatial Processing: All data are resampled to a unified latitude and longitude grid using bilinear interpolation. In this embodiment, a spatial resolution of 0.5° × 0.5° is used. This step ensures that all environmental variables are within the same spatial framework.

[0046] After the above processing in step one, a spatiotemporally consistent basic dataset is formed for all subsequent modules to call.

[0047] Step 2, Tropical Cyclone Formation Simulation: This step aims to simulate the formation event of a tropical cyclone (TC), that is, to determine when, where, and with what initial intensity it will occur.

[0048] First, based on the preprocessed monthly mean climate field data, a field called the Dynamic Generation Potential Index (DGPI) is calculated. The specific calculation formula is as follows: , In formula (1): Indicates the first Month in OK The dynamic generation potential index for grid points is a dimensionless number. The larger the value, the higher the probability of a tropical cyclone (TC) forming at that grid point in that month. The vertical wind shear at this grid point is calculated using the following formula: (2), in equation (2) , , and The zonal and meridional wind components at the 200hPa and 850hPa levels are respectively taken from the preprocessed monthly average data; The meridional gradient of the 500 hPa zonal wind at this grid point is calculated along the meridional direction using the central difference method. (3), in equation (3) It is the meridional distance; The vertical velocity at 500 hPa for this grid point is taken directly from the monthly average data. A negative value indicates upward motion, which is conducive to the development of convection. The absolute vorticity at 850 hPa for this grid point is calculated as follows: First, the relative vorticity at 850 hPa is calculated. (4), in equation (4) This is the latitudinal distance. All partial derivatives are obtained using the central difference method on a uniform latitude and longitude grid, and then the Coriolis parameter is added. (5), in equation (5) The angular velocity of Earth's rotation. Using latitude, we ultimately obtain the absolute vorticity. .

[0049] Next, the DGPI field is converted into a non-negative generation probability density field using an exponential function. A target annual average generation count for the simulation region (e.g., the Northwest Pacific) is set (e.g., 28 events / year). For each year of the simulation, random sampling is performed based on this probability density field to sequentially determine the generation grid for each TC event and its precise latitude and longitude location within that grid (resolution 0.5° × 0.5°). The specific date and time of generation It randomly assigns an initial intensity from a preset initial intensity distribution (such as a uniform distribution of 15kt-35kt). .

[0050] Step two's final output is a set of initial states containing multiple TC events. Each event records its initial position, initial time, and initial intensity, denoted as { }

[0051] Step 3, Tropical Cyclone Path Simulation: This step simulates the complete trajectory of each generated TC event from its generation to its dissipation using hourly integration.

[0052] At each integration time step t TC's movement speed Airflow guided by large-scale environment With Beta Drift Item The synthesis, and its calculation formula is: (6), In formula (6): This represents the steering airflow on a large scale, and its calculation includes the following steps: First, from the preprocessed daily average wind field data, bilinear interpolation is used to obtain the current TC center position. Wind vectors at 850 hPa and 250 hPa and According to weighting coefficients and The composite steering airflow is obtained by weighting and combining the wind vectors at the two levels mentioned above. (7); In order to simulate the instantaneous fluctuations at the weather scale, a random perturbation field is superimposed on the daily average wind field. The perturbation field is synthesized by the variance-covariance matrix of the monthly average wind field through the Cholesky decomposition method to maintain its spatiotemporal statistical characteristics. The drift wind speed term is calculated using a parameterization scheme applicable to the western Pacific: (8), (9), in equations (8) and (9) It is the latitude of the TC center point at the current time step.

[0053] Obtain total movement speed Update the center position of TC using the spherical coordinate formula: (10) (11), In equations (10) and (11), =3600 seconds (1 hour) It is the Earth's radius (approximately 6,371,000 meters).

[0054] Repeat the hourly integration process described above to form a simulation loop. The loop terminates when any of the following conditions are met: 1. The center of the TC is moved out of the preset simulation area; 2. The TC intensity decreased to below 10 kt; 3. The simulation duration reaches the preset maximum lifespan (e.g., 30 days).

[0055] This step outputs the complete trajectory sequence for each TC, which includes the center position, time, and speed information for each time step, denoted as {( )}.

[0056] Step 4, Tropical Cyclone Intensity Simulation: This step predicts the change in maximum sustained wind speed (intensity) over time along the simulated TC movement trajectory.

[0057] First, for each predicted point on the trajectory (e.g., one point every 12 hours), key environmental variables are extracted from the base dataset through interpolation, including: 1. Potential intensity is calculated by calling a standard algorithm library based on the thermodynamic equations of the air-sea interface, with the input being the vertical profiles of sea surface temperature and atmospheric temperature and humidity.

[0058] 2. 800hPa-200hPa deep vertical wind shear, calculated using the same method as the generation module. .

[0059] 3. The relative humidity in the middle layer at 500hPa-300hPa is the arithmetic mean of the relative humidity at 500hPa and 300hPa.

[0060] 4. Movement speed, calculated from the velocity vector output by the path module: .

[0061] 5. Changes in land area proportion. Calculate the land area proportion within a circular area with a radius of 200 kilometers, using the center of TC as the origin. (Pixel statistics are performed using a 0.5° resolution land-ocean mask grid), then .

[0062] Then, an autoregressive (AR) model is used to predict intensity changes, and the prediction equation is: (12), In equation (12): The intensity to be predicted for the next 12 hours is expressed in kt. The intensity of the current and previous time steps; The deterministic linear predictor is a multiple linear regression model, specifically in the form of:

[0063] (13) Let be a random disturbance term, from a value of 0 and a standard deviation of . The normal distribution N(0, Random sampling in ) with standard deviation This is determined by fitting the residual sequence of historical tropical cyclone data and deterministic linear forecasts, for example, the open ocean surface of the western Pacific. 4.5kt is acceptable.

[0064] Finally, the predicted wind speed intensity Core pressure differential required for conversion to wind farm module (Unit: hPa). Using the empirical conversion formula for the western Pacific region: When the center point of TC is located over the ocean (14) When the center point of TC is located on land, (15) In equations (14) and (15), The unit is hPa. The unit is kt.

[0065] This step ultimately outputs the intensity sequence of each TC on the trajectory every 12 hours. } and central pressure difference sequence {}

[0066] Step 5, Parametric wind field calculation: This step calculates the near-surface wind field at a height of 10 meters under the assumption of open terrain, based on the state of each tropical cyclone (TC) at each moment.

[0067] First, a circular computing region with a radius of 500 kilometers is established with the center of TC as the origin.

[0068] Subsequently, for each grid point within the computational region, its distance to the center of TC is calculated. and azimuth Then, the gradient wind speed is calculated point by point using a gradient wind balance model (specifically the Holland model in this embodiment). and gradient wind direction The calculation formula is as follows: (16) (16) In equations (15) and (16): For gradient wind speed; The gradient wind direction is defined with due north as 0°, increasing clockwise. To calculate the distance from the point to the center of TC; This refers to the movement speed of TC. The angle between the vector pointing from the calculation point to the center of TC and the direction of TC's movement; For the Coriolis parameter, calculate using the reference formula (5); The central pressure difference originates from the strength module; For air density, take a constant of 1.15; The radius of maximum wind speed is estimated using empirical formulas, for example: (17), coefficient Obtained by fitting historical data; Holland B parameters are estimated using empirical formulas, for example: ,coefficient Obtained by fitting historical data; The inflow angle describes the degree to which the wind is directed towards the center; it is a function of r / RMW and is determined by consulting an empirical table.

[0069] Since the gradient wind equation (16) is about The implicit equations need to be solved numerically using iterative methods (such as the Newton-Raphson method), setting initial iteration values ​​(such as 50 m / s) and convergence tolerances (such as 0.01 m / s).

[0070] Finally, the calculated gradient wind field is transformed into a near-surface wind field at a height of 10 meters using a boundary layer model: (19) (20) In equations (19) and (20): The surface wind speed at a height of 10 meters; The surface wind direction is at a height of 10 meters. The boundary layer height is taken as a constant of 1000 meters; This is the surface roughness length. For open sea surfaces, it is taken as 0.003m. For land, it is determined by referring to a table based on the land use type. The gradient height is set to 1000 meters. These are the boundary layer model coefficients, and is a constant, which can be found in relevant literature.

[0071] This step outputs the surface wind speed and direction field of each TC at each time step within the calculation area (e.g., 12.5 km resolution), as the first near-surface wind field.

[0072] Step 6, Terrain-Adaptive Wind Field Correction: This step corrects the first near-surface wind field (coarse resolution, such as 12.5km) obtained in step five to a high-resolution (such as 500m) terrain-adaptive second near-surface wind field. The calculation is completed through a deep learning model - Terrain Adaptive Wind Field Correction Module (TTWM).

[0073] The Terrain Adaptive Modification Model (TTWM) is a deep learning model based on a U-shaped transformer architecture. Its structure includes an encoder and a decoder, with skip connections between corresponding layers of the encoder and decoder, such as... Figure 2 As shown.

[0074] The model takes 5 channels of data as input, and its specific structure is as follows: 1. Wind speed scalar (calculated from the U / V components of the first near-surface wind field); 2. U-wind speed component; 3. V wind speed component; 4. Elevation data (from a high-resolution digital elevation model); 5. Land use type data.

[0075] The above input data has been uniformly cropped to: × The data is processed into a 5×5 tensor (corresponding to a 12.5km resolution) within a specific region. Before being input into the model, all channels of data are standardized to eliminate dimensional differences and accelerate model training convergence.

[0076] The training of the model is completed through the following process: 1. Training Dataset Construction: Select 4-5 representative historical tropical cyclone events (such as Tropical Cyclones "Lekima" and "Mufa"). Use the Weather Research and Forecasting Model (WRF) for high-precision numerical simulation, configuring a triple-nested grid with the innermost grid resolution reaching 500 meters. Use the 10-meter U / V wind speed components obtained from the innermost WRF simulation as the target ground truth for model training.

[0077] 2. Input data generation: The wind field simulated by WRF at the outermost layer (e.g., 12.5 km resolution), or the wind field calculated using the patented PWM module at 12.5 km resolution, is combined with the corresponding high-resolution terrain data as the input to the model.

[0078] 3. Sample Processing and Segmentation: Wind field and topographic data from the innermost region during each tropical cyclone process are cropped into multiple data sets of varying sizes. × The sample pairs consist of a 5-channel, 5×5 input tensor and a 2-channel, 100×100 ground truth tensor. All samples are randomly divided into training and validation sets at approximately 85% and 15% ratios, respectively.

[0079] 4. Model training: Using the Charbonnier loss function ( ,in It is the truth value of the target. These are model predictions. It is a small constant (let it be) This measures the error between the prediction and the true value. The Adam optimizer is used (parameters set to: =0.9, The algorithm is optimized with a learning rate of 0.999 and a weight decay coefficient of 0.02, and a cosine annealing learning rate scheduling strategy is adopted (initial learning rate 2×10). 4 The final learning rate is 1×10⁻⁶. 6 The batch size is set to 32, and training is typically performed for 500 epochs until the validation set loss no longer decreases significantly.

[0080] The core architecture and components of the model are as follows: 1. U-shaped transformer network: The model adopts an encoder-decoder structure. The encoder contains... Each stage (e.g.) =3), each stage consists of a downsampling layer and a feature transformation block, progressively extracting and compressing features. The decoder is symmetrical to the encoder, and each stage consists of an upsampling layer and a feature transformation block, progressively restoring spatial resolution and fusing it with the corresponding features from the encoder. The model's input and output are projected through convolutional layers.

[0081] 2. Local Enhancement Window Transform Block: As a basic feature transformation unit, its structure is as follows: Figure 3 As shown, it includes, in sequence: 3. Layer normalization: Standardize the input features.

[0082] 4. Window-based multi-head self-attention: Dividing the feature map into non-overlapping local windows (e.g., ... , =8), multi-head self-attention is computed within each window to capture long-range dependencies; and regular window and offset window partitions are alternately used in successive transform blocks to establish cross-window connections.

[0083] 5. Locally Enhanced Feedforward Network: Introducing depthwise separable convolution operations into the standard feedforward network enhances the model's ability to capture local neighborhood features, which is crucial for learning the local interaction between wind field and terrain.

[0084] After the model training is complete, it is used for wind field correction of new synthetic tropical cyclone events: 1. Input Preparation: For each first near-surface wind field output in step five, crop it in the exact same way as the training data. × The region is processed into a 5-channel, 5×5 input tensor and then subjected to the same normalization process.

[0085] 2. Model Inference: The prepared input tensor is fed into the trained TTWM model. The model operates via forward propagation, without the need to compute gradients.

[0086] 3. Post-processing and output: The 100×100 wind field tensor output by the model is denormalized to restore its physical quantity values, which yields the final 500-meter resolution, terrain-corrected U / V wind speed component field, denoted as the second near-surface wind field.

[0087] The second near-surface wind field output in this step is the final result of this invention. It has high resolution and accurately reflects the impact of complex terrain, and can be directly used for high-precision disaster risk analysis.

[0088] Example 2: This embodiment provides an environment-driven system for modeling tropical cyclones and high-resolution wind fields, used to implement the modeling method described in Embodiment 1.

[0089] Example 3: like Figure 4 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0090] The communication bus can be used to enable communication between the various components mentioned above.

[0091] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0092] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0093] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0094] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a modeling application program. The processor can be used to call the modeling application program stored in the memory and execute the steps of the modeling methods mentioned in the foregoing embodiments.

[0095] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figure 1 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0096] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0097] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0098] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0099] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0100] The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A systemic tropical cyclone and high-resolution wind field modeling method based on environment-driven approaches, characterized in that, Including the following steps: S1. Acquire and unify the spatiotemporal resolution of environmental field data to form a basic dataset. The environmental field data includes wind components, vertical velocity, relative vorticity, sea surface temperature, and relative humidity of the multi-layer atmosphere. S2. Based on the monthly average climate field in the basic dataset, calculate the dynamic generation potential index field and define the spatiotemporal generation probability distribution of tropical cyclones based on this. Generate an initial event set of multiple synthetic tropical cyclone events by random sampling. The initial record of each synthetic tropical cyclone event includes the initial location, initial time and initial intensity. S3. For each synthetic tropical cyclone event in the initial event set, perform an iterative simulation process until a preset termination condition is met. The iterative simulation process includes: S31. At each simulation time step, based on the monthly average wind field in the basic dataset and the preset drift parameterization scheme, calculate the moving speed of the event at the current center position and update its center position to generate the moving trajectory information of the event. S32. Along the path determined by the movement trajectory information, at the preset intensity prediction time interval, environmental variables are extracted from the basic dataset by interpolation, and the predicted intensity value is calculated by the intensity prediction model that includes physical factors and random disturbances, and then converted into the corresponding center pressure difference value. Through the iterative simulation process, the full life cycle information of each synthetic tropical cyclone event is output. The full life cycle information includes at least the movement trajectory information determined by the path simulation, the intensity sequence determined by the intensity simulation, and the central pressure difference sequence. S4. For the full life cycle information of each synthetic tropical cyclone event, based on the gradient wind balance physical model, calculate the location point in the movement trajectory information and its corresponding central pressure difference to obtain the near-surface wind field under the open terrain assumption, which is used as the first near-surface wind field. S5. Acquire high-resolution terrain data, input the first near-surface wind field and the high-resolution terrain data into a pre-trained terrain adaptive correction model, the deep learning model outputs a near-surface wind field corrected by terrain effect as the second near-surface wind field, and the spatial resolution of the second near-surface wind field is higher than that of the first near-surface wind field.

2. The method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches according to claim 1, characterized in that, The formula for calculating the dynamic generation potential index of a grid point in the dynamic generation potential index field is as follows: ， In the formula, Indicates the first Month in OK Dynamic generation potential index on column grid points The vertical wind shear at this grid point This represents the meridional gradient of the 500 hPa zonal wind at this grid point. The vertical velocity at this grid point is 500 hPa. The absolute vorticity at this grid point is 850 hPa.

3. The method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches according to claim 1, characterized in that, In step S31: The preset drift parameterization scheme is the Beta drift scheme; The moving speed is a composite of the large-scale environmental guiding airflow and the Beta drift wind speed term; The large-scale environmental guiding airflow is obtained by weighted synthesis of wind vectors at 850 hPa and 250 hPa and superimposed with random perturbations generated based on the statistical characteristics of the monthly average wind field.

4. The method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches according to claim 1, characterized in that, In step S32: The environmental variables include potential intensity, deep vertical wind shear, mid-level relative humidity, tropical cyclone movement speed, and the change in the land area ratio between the current point and the previous time point. The intensity prediction model is an autoregressive model. Its deterministic linear prediction term is a multiple linear regression model for current intensity, intensity change, movement speed, difference between current and potential intensity, vertical wind shear, mid-level relative humidity, and change in land area. Different regression coefficients are used for sea and land.

5. The method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches according to claim 1, characterized in that, The calculation of the first near-surface wind field based on the gradient wind balance physical model in step S4 is as follows: Using the Holland model with the center of the tropical cyclone as the origin, the gradient wind field is obtained by iteratively solving the gradient wind equation based on the central pressure difference, the maximum wind speed radius, the Holland B parameter, the Coriolis parameter, and the moving speed. Then, the gradient wind field is reduced to the near-surface wind field at a height of 10 meters through the boundary layer model.

6. The method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches according to claim 1, characterized in that, In step S5: The terrain adaptive correction model is a Transformer neural network based on a U-shaped encoder-decoder architecture; The inputs to the terrain adaptive correction model include wind speed scalar and U / V wind speed components derived from the first near-surface wind field, as well as elevation and land use type information derived from the high-resolution terrain data. The terrain adaptive correction model is obtained by training with high-precision numerical model simulation results as ground truth.

7. The method for modeling systemic tropical cyclones and high-resolution wind fields based on environment-driven approaches according to claim 6, characterized in that: The basic feature transformation unit of the Transformer neural network is a local enhancement window transformation block; The local enhancement window transformation block sequentially includes layer normalization, a multi-head self-attention mechanism based on local windows, and a feedforward network containing depthwise convolution operations.

8. A system for modeling tropical cyclones and high-resolution wind fields based on environment-driven systems, characterized in that, Used to implement the modeling method as described in any one of claims 1 to 7.

9. A computer device, the computer device comprising a memory, a processor, and a computer program, characterized in that, When the computer program is executed by a processor, it implements the modeling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the modeling method as described in any one of claims 1 to 7.

Images