A method for reconstructing typhoon wind fields based on parametric wind field and diffusion models

Abstract

This invention discloses a typhoon wind field reconstruction method based on a parametric wind field and diffusion model, relating to the technical field of typhoon wind field reconstruction and typhoon disaster assessment. The method includes the following steps: Step 1: Obtain a reanalysis typhoon sample set; Step 2: Generate a Holland-type parametric typhoon sample set corresponding to the reanalysis typhoon samples; Step 3: Construct a training database for typhoon wind field reconstruction; Step 4: Build and train a typhoon wind field reconstruction model based on a conditional diffusion model; Step 5: Obtain the optimal typhoon path parameters of the Joint Typhoon Warning Center (JTWC) for the target typhoon, and generate Holland-type parametric typhoon data on a preset local grid; Step 6: Input the Holland-type parametric typhoon data generated in Step 5 into the conditional diffusion model trained in Step 4, execute a back-diffusion process, and generate the reconstructed wind field of the target typhoon process. This method is used to reconstruct typhoon wind fields.

Description

Technical Field

[0001] This invention relates to the field of typhoon wind field reconstruction and typhoon disaster assessment technology, and in particular to a typhoon wind field reconstruction method based on a parametric wind field and diffusion model. Background Technology

[0002] Typhoon wind fields are one of the essential foundational data for offshore and coastal engineering design, storm surge simulation, and wind disaster risk assessment. In practical engineering applications, high spatial resolution and high accuracy typhoon wind fields are required to accurately describe the maximum wind speed near the typhoon center and the overall wind field structure. Currently, typhoon wind fields are mainly obtained through two methods: one is reanalysis wind fields generated based on numerical models and data assimilation, and the other is optimal typhoon track and intensity parameters provided by meteorological operational agencies.

[0003] While reanalysis wind fields offer good continuity and integrity in time and space, limitations in model physics and data assimilation schemes often lead to systematic biases in extreme wind speeds and local structures during severe typhoon events. This is particularly evident in the underestimation of extreme wind speeds near the center, making direct load calculations and risk assessments based on reanalysis wind fields prone to error. Typhoon warning centers and joint typhoon warning centers typically provide key parameters such as typhoon track, maximum wind speed, maximum wind speed radius, and central pressure based on observations and operational analysis. These parameters offer high timeliness and reliability, but their data dimensions are limited, lacking a complete spatial wind field distribution, making it difficult to directly meet the needs of high-resolution typhoon wind field reconstruction.

[0004] On the other hand, regarding data sources, there is a significant "modal difference" between reanalysis wind fields and typhoon parameters from the Joint Typhoon Warning Center (JTWC): reanalysis data is presented as wind speed vector fields on a regular spatial grid, offering high information dimensionality but limited accuracy; JTWC data, on the other hand, is presented as a limited number of parameters such as typhoon path, maximum wind speed, maximum wind speed radius, and central pressure, offering higher accuracy but lacking spatial distribution information. Existing methods, if trained solely on reanalysis wind fields, struggle to directly utilize warning center data containing only path and intensity parameters in operational phases. Even attempts to incorporate both types of data into the same model often lack a unified intermediate representation, making it difficult for the model to effectively transfer and fuse between high-dimensional wind fields and low-dimensional parameters.

[0005] In summary, the existing technology has the following technical problems that need to be solved:

[0006] 1. Although the wind field data is complete, it is insufficient in terms of the accuracy of extreme wind speeds and wind circle size, making it difficult to accurately reflect the true wind field intensity near the typhoon center.

[0007] 2. Although the typhoon path and intensity parameters provided by various typhoon warning centers are accurate, the data are mostly in the form of a small number of scalars or path parameters, which are difficult to use directly to construct high spatiotemporal resolution two-dimensional or three-dimensional typhoon wind fields.

[0008] 3. The reanalysis wind field is given in the form of wind speed vector field on a regular spatial grid, while the data of various typhoon warning centers are given in the form of a few parameters such as typhoon path, maximum wind speed, maximum wind speed radius, and central pressure. There is a lack of a unified intermediate representation between the two, which makes it difficult for the model trained solely on the reanalysis wind field to be directly transferred to the case of only inputting warning center parameters in the operational stage. It is impossible to make full use of the information of both types of data within the same framework.

[0009] In view of this, this invention is hereby proposed. Summary of the Invention

[0010] The purpose of this invention is to address the shortcomings of existing technologies by proposing a typhoon wind field reconstruction method based on a parametric wind field and diffusion model. This method can comprehensively utilize reanalysis wind fields and typhoon parameters from meteorological operational agencies to construct typhoon wind fields. While ensuring the integrity of the spatial distribution of the wind field, it can also statistically characterize the uncertainty of the typhoon wind field, thereby improving the intensity and structural accuracy of the typhoon wind field.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A typhoon wind field reconstruction method based on a parametric wind field and diffusion model, characterized by the following steps:

[0013] Step 1: Obtain the optimal path data of the Joint Typhoon Warning Center corresponding to the historical typhoon samples, determine the occurrence time series and typhoon center location series of the historical typhoon samples, and obtain the corresponding reanalysis wind field data; preprocess the reanalysis wind field data to align it in time and unify it in space to a preset local grid based on the typhoon center, to obtain the reanalysis typhoon sample set; the samples in the reanalysis typhoon sample set should include at least the zonal component u0 of wind speed, the meridional component v0 of wind speed, and sea level pressure;

[0014] Step 2: Select the Holland-type parametric typhoon wind field model, and solve the parameter set of the Holland-type model based on the reanalysis typhoon sample set obtained in Step 1. Generate the Holland-type parametric typhoon sample set corresponding to the reanalysis typhoon sample on the preset local grid. The samples in the Holland-type parametric typhoon sample set include at least the zonal component u0 of wind speed, the meridional component v0 of wind speed, and the sea level pressure.

[0015] Step 3: Spatiotemporally pair the reanalysis typhoon sample set obtained in Step 1 with the Holland-type parameterized typhoon sample set obtained in Step 2 to construct a training database for typhoon wind field reconstruction.

[0016] Step 4: Build and train a typhoon wind field reconstruction model based on the conditional diffusion model. Use the Holland-type parameterized typhoon samples in the training database used for typhoon wind field reconstruction in Step 3 as conditional inputs and the reanalysis typhoon samples as the learning target to iteratively train the conditional diffusion model until convergence.

[0017] Step 5: Obtain the optimal typhoon track parameters from the Joint Typhoon Warning Center for the target typhoon, and generate Holland-type parameterized typhoon data on the preset local grid.

[0018] Step 6: Input the Holland-type parameterized typhoon data generated in Step 5 into the conditional diffusion model trained in Step 4, execute the reverse diffusion process, and generate the reconstructed wind field of the target typhoon process.

[0019] Furthermore, step 1 includes the following steps:

[0020] Step 1.1: Obtain the typhoon center location, typhoon intensity level and path information of several historical typhoon processes, and create a typhoon sample set vector. Each sample includes: typhoon number, time, typhoon center longitude, typhoon center latitude, typhoon level, and maximum wind speed radius.

[0021] Step 1.2: Determine the temporal search range of historical typhoons based on the time, construct the spatial search range based on the longitude and latitude of the typhoon center, and obtain low spatial resolution reanalysis wind field data from the reanalysis database based on the temporal and spatial search ranges; the low spatial resolution reanalysis wind field data shall include at least the rasterized zonal component u0 of wind speed, the meridional component v0 of wind speed, and sea level pressure.

[0022] Step 1.3: Using the typhoon center position at each moment as a reference, perform localized cropping, coordinate unification and interpolation processing on the low spatial resolution reanalysis wind field data to unify the data of each typhoon process to a preset time interval and preset local grid; coordinate transformation is to transfer the longitude from the rasterized grid of [-180, 180] to the raster grid of [0-360].

[0023] Furthermore, step 2 includes the following steps:

[0024] Step 2.1: Determine the required parameter set for the Holland-type parametric typhoon wind field model, including the typhoon center location, central pressure, and ambient pressure;

[0025] Step 2.2: Extract the radial wind speed profile and parameter value set of the typhoon center area from the reanalysis wind field data in multiple azimuth directions. The parameter value set is the set of parameter values ​​in Step 2.1.

[0026] Step 2.3: Based on the radial wind speed profile and parameter set extracted in Step 2.2, solve for the maximum wind speed and its corresponding radius, and combine the central pressure and ambient pressure to determine the shape parameters, thus obtaining the complete set of values ​​for each parameter at each time.

[0027] Step 2.4: Based on the numerical set obtained in Step 2.3, calculate the radial distance and azimuth from the grid point to the typhoon center on the preset local grid, and substitute them into the Holland-type parametric wind field model to generate a Holland-type parametric typhoon sample set containing the zonal component u0 of wind speed, the meridional component v0 of wind speed, and the sea level pressure.

[0028] Furthermore, step 3 includes the following steps:

[0029] Step 3.1: Pair reanalysis typhoon samples from the same typhoon at the same time with Holland-type parametric typhoon samples to form sample pairs;

[0030] Step 3.2: Perform data preprocessing on the sample pairs to form a training database; data preprocessing includes one or more of the following: window partitioning, time stacking, standardization, and normalization; the window length is set according to the business time resolution.

[0031] Furthermore, in step 4, a typhoon wind field reconstruction model is constructed based on the conditional diffusion model, including a diffusion process and a reverse generation process. The diffusion process progressively adds Gaussian noise to the reanalyzed typhoon samples, while the reverse generation process uses a denoising network to progressively remove noise and reconstruct the target wind field under Holland-type parameterized typhoon sample conditions. The noise addition process of the diffusion model is as follows:

[0032] ;

[0033] ;

[0034] In the formula, and This refers to the zonal and meridional components of wind speed in Holland-type parametric typhoon wind field rasterized data. , The noise is randomly generated Gaussian noise. Here, i and j are the noise intensity parameters of the diffusion model, and i and j are the grid point labels in the wind field grid.

[0035] Furthermore, in step 4, the denoising network based on the conditional diffusion model reverse generation process adopts an autoencoder neural network structure, including an encoder, a decoder, and skip connections. The encoder is used to extract multi-scale spatial features, the decoder is used to reconstruct the spatial distribution of the wind field, and the skip connections are used to fuse feature information at different scales.

[0036] Furthermore, in step 4, the zonal component of wind speed is predicted based on the conditional diffusion model. and wind speed meridional component The formula is as follows:

[0037] ;

[0038] ;

[0039] In the formula, , This is the Gaussian noise predicted by the diffusion model.

[0040] Furthermore, in step 4, the loss function based on the conditional diffusion model includes the main loss. and wind speed modulus loss The calculation formula is as follows:

[0041] ;

[0042] In the formula, , These are the weighting coefficients;

[0043] In the formula,

[0044] ;

[0045] .

[0046] Furthermore, step 5 includes the following steps:

[0047] Step 5.1: Obtain the optimal typhoon path parameters, including: typhoon center longitude. Longitude and latitude of the typhoon center The central pressure of the typhoon is The ambient air pressure is Maximum wind speed The radius corresponding to the maximum wind speed is Construct a pre-defined local grid, where the grid points are... The longitude of the grid points is The grid point latitude is ;

[0048] Step 5.2: Calculate the radial distance from the typhoon center to the grid points. The formula is as follows:

[0049] ;

[0050] In the formula, Re is the Earth's radius;

[0051] In the formula,

[0052] ;

[0053] ;

[0054] ;

[0055] Step 5.3: Calculate the azimuth of the grid points The formula is as follows:

[0056] ;

[0057] Step 5.4: Calculate the shape parameter B(t), using the following formula:

[0058] ;

[0059] In the formula, Where is the air density, and e is an adjustable constant;

[0060] In the formula,

[0061] ;

[0062] ;

[0063] Step 5.5: Calculate the Holland pressure distribution The formula is as follows:

[0064] ;

[0065] Step 5.6: Calculate the gradient wind speed of the tangential wind speed The formula is as follows:

[0066] ;

[0067] in,

[0068] ;

[0069] Step 5.7: Calculate the rasterized data u0 and v0 of the Holland-type parametric typhoon wind field, using the following formula:

[0070] ;

[0071] .

[0072] Compared with the prior art, the beneficial effects of this invention are as follows:

[0073] 1. A method for reconstructing typhoon wind fields based on Holland-type parametric wind fields and diffusion models is proposed. This method introduces Holland-type parametric wind fields as an intermediate representation between the reanalysis wind field and the typhoon parameters from the Joint Typhoon Warning Center. During the training phase, the Holland-type parametric wind field obtained from the reanalysis wind field inversion is used to constrain the diffusion model to learn the probabilistic mapping relationship between the "parametric wind field and the high-precision wind field". During the operational application phase, a small number of high-precision typhoon parameters provided by the Joint Typhoon Warning Center are used to construct a Holland-type parametric wind field and drive the diffusion model to generate the reconstructed wind field. Thus, a complete spatial typhoon wind field with uncertain representation is generated from a small number of high-precision typhoon parameters, improving the accuracy of extreme wind speeds and wind field structure while maintaining the integrity of the spatial distribution of the wind field.

[0074] 2. The Holland-type parameterized typhoon wind field model with fewer parameters and clear physical meaning can generate a complete spatial typhoon wind field under the condition of relatively scarce but reliable data (a small number of high-precision optimal typhoon path parameters), thereby reducing the dependence on observation resources and having good applicability.

[0075] 3. A nonlinear mapping relationship between Holland-type parameterized wind field and reanalysis wind field was established through a conditional diffusion model. The final reconstructed wind field was obtained by executing the reverse diffusion process multiple times. Under the premise of ensuring reasonable overall structure and intensity, it can effectively reflect the uncertainty and random fluctuation characteristics of typhoon wind field, which is conducive to wind disaster risk assessment and scenario analysis.

[0076] 4. It can be directly used for wind load calculation of marine engineering structures, numerical simulation of storm surges and wind disaster risk assessment, improving the reliability of engineering design and disaster early warning. Attached Figure Description

[0077] Figure 1 The flowchart shows a typhoon wind field reconstruction method based on a parametric wind field and diffusion model.

[0078] Figure 2 Schematic diagram of the forward direction and azimuth of Typhoon MAWAR

[0079] Figure 3 The reconstruction results for the northeast quadrant of Typhoon MAWAR;

[0080] Figure 4 The reconstruction results for the northwestern quadrant of Typhoon MAWAR;

[0081] Figure 5 The reconstruction results for the southwest quadrant of Typhoon MAWAR;

[0082] Figure 6 This is the reconstruction result for the southeast quadrant of Typhoon MAWAR. Detailed Implementation

[0083] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0084] Example 1:

[0085] A method for reconstructing typhoon wind fields based on a parametric wind field and diffusion model, such as... Figure 1 As shown, it includes the following steps:

[0086] Step 1: Obtain the optimal path data of the Joint Typhoon Warning Center (JTWC) corresponding to the historical typhoon samples, determine the occurrence time series and typhoon center location series of the historical typhoon samples, and obtain the corresponding reanalysis wind field data; preprocess the reanalysis wind field data to align it in time and unify it in space to a preset local grid based on the typhoon center, to obtain the reanalysis typhoon sample set; the samples in the reanalysis typhoon sample set should at least include the zonal component u0 of wind speed, the meridional component v0 of wind speed, and sea level pressure.

[0087] Specifically, step 1 includes the following steps:

[0088] Step 1.1: Obtain information on the typhoon center location, typhoon intensity level, and path of several historical typhoon events, and create a typhoon sample set vector. Each sample includes: typhoon number, time, typhoon center longitude, typhoon center latitude, typhoon level, and maximum wind speed radius.

[0089] In this embodiment, the same typhoon consists of samples from several moments; the samples are obtained based on the best path data provided by the Joint Typhoon Warning Center.

[0090] Step 1.2: Determine the temporal search range of historical typhoons based on the time, construct the spatial search range based on the longitude and latitude of the typhoon center, and obtain low spatial resolution reanalysis wind field data from the reanalysis database based on the temporal and spatial search ranges; the low spatial resolution reanalysis wind field data includes at least the rasterized zonal component u0 of wind speed, the meridional component v0 of wind speed, and sea level pressure.

[0091] In this embodiment, the time search range can be determined based on the initial and final times of the typhoon; the spatial search range can be determined by adding n° (e.g., 6°) to the typhoon center longitude and latitude at the initial and final times; thereafter, low spatial resolution reanalysis wind field data is indexed and obtained from a reanalysis database (e.g., ERA5).

[0092] Step 1.3: Using the typhoon center position at each moment as a reference, perform localized cropping, coordinate unification and interpolation processing on the low spatial resolution reanalysis wind field data to unify the data of each typhoon process to a preset time interval and preset local grid; coordinate transformation is to transfer the longitude from the rasterized grid of [-180, 180] to the raster grid of [0-360].

[0093] In this embodiment, the reanalysis wind field data is time-differenced according to a preset time interval, and the longitude is transferred from a rasterized grid of [-180, 180] to a raster grid of [0-360], thereby unifying the reanalysis wind field data to the preset time interval and spatial resolution, and completing the temporal and spatial preprocessing of the low spatial resolution reanalysis wind field data.

[0094] Step 2: Select the Holland-type parametric typhoon wind field model, solve the parameter set of the Holland-type model based on the reanalysis typhoon sample set obtained in Step 1, and generate the Holland-type parametric typhoon sample set corresponding to the reanalysis typhoon sample on the preset local grid; the samples in the Holland-type parametric typhoon sample set include at least the zonal component u0 of wind speed, the meridional component v0 of wind speed, and the sea level pressure.

[0095] Specifically, step 2 includes the following steps:

[0096] Step 2.1: Determine the required parameter set for the Holland-type parametric typhoon wind field model, including the typhoon center location, central pressure, and ambient pressure.

[0097] Step 2.2: Extract the radial wind speed profile and parameter value set of the typhoon center area from the reanalysis wind field data in multiple azimuth directions. The parameter value set is the set of parameter values ​​in Step 2.1.

[0098] In this embodiment, the multiple azimuth angles are azimuth angles in the first, second, third, and fourth quadrants, with the typhoon's direction of travel as the positive direction and the typhoon's center as the center, and spaced 90° apart.

[0099] In this embodiment, the typhoon center range can be the spatial search range in step 1.2.

[0100] Step 2.3: Based on the radial wind speed profile and parameter set extracted in Step 2.2, solve for the maximum wind speed and its corresponding radius, and determine the shape parameters by combining the central pressure and ambient pressure, so as to obtain the complete set of values ​​for each parameter at each time.

[0101] Step 2.4: Based on the numerical set obtained in Step 2.3, calculate the radial distance and azimuth from the grid point to the typhoon center on the preset local grid, and substitute them into the Holland-type parametric wind field model to generate a Holland-type parametric typhoon sample set containing the zonal component u0 of wind speed, the meridional component v0 of wind speed, and the sea level pressure.

[0102] It should be noted that, in specific implementation, other types of parametric typhoon wind field models can also be selected, such as the non-standard circular Chavas wind field.

[0103] Step 3: Spatiotemporally pair the reanalysis typhoon sample set obtained in Step 1 with the Holland-type parameterized typhoon sample set obtained in Step 2 to construct a training database for typhoon wind field reconstruction.

[0104] Specifically, step 3 includes the following steps:

[0105] Step 3.1: Pair reanalysis typhoon samples from the same typhoon at the same time with Holland-type parametric typhoon samples to form sample pairs.

[0106] Step 3.2: Perform data preprocessing on the sample pairs to form a training database; data preprocessing includes one or more of the following: window partitioning, time stacking, standardization, and normalization; the window length is set according to the business time resolution.

[0107] In this embodiment, a window length of 6 hours, 12 hours, or 24 hours can be formed based on 3 hours of data.

[0108] In this embodiment, the time window can be 6 hours, 12 hours, or 24 hours.

[0109] Step 4: Build and train a typhoon wind field reconstruction model based on the conditional diffusion model. Use the Holland-type parameterized typhoon samples in the training database used for typhoon wind field reconstruction in Step 3 as conditional inputs and the reanalysis typhoon samples as learning targets to iteratively train the conditional diffusion model until convergence.

[0110] In this embodiment, a nonlinear mapping relationship between Holland-type parameterized typhoon samples and reanalysis typhoon samples is constructed through step 4.

[0111] Typhoon wind fields exhibit inherent uncertainties due to the combined effects of multi-scale turbulence, convective systems, and observation and assimilation errors. Even for the same typhoon event, with similar path and intensity parameters, the wind circle structure and local wind speed distribution may differ, making a single deterministic wind field insufficient for comprehensive characterization. This embodiment constructs a typhoon wind field reconstruction model based on a conditional diffusion model, which can realistically reflect the detailed characteristics and randomness of typhoon wind fields.

[0112] Specifically, in step 4, a typhoon wind field reconstruction model is constructed based on the conditional diffusion model, including a diffusion process and a reverse generation process. The diffusion process progressively adds Gaussian noise to the reanalyzed typhoon samples, while the reverse generation process uses a denoising network to progressively remove noise and reconstruct the target wind field under Holland-type parameterized typhoon sample conditions. The noise addition process of the diffusion model is as follows:

[0113] ;

[0114] ;

[0115] In the formula, and This refers to the zonal and meridional components of wind speed in Holland-type parametric typhoon wind field rasterized data. , The noise is randomly generated Gaussian noise. Here, i and j are the noise intensity parameters of the diffusion model, and i and j are the grid numbers in the wind field grid.

[0116] On the other hand, in step 4, the denoising network based on the conditional diffusion model reverse generation process adopts an autoencoder neural network structure, including an encoder, a decoder, and skip connections. The encoder is used to extract multi-scale spatial features, the decoder is used to reconstruct the spatial distribution of the wind field, and the skip connections are used to fuse feature information at different scales.

[0117] On the other hand, in step 4, the zonal component of wind speed is predicted based on the conditional diffusion model. and wind speed meridional component The formula is as follows:

[0118] ;

[0119] ;

[0120] In the formula, , This is the Gaussian noise predicted by the diffusion model.

[0121] Furthermore, in step 4, the loss function based on the conditional diffusion model includes the main loss. and wind speed modulus loss The calculation formula is as follows:

[0122] ;

[0123] In the formula, , These are the weighting coefficients;

[0124] In the formula,

[0125] ;

[0126] .

[0127] In this embodiment, wind speed modulus loss is added. It can improve the physical rationality of the generated wind field.

[0128] Step 5: Obtain the optimal typhoon track parameters from the Joint Typhoon Warning Center for the target typhoon, and generate Holland-type parameterized typhoon data on a preset local grid.

[0129] In this embodiment, steps 5 and 6 are the service broadcasting stage.

[0130] Specifically, step 5 includes the following steps:

[0131] Step 5.1: Obtain the optimal typhoon path parameters, including: typhoon center longitude. Longitude and latitude of the typhoon center The central pressure of the typhoon is The ambient air pressure is Maximum wind speed The radius corresponding to the maximum wind speed is Construct a pre-defined local grid, where the grid points are... The longitude of the grid points is The grid point latitude is .

[0132] In this embodiment, the optimal typhoon path parameters of the target typhoon are preferably obtained from the Joint Typhoon Warning Center and / or the China Meteorological Administration; the preset local grid is a grid constructed with the typhoon's forward wind direction as north and the typhoon center as the center; in specific implementation, the grid resolution is set as needed.

[0133] Step 5.2: Calculate the radial distance from the typhoon center to the grid points. The formula is as follows:

[0134] ;

[0135] In the formula, Re is the Earth's radius;

[0136] In the formula,

[0137] ;

[0138] ;

[0139] .

[0140] Step 5.3: Calculate the azimuth of the grid points The formula is as follows:

[0141] .

[0142] Step 5.4: Calculate the shape parameter B(t), using the following formula:

[0143] ;

[0144] In the formula, Let be the air density, and e be an adjustable constant.

[0145] In the formula,

[0146] ;

[0147] .

[0148] Step 5.5: Calculate the Holland pressure distribution The formula is as follows:

[0149] .

[0150] Step 5.6: Calculate the gradient wind speed of the tangential wind speed The formula is as follows:

[0151] ;

[0152] in,

[0153] .

[0154] Step 5.7: Calculate the rasterized data u0 and v0 of the Holland-type parametric typhoon wind field, using the following formula:

[0155] ;

[0156] .

[0157] Step 6: Input the Holland-type parameterized typhoon data generated in Step 5 into the conditional diffusion model trained in Step 4, execute the reverse diffusion process, and generate the reconstructed wind field of the target typhoon process.

[0158] This embodiment proposes a typhoon wind field reconstruction method based on a Holland-type parametric wind field and diffusion model. This method introduces a Holland-type parametric wind field as an intermediate representation between the reanalysis wind field and the typhoon parameters from the Joint Typhoon Warning Center (JTWC). During the training phase, the Holland-type parametric wind field obtained from the reanalysis wind field inversion constrains the diffusion model to learn the probabilistic mapping relationship between the parametric wind field and the high-precision wind field. In the operational application phase, a small number of high-precision typhoon parameters provided by the JTWC are used to construct a Holland-type parametric wind field and drive the diffusion model to generate a reconstructed wind field. This allows for the generation of a complete spatial typhoon wind field with uncertain representations using a small number of high-precision typhoon parameters, improving the accuracy of extreme wind speeds and wind field structure while maintaining the integrity of the wind field's spatial distribution.

[0159] The method in this embodiment uses a Holland-type parameterized typhoon wind field model with fewer parameters and clear physical meaning. It can generate a complete spatial typhoon wind field under the condition of relatively scarce but reliable data (a small number of high-precision optimal typhoon path parameters), thereby reducing the dependence on observation resources and having good applicability.

[0160] The method in this embodiment establishes a nonlinear mapping relationship between the Holland-type parameterized wind field and the reanalysis wind field through a conditional diffusion model. The final reconstructed wind field is obtained by executing the reverse diffusion process multiple times. Under the premise of ensuring reasonable overall structure and intensity, it can effectively reflect the uncertainty and random fluctuation characteristics of typhoon wind fields, which is beneficial for wind disaster risk assessment and scenario analysis.

[0161] In summary, the method of this embodiment can be directly used for wind load calculation of marine engineering structures, storm surge numerical simulation, and wind disaster risk assessment, thereby improving the reliability of engineering design and disaster early warning.

[0162] To verify the effectiveness of the method in this embodiment, Typhoon MAWAR (times at 20:00 and 21:00 on May 27, 2023) was used as an example for verification. Figure 2 This diagram illustrates the direction and azimuth of Typhoon MAWAR. The circles in the diagram represent the typhoon's wind circle outline. The change in the typhoon's center position from time T-1 to time T is defined as the typhoon's path. The direction from time T-1 to time T is the typhoon's direction of movement at time T-1. Using the direction of movement as local due north, the typhoon is divided into four quadrants perpendicular to the direction of movement. The first quadrant is the front half of the right hemisphere along the direction of movement, and the quadrants are designated 1, 2, 3, and 4 in a clockwise direction. Figures 3-6The figures show the reconstruction results for the MAWAR typhoon in the northeast, northwest, southwest, and southeast quadrants, respectively. The solid lines represent the average wind speed profiles of the typhoon's wind field in the four quadrants. The blue solid line represents the input wind speed profile of the Holland-type parametric wind field, the orange solid line represents the wind speed profile generated by the conditional diffusion model in this embodiment, and the green solid line represents the wind speed profile of the reanalysis wind field (obtained from ERA). The X-axis represents the distance from the typhoon center, and the Y-axis represents the wind speed magnitude. It can be seen that the green and orange solid lines show good consistency in trend. The RMSE in the southwest quadrant at 20:00 is low, with the lowest being only 0.9 m / s. Furthermore, the RMSE in the southwest quadrant at 21:00 is 1.68 m / s. The northeast quadrant at 20:00 and 21:00 are 2.24 m / s and 3.47 m / s, respectively. The southeast quadrant at 20:00 and 21:00 are 3.07 m / s and 4.57 m / s, respectively. The northwest quadrant at 20:00 and 21:00 are 2.24 m / s and 3.18 m / s, respectively. In summary, the conditional diffusion model in this embodiment can effectively reconstruct and reanalyze the wind field's wind speed distribution pattern and magnitude from the parameterized wind field.

[0163] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for reconstructing typhoon wind fields based on a parametric wind field and diffusion model, characterized in that, Includes the following steps: Step 1: Obtain the optimal typhoon path data of the Joint Typhoon Warning Center corresponding to the historical typhoon samples, determine the occurrence time series and typhoon center location series of the historical typhoon samples, and obtain the corresponding reanalysis wind field data; preprocess the reanalysis wind field data to align it in time and unify it in space to a preset local grid based on the typhoon center, and obtain the reanalysis typhoon sample set. Further analysis of the typhoon sample set reveals that the samples must include at least the zonal component of wind speed u0, the meridional component of wind speed v0, and sea level pressure. Step 2: Select the Holland-type parametric typhoon wind field model, and solve the parameter set of the Holland-type model based on the reanalysis typhoon sample set obtained in Step 1. Generate the Holland-type parametric typhoon sample set corresponding to the reanalysis typhoon sample on the preset local grid. The samples in the Holland-type parametric typhoon sample set include at least the zonal component u0 of wind speed, the meridional component v0 of wind speed, and the sea level pressure. Step 3: Spatiotemporally pair the reanalysis typhoon sample set obtained in Step 1 with the Holland-type parameterized typhoon sample set obtained in Step 2 to construct a training database for typhoon wind field reconstruction. Step 4: Build and train a typhoon wind field reconstruction model based on the conditional diffusion model. Use the Holland-type parameterized typhoon samples in the training database used for typhoon wind field reconstruction in Step 3 as conditional inputs and the reanalysis typhoon samples as the learning target to iteratively train the conditional diffusion model until convergence. Step 5: Obtain the optimal typhoon track parameters from the Joint Typhoon Warning Center for the target typhoon, and generate Holland-type parameterized typhoon data on the preset local grid. Step 6: Input the Holland-type parameterized typhoon data generated in Step 5 into the conditional diffusion model trained in Step 4, execute the reverse diffusion process, and generate the reconstructed wind field of the target typhoon process.

2. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 1, characterized in that, Step 1 includes the following steps: Step 1.1: Obtain the typhoon center location, typhoon intensity level and path information of several historical typhoon processes, and create a typhoon sample set vector. Each sample includes: typhoon number, time, typhoon center longitude, typhoon center latitude, typhoon level, and maximum wind speed radius. Step 1.2: Determine the temporal search range of historical typhoons based on the time, construct the spatial search range based on the longitude and latitude of the typhoon center, and obtain low spatial resolution reanalysis wind field data from the reanalysis database based on the temporal and spatial search ranges; the low spatial resolution reanalysis wind field data shall include at least the rasterized zonal component u0 of wind speed, the meridional component v0 of wind speed, and sea level pressure. Step 1.3: Using the typhoon center position at each moment as a reference, perform localized cropping, coordinate unification and interpolation processing on the low spatial resolution reanalysis wind field data to unify the data of each typhoon process to a preset time interval and preset local grid; coordinate transformation is to transfer the longitude from the rasterized grid of [-180, 180] to the raster grid of [0-360].

3. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 1, characterized in that, Step 2 includes the following steps: Step 2.1: Determine the required parameter set for the Holland-type parametric typhoon wind field model, including the typhoon center location, central pressure, and ambient pressure; Step 2.2: Extract wind field data from reanalysis data in multiple azimuth directions. Extract the radial wind speed profile and parameter value set within the typhoon's center area. The parameter value set is the set of parameter values ​​from step 2.

1. Step 2.3: Based on the radial wind speed profile and parameter set extracted in Step 2.2, solve for the maximum wind speed and its corresponding radius, and combine the central pressure and ambient pressure to determine the shape parameters, thus obtaining the complete set of values ​​for each parameter at each time. Step 2.4: Based on the numerical set obtained in Step 2.3, calculate the radial distance and azimuth from the grid point to the typhoon center on the preset local grid, and substitute them into the Holland-type parametric wind field model to generate a Holland-type parametric typhoon sample set containing the zonal component u0 of wind speed, the meridional component v0 of wind speed, and the sea level pressure.

4. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1: Pair reanalysis typhoon samples from the same typhoon at the same time with Holland-type parametric typhoon samples to form sample pairs; Step 3.2: Preprocess the sample pairs to form a training database; Data preprocessing includes one or more of the following: window partitioning, time stacking, standardization, and normalization; the window length is set according to the business time resolution.

5. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 1, characterized in that, In step 4, a typhoon wind field reconstruction model is constructed based on the conditional diffusion model, including the diffusion process and the reverse generation process; The diffusion process progressively adds Gaussian noise to the reanalysis typhoon samples, while the reverse generation process uses a denoising network to progressively remove noise and reconstruct the target wind field under Holland-type parameterized typhoon sample conditions. The noise addition process of the diffusion model is as follows: ； ； In the formula, and This refers to the zonal and meridional components of wind speed in Holland-type parametric typhoon wind field rasterized data. , The noise is randomly generated Gaussian noise. Here, i and j are the noise intensity parameters of the diffusion model, and i and j are the grid point labels in the wind field grid.

6. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 5, characterized in that, In step 4, the denoising network based on the conditional diffusion model reverse generation process adopts an autoencoder neural network structure, including an encoder, a decoder, and skip connections. The encoder is used to extract multi-scale spatial features, the decoder is used to reconstruct the spatial distribution of the wind field, and the skip connections are used to fuse feature information at different scales.

7. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 5, characterized in that, In step 4, the zonal component of wind speed is predicted based on the conditional diffusion model. and wind speed meridional component The formula is as follows: ； ； In the formula, , This is the Gaussian noise predicted by the diffusion model.

8. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 7, characterized in that, In step 4, the loss function based on the conditional diffusion model includes the main loss. and wind speed modulus loss The calculation formula is as follows: ； In the formula, , These are the weighting coefficients; In the formula, ； 。 9. The typhoon wind field reconstruction method based on a parametric wind field and diffusion model according to claim 1, characterized in that, Step 5 includes the following steps: Step 5.1: Obtain the optimal typhoon path parameters, including: typhoon center longitude. Longitude and latitude of the typhoon center The central pressure of the typhoon is The ambient air pressure is Maximum wind speed The radius corresponding to the maximum wind speed is Construct a pre-defined local grid, where the grid points are... The longitude of the grid points is The grid point latitude is ; Step 5.2: Calculate the radial distance from the typhoon center to the grid points. The formula is as follows: ； In the formula, Re is the Earth's radius; In the formula, ； ； ； Step 5.3: Calculate the azimuth of the grid points The formula is as follows: ； Step 5.4: Calculate the shape parameter B(t), using the following formula: ； In the formula, Where is the air density, and e is an adjustable constant; In the formula, ； ； Step 5.5: Calculate the Holland pressure distribution The formula is as follows: ； Step 5.6: Calculate the gradient wind speed of the tangential wind speed The formula is as follows: ； in, ； Step 5.7: Calculate the rasterized data u0 and v0 of the Holland-type parametric typhoon wind field, using the following formula: ； 。

Images