A typhoon full path simulation method based on conditional generative adversarial network

Abstract

The present application belongs to the technical field of structural wind resistance design, and particularly relates to a typhoon full-path simulation method based on conditional generative adversarial network. The present application firstly establishes a typhoon occurrence probability model and a typhoon movement speed random generator, generates random samples of typhoon movement speed under given 250hPa and 850hPa height air flow conditions, and simulates typhoon movement trajectory; then establishes a typhoon intensity model to simulate typhoon intensity evolution; based on the randomly generated typhoon path and intensity samples, the basic meteorological parameters of each point of the coastal shoreline are statistically compared with historical data; a typhoon three-dimensional nonlinear boundary layer wind field model is established to simulate typhoon annual extreme wind speed samples, and a typhoon ground surface annual extreme wind speed probability distribution is established. The present application establishes a nonlinear mapping relationship between typhoon movement and intensity change and environmental variables, and reflects the non-Gaussian probability characteristics of typhoon intensity evolution.

Description

Technical Field

[0001] This invention belongs to the field of structural wind resistance performance design technology, and in particular relates to a typhoon full-path simulation method based on conditional generative adversarial networks. Background Technology

[0002] Every time a typhoon passes through, it is accompanied by secondary disasters such as strong winds, torrential rains, and storm surges. These disasters not only destroy infrastructure such as buildings, bridges, and roads on land, but also trigger natural disasters such as tsunamis, floods, and mudslides. When a typhoon occurs, meteorological departments usually report important information such as the typhoon's trajectory, expected landfall time, intensity level, cities it will pass through, and the scope of its impact, reminding people to take appropriate protective measures. After a high-intensity natural disaster like a typhoon, if the development of the typhoon is not predicted in time, and people are not evacuated and infrastructure is not reinforced in advance, it may cause significant casualties and economic losses. The key to typhoon simulation and forecasting, and to conducting typhoon hazard analysis, lies in simulating important information such as the typhoon's path and wind intensity.

[0003] In the field of wind-resistant engineering design, the extreme wind speed that occurs once every 50 years is typically used to determine the magnitude of the wind load. In typhoon-affected areas, due to the scarcity of surface wind speed observation records, it is necessary to determine the 50-year return period wind speed through simulation-based extreme wind speed analysis methods. Among these methods, typhoon track prediction and simulation are crucial for extreme wind speed analysis. Current typhoon track simulation and prediction methods mainly include two numerical simulation methods: the simulated circle method and the full-track simulation method. Both are based on measured data of passing typhoons within a local area, applying typhoon wind field models and using statistical methods to simulate the typhoon. The difference lies in Vickery's proposed full-track simulation method, which divides the entire sea area into sections and performs linear regression analysis on key typhoon information within each section to establish a typhoon parameter model covering the entire sea area, thereby simulating the typhoon's development process across the entire ocean surface. The full-path simulation method not only improves upon the shortcomings of the circular simulation method for some high-latitude regions where measured typhoon data is scarce and statistical analysis of key typhoon parameters at simulation points is required, but it can also simulate the entire typhoon development process over the ocean. However, Vickery's linear regression model requires extensive, somewhat subjective adjustments to the model parameters to ensure consistency between the statistical values ​​of the simulated typhoon's key parameters and those of historical typhoons. Furthermore, this model cannot reflect the nonlinear relationship between environmental variables and typhoon movement and intensity evolution, making it difficult to implement widely.

[0004] Therefore, there is an urgent need for a typhoon full-path simulation method based on conditional generative adversarial networks to solve the above problems. Summary of the Invention

[0005] To overcome the shortcomings of the existing technology, this invention provides a typhoon full-path simulation method based on conditional generative adversarial networks. This invention establishes a nonlinear mapping relationship between typhoon motion and intensity changes and environmental variables, and reflects the non-Gaussian probabilistic characteristics of typhoon intensity evolution.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for simulating the full path of a typhoon based on conditional generative adversarial networks, the specific steps of which are as follows:

[0008] S1. Establish a typhoon occurrence probability model and generate samples of typhoon occurrence frequency, occurrence time, occurrence location, and initial intensity.

[0009] S2. Establish a random generator for typhoon movement speed to generate random samples of typhoon movement speed under given airflow conditions at 250hPa and 850hPa altitudes, and simulate the typhoon movement trajectory.

[0010] S3. Establish a typhoon intensity model, including a random generator for ocean typhoon intensity and a random generator for land typhoon intensity, generate random samples of typhoon intensity changes under given environmental physical variables, and simulate the evolution of typhoon intensity.

[0011] S4. Verify the simulation effect of typhoon path and intensity statistical characteristics. Based on the randomly generated typhoon path and intensity evolution simulation samples, statistically analyze the basic meteorological parameters of typhoons at various points along the coastal coastline and compare them with historical typhoon sample data.

[0012] S5. Establish a three-dimensional nonlinear boundary layer wind field model of typhoon, and simulate the surface wind speed process of typhoon by combining typhoon path and intensity samples.

[0013] S6. Based on simulated multi-year extreme typhoon wind speed samples, establish the probability distribution of extreme typhoon wind speeds on the ground.

[0014] Preferably, in step S1, the probability distribution of the annual number of typhoon occurrences is fitted using a Poisson distribution, and the number of typhoon occurrences within a given period (i.e., the probability of a typhoon occurring within a given period) is sampled; the joint temporal-spatial probability distribution of typhoon occurrences is fitted using a kernel density estimation method, and the typhoon occurrence time and location are sampled; the probability distribution of the initial intensity of typhoons is fitted using a maximum likelihood estimation method, and the initial intensity samples of typhoons are sampled.

[0015] Preferably, in step S2, based on historical typhoon data observed over the past 30-40 years (which can be obtained from the China Meteorological Administration's optimal path dataset and data from the third-generation operational meteorological observation satellite (NOAA) of the National Oceanic and Atmospheric Administration), a random sample generator for typhoon velocity under given environmental airflow conditions is trained using a conditional generative adversarial network method (e.g., the Wasserstein conditional generative adversarial network method). The formula for typhoon velocity is as follows:

[0016] U t =G U (ε U |x U (1)

[0017] Among them: U t =(u t ,v t x represents the typhoon's velocity at a given moment; U =(u 250 ,v 250 ,u 850 ,v 850 ,φ,λ) T Represents the environmental conditions at that moment; u 250 The eastward component of the basic airflow velocity at 250 hPa altitude; v 250 The northward component of the basic airflow velocity at 250 hPa altitude; u 850 The eastward component of the basic airflow velocity at 850 hPa altitude; v 850 The northward component of the basic airflow velocity at 850 hPa altitude; ε U Let G be a standard normal random variable; λ be the longitude of the typhoon center at that moment; φ be the latitude of the typhoon center at that moment; and G be a standard normal random variable. U It is a generator model, represented by a fully connected neural network.

[0018] Preferably, a random sample generator for typhoon velocity is used to generate random samples of typhoon velocity, and the typhoon center position at the next moment is simulated based on the current moment:

[0019]

[0020]

[0021] Where i represents the (i+1)th time; λ i+1 φ is the longitude of the typhoon center at time i+1; i+1 R represents the latitude of the typhoon center at time i+1; e =6371km, representing the Earth's radius; Δt =6 hours, representing the time interval between time i+1 and time i.

[0022] Preferably, before proceeding to step S3, the location of the typhoon center obtained in step S2 is used to determine whether the typhoon will be located over the ocean or make landfall at the next moment, and the corresponding typhoon intensity generator is used to simulate the evolution of the typhoon intensity.

[0023] Preferably, in step S3, a generator for ocean typhoon intensity under given environmental conditions is trained using a conditional generative adversarial network method (e.g., the Wasserstein conditional generative adversarial network method) based on historical typhoon data observed over the past 30-40 years (which can be obtained from the China Meteorological Administration's optimal track dataset and NOAA data).

[0024] I i+1 =G I (ε I,i |x I,i (4)

[0025] Where: I represents the relative intensity; i represents the (i+1)th time; ε I,i The zero-mean random disturbance term conforming to a normal distribution; x I,i =(λ i ,φ i ,T s,i ,T 0,i ,p 0,i ,S i V t,i ,h m,i ,Γ i ) T Let λ represent the ocean surface environmental conditions at time i; i Let φ be the longitude of the typhoon center at time i; i Let T be the latitude of the typhoon center at time i; s T0 is the ocean surface temperature; p0 is the tropopause temperature; S is the minimum ocean surface pressure; V is the vertical wind shear. t h represents the speed of the typhoon's movement. m Γ represents the thickness of the ocean surface mixing layer; Γ represents the ocean submixing temperature layer; G I This is a model for generating ocean typhoon intensity.

[0026] Preferably, in step S3, a land typhoon intensity generator for a given environmental condition is trained using a conditional generative adversarial network method (e.g., the Wasserstein conditional generative adversarial network method) based on historical typhoon data observed over the past 30-40 years (which can be obtained from the China Meteorological Administration's optimal track dataset and NOAA data).

[0027] Δp i+1 =G Δp (ε Δp,i |x Δp,i (5)

[0028] Where: Δp is the central pressure difference; i represents the (i+1)th time; ε Δp,i The zero-mean random disturbance term conforming to a normal distribution; x Δp,i =(λ i ,φ i ,p 0,i ,S i V t,i ) T λ represents the land environment conditions at time i; i Let φ be the longitude of the typhoon center at time i; i p0 is the latitude of the typhoon center at time i; S is the minimum sea surface pressure; V is the vertical wind shear; t G represents the speed of the typhoon. Δp This represents a land-based typhoon intensity generator model.

[0029] Preferably, in step S4, the historical data consists of historical meteorological parameters of typhoons at various measuring points along the coastal coastline over the past 30-40 years. To verify the statistical characteristics of the simulated typhoon movement path and intensity evolution samples, stations are set up along the coastal coastline at intervals of 50-100km. Then, the annual occurrence rate, movement direction, movement speed, and central pressure difference of the simulated typhoon samples within a specified radius (here, a range of 250km) of each station are statistically analyzed. The Kolmogorov-Smirnov test method is used to compare and analyze the data with the historical typhoon data. If the test results meet the requirements of a confidence level of 0.95, the model can be used for subsequent analysis. Otherwise, the hyperparameters and structure of the neural network need to be adjusted and retrained.

[0030] Preferably, in step S5, during the typhoon wind field simulation, information on passing typhoons within a 150-250km range of the simulated observation station is collected. Based on the typhoon motion and intensity information, the typhoon wind speed at the simulated station is calculated using a three-dimensional nonlinear typhoon boundary layer wind field model. The specific process is as follows:

[0031] (1) Using the Holland pressure model, the pressure field at the top of the typhoon boundary layer is established, and the sea level pressure is expressed by the following formula:

[0032] p=p0+Δpexp[-(R max / r) B (6)

[0033] Where: p is the sea level pressure at a radial distance r from the typhoon center; R max B represents the radius of maximum wind speed; B represents the air pressure profile parameters.

[0034] The maximum wind speed radius R is estimated using the Vickery empirical model. max Holland pressure profile parameter B:

[0035]

[0036]

[0037] In the formula, Δp is the central pressure difference; φ is the latitude; fc = 2 × 7.273 × 10⁻⁵ sinφ; and ε B This is a zero-mean random disturbance term that conforms to a normal distribution;

[0038] (2) Establish the basic equations for atmospheric flow in the typhoon boundary layer.

[0039] The Navier-Stokes equations, using cylindrical coordinates and considering the hydrostatic approximation and the incompressibility assumption, are as follows:

[0040]

[0041]

[0042]

[0043]

[0044] Where: (u, v, w) represent the radial, tangential, and vertical wind speeds in the cylindrical coordinate system (r, θ, z), respectively, with the coordinate system moving along with the typhoon center; r is the horizontal distance from the observation station to the typhoon center; θ is the argument of the projection of the line connecting the observation station and the origin, rotated counterclockwise from due east to the origin, onto the horizontal plane; z is the altitude of the observation station; K V It is the vertical eddy viscosity coefficient, calculated by the following formula:

[0045] K V =A -1 / 3 q t l (13)

[0046] Where: l is the turbulence length scale; A is the model parameter, with a value of 25; and q is the turbulence characteristic velocity. t Satisfy the following equation:

[0047]

[0048] Where: B is the model parameter, taken as 0.2; the formula for the turbulence characteristic scale in equation (14) is as follows:

[0049]

[0050]

[0051] Where: α is the model parameter, which is set to 0.1; κ is the von Karman constant, which is set to 0.4;

[0052] (3) Establish boundary conditions for atmospheric flow in the typhoon boundary layer

[0053] At the top of the typhoon boundary layer (z = 2000m), the following boundary conditions are applied:

[0054]

[0055] The vertical distribution of wind speed in the outer region of the typhoon boundary layer (r = 500 km) is described using an Ekman spiral:

[0056]

[0057]

[0058] Where: In the formula: K V =50m 2 / s 2 The gradient wind speed V can be solved by the following formula:

[0059]

[0060] At the bottom of the boundary layer (z = 10m), the following boundary conditions are applied:

[0061]

[0062]

[0063] w = 0 (23)

[0064] q t =A 1 / 3 u * (twenty four)

[0065] Where, ω b The argument of the surface wind speed relative to the radial direction; u e and v e These are the radial and tangential components of the atmospheric velocity relative to the Earth's surface in cylindrical coordinates, respectively; z0 is the surface roughness, which is 0.05 m on land and can be determined by the following formula on the ocean surface.

[0066]

[0067] C D10 It is the surface drag coefficient at a height of 10 meters above sea level, which is determined by the wind speed at that height.

[0068] C D10=(0.49+0.065U) 10 )×10 -3 (26);

[0069] (4) Calculation process

[0070] First, take the vertical wind speed and eddy viscosity coefficient as the calculation results of the previous step, solve the momentum conservation equation according to formula (9) and formula (10) and update the horizontal wind speed; then, based on the updated horizontal wind speed, solve the mass conservation equation according to formula (12) and solve the turbulent kinetic energy dynamic equation according to formula (14) to update the vertical wind speed, turbulent kinetic energy and eddy viscosity coefficient; iterate continuously in this process until the error of the calculation results of the two steps is less than the specified tolerance limit, and the convergence result is obtained.

[0071] Preferably, in step S5, it is necessary to first determine whether the position of the next point in the typhoon's path is within the simulation radius centered on the simulation station. If the straight-line distance between the typhoon center and the center of the simulation station is less than 250km, then step S5 continues; if it is greater than 250km, then the simulation ends.

[0072] Preferably, in step S6, after obtaining the long-term typhoon wind speed process and annual extreme wind speed sequence based on the above simulation process, the annual extreme wind speed sample is fitted to multiple extreme probability distribution functions using the maximum likelihood estimation method, and the optimal annual extreme probability distribution is selected using the AIC criterion, thereby calculating the typhoon wind speeds for each station during the 50-year and 100-year periods.

[0073] The advantages of this invention are:

[0074] (1) The method provided by this invention can improve upon the shortcomings of linear regression analysis in traditional typhoon full-path simulation methods, uncover nonlinear mapping relationships in key historical typhoon information, and reflect the non-Gaussian probability characteristics of typhoon intensity evolution. This method employs a generative adversarial network algorithm based on adversarial self-learning, which can analyze and process historical typhoon data falling within the divided latitude and longitude grid. Through continuous self-learning, it can obtain an ideal regression model between the current path point of the typhoon and the next point in the grid, thereby predicting information about the next point during the typhoon's movement. This method can perform homogeneous analysis based on historical typhoon data from different regions, possessing universality and can be extended to other regions for typhoon risk and disaster assessment.

[0075] (2) The conditional generative adversarial network of this invention, based on game theory, trains a conditional generative model through continuous adversarial learning, enabling it to accurately describe the complex conditional probability distribution model of given data. Therefore, the introduction of conditional generative adversarial network technology in this invention can, on the one hand, improve the shortcomings of traditional linear regression methods in reflecting the nonlinear dependence of typhoon path and intensity changes on environmental variables; on the other hand, it can accurately describe the complex stochastic characteristics of typhoon path and intensity changes, such as non-Gaussian patterns. Based on this method, the evolution process of typhoons in the Northwest Pacific region from occurrence to impact on engineering sites can be simulated, providing sufficient samples of typhoon path and intensity evolution that conform to historical statistical characteristics for typhoon disaster risk assessment. This method ensures good accuracy and universality and can be extended to other regions for typhoon disaster risk assessment and analysis. Attached Figure Description

[0076] Figure 1 This is a flowchart illustrating the technical process of the present invention.

[0077] Figure 2 This is a probability distribution chart of the number of typhoons occurring in historical years, as presented in this invention.

[0078] Figure 3 This is a spatial probability density distribution map of the typhoon initiation points for different months according to the present invention, wherein, Figure 3 (a) is a spatial probability density distribution map of the typhoon initiation points in July; Figure 3 (b) is a spatial probability density distribution map of the typhoon initiation point corresponding to October.

[0079] Figure 4 This is a flowchart of the typhoon movement model simulation of the present invention.

[0080] Figure 5 This is a training structure diagram of the conditional generative adversarial network of the present invention.

[0081] Figure 6 This invention provides a comparison of the simulated typhoon path and intensity with that of an actual typhoon.

[0082] Figure 7 This is the probability density distribution diagram of the present invention, wherein Figure 7 (a) is the probability density distribution of the fitted typhoon's velocity; Figure 7 (b) is the probability density distribution of the fitted typhoon azimuth angle; Figure 7 (c) is a probability density distribution map of the distance between the typhoon center and the engineering site; Figure 7 (d) is the probability density distribution of the pressure difference at the center of the typhoon.

[0083] Figure 8 The flowchart for establishing a nonlinear typhoon boundary layer wind field model for this invention is shown.

[0084] Figure 9 This is a distribution map of annual extreme wind speeds under different return periods according to the present invention. Detailed Implementation

[0085] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0086] like Figure 1-9 As shown, a typhoon full-path simulation method based on conditional generative adversarial networks includes the following specific steps:

[0087] Step 1: Fit the probability distribution of the annual number of typhoons using the Poisson distribution, and sample the number of typhoons within a given period; fit the temporal-spatial joint probability distribution of typhoons using the kernel density estimation method, and sample the time and location of typhoons; fit the probability distribution of the initial intensity of typhoons using the maximum likelihood estimation method, and sample the initial intensity of typhoons.

[0088] Step 2: Based on the China Meteorological Administration's optimal path dataset and NOAA reanalysis data, a random sample generator for typhoon velocity under given environmental airflow conditions is trained using the Wasserstein conditional generative adversarial network method.

[0089] U t =G U (ε U |x U (1)

[0090] Among them: U t =(u t ,v t (x) represents the typhoon's speed at a given moment; U =(u 250 ,v 250 ,u 850 ,v 850 ,φ,λ) T Represents the environmental conditions at that moment; u 250 It is the eastward component of the basic airflow velocity at 250 hPa altitude; v 250 It is the northward component of the basic airflow velocity at 250 hPa altitude; u 850 It is the eastward component of the basic airflow velocity at 850 hPa altitude; v 850 It is the northward component of the basic airflow velocity at 850 hPa altitude; ε U It is a standard normally distributed random variable; λ and φ represent the longitude and latitude of the typhoon center at that moment, respectively; GU It is a generator model, represented by a fully connected neural network.

[0091] A random typhoon velocity generator is used to generate typhoon velocity samples, and the typhoon center position at the next moment is simulated:

[0092]

[0093]

[0094] Where i represents the (i+1)th time; λ i+1 φ i+1 These are the longitude and latitude of the typhoon center at time i+1; R e =6371km, representing the Earth's radius; Δt =6 hours, representing the time interval between time i+1 and time i.

[0095] Step 3: Based on the China Meteorological Administration's optimal path dataset and NOAA reanalysis data, train a generator for ocean typhoon intensity under given environmental conditions using the Wasserstein conditional generative adversarial network method.

[0096] I i+1 =G I (ε I,i |x I,i (4)

[0097] Where: I represents the relative intensity; i represents the (i+1)th time; ε I,i The zero-mean random disturbance term conforming to a normal distribution; x I,i =(λ i ,φ i ,T s,i ,T 0,i ,p 0,i ,S i V t,i ,h m,i ,Γ i ) T Let λ represent the ocean surface environmental conditions at time i; i φ i T represents the longitude and latitude of the typhoon center at time i, respectively; s T0 represents ocean surface temperature; P0 represents tropopause temperature; S represents minimum ocean surface pressure; V represents vertical wind shear; t h represents the speed of the typhoon's movement. m G represents the thickness of the ocean surface mixing layer; Γ represents the ocean sub-mixing temperature layer; I This represents a model for generating ocean typhoon intensity.

[0098] Then, based on the China Meteorological Administration's optimal path dataset and NOAA reanalysis data, a land typhoon intensity generator for given environmental conditions was trained using the Wasserstein conditional generative adversarial network method.

[0099] Δp i+1 =G Δp (ε Δp,i |x Δp,i (5)

[0100] Where: Δp is the central air pressure difference; ε Δp,i The zero-mean random disturbance term conforming to a normal distribution; x Δp,i =(λ i ,φ i ,p 0,i ,S i V t,i ) T G represents the land environment conditions at time i; Δp This represents a land-based typhoon intensity generator model.

[0101] Before proceeding to the third step, the location of the typhoon center obtained in the second step is used to determine whether the typhoon will be located over the ocean or make landfall at the next moment, and different typhoon intensity random generators are used to simulate the evolution of typhoon intensity.

[0102] Step 4: Validate the statistical characteristics of the simulated typhoon path and intensity evolution samples. Set up stations at intervals of 50-100km along the coastal coastline, and then collect important information such as the annual occurrence rate, direction of movement, speed of movement, and central pressure difference of simulated typhoons within a range of 150-250km for each station. Then, use hypothesis testing methods to compare and analyze the data with historical typhoon data.

[0103] Step 5: In the typhoon wind field simulation, information on passing typhoons within a 250km radius of the simulated observation station is collected. Based on information such as typhoon motion and intensity, the typhoon wind speed at the simulated station is calculated using a three-dimensional nonlinear typhoon boundary layer wind field model. The specific process is as follows:

[0104] (1) Establish the pressure field at the top of the typhoon boundary layer

[0105] The Holland pressure model was used.

[0106] p=p0+Δpexp[-(R max / r) B (6)

[0107] Where: p is the sea level pressure at a radial distance r from the typhoon center; R max R represents the radius of maximum wind speed; B represents the pressure profile parameters; where the maximum wind speed radius R is estimated using the Vickery empirical model.max Holland pressure profile parameter B:

[0108]

[0109]

[0110] In the formula, Δp is the central pressure difference; φ is the latitude; fc = 2 × 7.273 × 10⁻⁵ sinφ; and ε B This is a zero-mean random disturbance term that conforms to a normal distribution;

[0111] (2) Establish the basic equations for atmospheric flow in the typhoon boundary layer.

[0112] Using the Navier-Stokes equations in cylindrical coordinates, considering the hydrostatic approximation and the incompressibility assumption:

[0113]

[0114]

[0115]

[0116]

[0117] Where: (u, v, w) are the radial, tangential, and vertical wind speeds in the cylindrical coordinate system (r, θ, z), respectively, which moves along with the typhoon center; r is the horizontal distance from the observation station to the typhoon center; θ is the argument of the projection of the line connecting the observation station and the origin, rotated counterclockwise from due east to the origin, onto the horizontal plane; z is the altitude of the observation station; K V It is the vertical eddy viscosity coefficient, calculated by the following formula:

[0118] K V =A -1 / 3 q t l (13)

[0119] Where: l is the turbulence length scale; A is the model parameter, taken as 25; and q is the turbulence characteristic velocity. t Satisfy the following equation:

[0120]

[0121] Where B is the model parameter, taken as 0.2. The turbulence characteristic scale in equation (14) is calculated by the following formula:

[0122]

[0123]

[0124] Where: α is the model parameter, which is set to 0.1; κ is the von Karman constant, which is set to 0.4;

[0125] (3) Establish boundary conditions for atmospheric flow in the typhoon boundary layer

[0126] At the top of the typhoon boundary layer (z = 2000m), the following boundary conditions are applied:

[0127]

[0128] The vertical distribution of wind speed in the outer region of the typhoon boundary layer (r = 500 km) is described using an Ekman spiral:

[0129]

[0130]

[0131] Where: In the formula: K V =50m 2 / s 2 V is the gradient wind speed, which can be solved by the following formula:

[0132]

[0133] At the bottom of the boundary layer (z = 10 m), the following boundary conditions are applied:

[0134]

[0135]

[0136] w = 0 (23)

[0137] q t =A 1 / 3 u * (twenty four)

[0138] Where, ω b It is the argument of surface wind speed relative to the radial direction; u e and v e Let z0 be the radial and tangential components of the atmospheric velocity relative to the Earth's surface in cylindrical coordinates; z0 is the surface roughness, which is 0.05 m on land and can be determined by the following formula (25) on the ocean surface.

[0139]

[0140] C D10 It is the surface drag coefficient at a height of 10 meters above sea level, which is determined by the wind speed at that height.

[0141] C D10=(0.49+0.065U) 10 )×10 -3 (26);

[0142] (4) Calculation process

[0143] First, the vertical wind speed and eddy viscosity coefficient are taken as the calculation results of the previous step to solve the momentum conservation equation (Equations (9) and (10)) and update the horizontal wind speed. Then, based on the updated horizontal wind speed, the mass conservation equation (Equation (12)) and the turbulent kinetic energy dynamic equation (Equation (14)) are solved to update the vertical wind speed, turbulent kinetic energy, and eddy viscosity coefficient. This process is iterated continuously until the error between the calculation results of the two steps is less than the specified tolerance limit, at which point the convergence result is obtained.

[0144] In the fifth step, it is necessary to determine whether the next point in the typhoon's path is within the simulation radius centered on the simulation station. If the straight-line distance between the typhoon center and the center of the simulation station is less than 250km, then continue with the fifth step; if it is greater than 250km, then end the simulation.

[0145] Step 6: Based on the long-term typhoon wind speed process and annual extreme wind speed sequence obtained from the above simulation process, the annual extreme wind speed sample is fitted to multiple extreme probability distribution functions using maximum likelihood estimation. The optimal annual extreme probability distribution is selected using the AIC criterion, and then the 50-year and 100-year typhoon wind speeds for each station are calculated.

[0146] Example 1

[0147] This invention patent is being used to assess the typhoon disaster risk in the Xiamen area based on measured typhoon data recorded by the China Meteorological Administration, in order to verify the feasibility of the invention. Details are as follows:

[0148] (1) Establish a typhoon origin point information model

[0149] Figure 2 The figure shows the probability distribution of the number of typhoons in historical years fitted with a Poisson distribution. Then, based on the fitted probability distribution model, the number of typhoons in a given period is generated by random sampling.

[0150] By fitting the starting point information of historical typhoons using the kernel density estimation method, the joint temporal-spatial probability distribution of typhoon occurrence can be obtained. Based on the obtained probability distribution model, random sampling is performed to generate the time and location of typhoon occurrence. Figure 3 The figure shows the temporal-spatial joint probability distribution of typhoon occurrence.

[0151] (2) Establish a travel model

[0152] Figure 4The diagram shows the flowchart of the simulated typhoon movement model. Based on the optimal typhoon path dataset recorded by the China Meteorological Administration and NOAA reanalysis data, a random sample generator for typhoon movement velocity under given environmental airflow conditions is trained using the Wasserstein conditional generative adversarial network method. Through continuous random sampling, the latitude and longitude information of the typhoon center position at each moment can be simulated, thus obtaining the entire typhoon's movement path. Figure 5 The diagram shows the training structure of the Wasserstein Conditional Generative Adversarial Network.

[0153] (3) Verification of typhoon track and intensity simulation results

[0154] Figure 6 The image shows a comparison between the actual and simulated paths of Typhoon Gloria recorded by the China Meteorological Administration from 1980 to 2021. The comparison reveals that the simulated path and the historical path are largely consistent in spatial distribution trends.

[0155] (4) Probability distribution test of key typhoon parameters

[0156] Taking Xiamen as an example, an environmental variable model for typhoon occurrence is established based on historical typhoon environmental information. Figure 7 The figure shows the probability distribution of the corresponding key parameters.

[0157] (5) Typhoon Hazard Analysis

[0158] Figure 8 The diagram illustrates the process of establishing a nonlinear typhoon boundary layer wind field model. During the full typhoon path simulation, the typhoon wind field model is activated when the straight-line distance from the typhoon center to the simulation point is less than or equal to 250 km. Maximum likelihood estimation is used to fit the annual extreme wind speed samples to various extreme probability distribution functions. The optimal annual extreme probability distribution is selected using the AIC criterion, and then the 50-year and 100-year typhoon wind speeds at each station are calculated.

[0159] Figure 9 The figure shows the annual extreme wind speed distribution under different return periods.

[0160] The application of Generative Adversarial Networks (GANs) algorithms can help in numerical simulations of random probabilistic events where the exact relationship between input and output data is difficult to determine. Through continuous self-adversarial learning, the gap between output values ​​and actual values ​​can be gradually reduced, ultimately yielding a near-realistic regression model. In typhoon full-path simulation, GANs not only compensate for the shortcomings of traditional autoregressive models in providing the nonlinear relationship between environmental variables and typhoon movement and intensity evolution, but also reflect the non-Gaussian statistical characteristics of typhoon movement and intensity evolution.

[0161] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A typhoon full-path simulation method based on conditional generative adversarial networks, characterized in that, The specific steps are as follows: S1. Establish a typhoon occurrence probability model and generate samples of typhoon occurrence frequency, occurrence time, occurrence location, and initial intensity. S2. Establish a random generator for typhoon movement speed to generate random samples of typhoon movement speed under given airflow conditions at 250hPa and 850hPa altitudes, and simulate the typhoon movement trajectory. S3. Establish a typhoon intensity model, including a random generator for ocean typhoon intensity and a random generator for land typhoon intensity, generate random samples of typhoon intensity changes under given environmental physical variables, and simulate the evolution of typhoon intensity. S4. Verify the simulation effect of typhoon path and intensity statistical characteristics. Based on the randomly generated typhoon path and intensity evolution simulation samples, statistically analyze the basic meteorological parameters of typhoons at various points along the coastal coastline and compare them with historical data. S5. Establish a three-dimensional nonlinear boundary layer wind field model of typhoon, and simulate the surface wind speed process of typhoon by combining typhoon path and intensity samples. S6. Based on the simulated annual extreme wind speed samples of typhoons over many years, establish the probability distribution of annual extreme wind speeds of typhoons on the ground. In step S2, based on historical typhoon data from the past 30-40 years, a random sample generator for typhoon velocity under given environmental airflow conditions is trained using a conditional generative adversarial network (GAN) method. The formula for typhoon velocity is as follows: (1) in: The speed of the typhoon at a given moment; This represents the environmental conditions at that moment; The eastward component of the basic airflow velocity at 250 hPa altitude; The northward component of the basic airflow velocity at 250 hPa altitude; The eastward component of the basic airflow velocity at 850 hPa altitude; The northward component of the basic airflow velocity at 850 hPa altitude; It is a standard normally distributed random variable; This is the longitude of the typhoon's center at that moment; This is the latitude of the typhoon's center at that moment; It is a generator model, represented by a fully connected neural network; In step S3, based on historical typhoon data from the past 30-40 years, a generator for generating ocean typhoon intensity under given environmental conditions is trained using a conditional generative adversarial network method. （4） in: Indicates relative intensity; Indicates the first i +1 moment; This is a zero-mean random disturbance term that conforms to a normal distribution; for Constant ocean surface environmental conditions; For the first Longitude of the typhoon center at any time; For the first Latitude of the typhoon center at any time; The ocean surface temperature; This refers to the tropopause temperature. This is the lowest atmospheric pressure over the ocean surface; This is a vertical wind shear; The speed of the typhoon's movement; This refers to the thickness of the mixed layer on the ocean surface. This is a sub-mixing temperature layer in the ocean; Model for ocean typhoon intensity generator; In step S3, based on historical typhoon data from the past 30-40 years, a land typhoon intensity generator is trained using a conditional generative adversarial network (GAN) method under given environmental conditions. （5） in: The central pressure difference; Indicates the first i +1 moment; This is a zero-mean random disturbance term that conforms to a normal distribution; represent Constant terrestrial environmental conditions; For the first Longitude of the typhoon center at any time; For the first Latitude of the typhoon center at any time; This is the lowest atmospheric pressure over the ocean surface; This is a vertical wind shear; The speed of the typhoon's movement; This represents a land-based typhoon intensity generator model.

2. The typhoon full-path simulation method based on conditional generative adversarial networks according to claim 1, characterized in that: In step S1, the probability distribution of the annual number of typhoons is fitted using a Poisson distribution, and the number of typhoons within a given period is sampled; the joint temporal-spatial probability distribution of typhoons is fitted using a kernel density estimation method, and the time and location of typhoon occurrence are sampled; the probability distribution of the initial intensity of typhoons is fitted using a maximum likelihood estimation method, and samples of the initial intensity of typhoons are sampled.

3. The typhoon full-path simulation method based on conditional generative adversarial networks according to claim 1, characterized in that: A random sample generator for typhoon velocity is used to generate random samples of typhoon velocity, and the location of the typhoon center at the next moment is simulated: （2） （3） in, Indicates the first i +1 moment; For the first i Typhoon center longitude at +1 o'clock; For the first i +1 hour latitude of the typhoon center; , representing the Earth's radius; , representing the i +1 moment and the i The time interval between moments.

4. The typhoon full-path simulation method based on conditional generative adversarial networks according to claim 3, characterized in that: Before proceeding to step S3, the location of the typhoon center obtained in step S2 is used to determine whether the typhoon will be located over the ocean or make landfall at the next moment, and the corresponding typhoon intensity generator is used to simulate the evolution of the typhoon intensity.

5. The typhoon full-path simulation method based on conditional generative adversarial networks according to claim 1, characterized in that: In step S4, the historical data consists of historical meteorological parameters of typhoons at various measuring points along the coastal coastline over the past 30-40 years. To verify the statistical characteristics of the simulated typhoon movement path and intensity evolution samples, stations are set up along the coastal coastline at intervals of 50-100 km. Then, the annual occurrence rate, movement direction, movement speed, and central pressure difference of simulated typhoons within a specified radius of each station are statistically analyzed. The KS test method is then used to compare and analyze the data with the historical typhoon data. If the test results meet the requirements of a confidence level of 0.95, the model can be used for subsequent analysis. Otherwise, the hyperparameters and structure of the neural network need to be adjusted and retrained.

6. The typhoon full-path simulation method based on conditional generative adversarial networks according to claim 1, characterized in that, In step S5, during the typhoon wind field simulation, information on passing typhoons within a 250km radius of the simulated observation station is collected. Based on the typhoon motion and intensity information, the typhoon wind speed at the simulated station is calculated using a three-dimensional nonlinear typhoon boundary layer wind field model. The specific process is as follows: (1) Using the Holland pressure model, the pressure field at the top of the typhoon boundary layer is established, and the sea level pressure is expressed by the following formula: (6) in: The radial distance from the typhoon center is Sea level pressure at that location; The radius of maximum wind speed; These are the parameters of the air pressure profile; The maximum wind speed radius was estimated using the Vickery empirical model. Holland pressure profile parameters : (7) (8) In the formula The central pressure difference; Latitude; ; and This is a zero-mean random disturbance term that conforms to a normal distribution; (2) Establish the basic equations for atmospheric flow in the typhoon boundary layer. The Navier-Stokes equations, using cylindrical coordinates and considering the hydrostatic approximation and the incompressibility assumption, are as follows: (9) (10) (11) (12) in: They are cylindrical coordinate systems The radial wind speed, tangential wind speed, and vertical wind speed in the coordinate system move along with the typhoon center; r To investigate the horizontal distance from the station to the typhoon center; It is the argument of the projection onto the horizontal plane of the line connecting the origin and the observation station, which is rotated counterclockwise from due east. It is to examine the height of the site; It is the vertical eddy viscosity coefficient, calculated by the following formula: (13) in: For turbulent length scale; The model parameter is set to 25; the turbulent characteristic velocity. Satisfy the following equation: (14) in: The model parameter is set to 0.2; the formula for the turbulence characteristic scale in equation (14) is as follows: (15) (16) in: The model parameter is set to 0.1; is the von Karman constant, which is set to 0.4; (3) Establish boundary conditions for atmospheric flow in the typhoon boundary layer At the top of the typhoon boundary layer ( The following boundary conditions are adopted: (17)； The outer region of the typhoon boundary layer ( The vertical distribution of wind speed is described using an Ekman spiral: (18) (19) Where: In the formula: , ; The gradient wind speed can be solved by the following formula: (20)； At the bottom of the boundary layer ( The following boundary conditions are adopted: (21) (22) (23) (24) in, The argument of the surface wind speed relative to the radial direction; and These are the radial and tangential components of the velocity of atmospheric motion relative to the Earth's surface in cylindrical coordinates, respectively. For surface roughness, the value is 0.05m on land and can be determined by the following formula on ocean surface; (25) It is the surface drag coefficient at a height of 10 meters above sea level, which is determined by the wind speed at that height. (26)； (4) Calculation process First, take the vertical wind speed and eddy viscosity coefficient as the calculation results of the previous step, solve the momentum conservation equation according to formula (9) and formula (10) and update the horizontal wind speed; then, based on the updated horizontal wind speed, solve the mass conservation equation according to formula (12) and solve the turbulent kinetic energy dynamic equation according to formula (14) to update the vertical wind speed, turbulent kinetic energy and eddy viscosity coefficient; iterate continuously in this process until the error of the calculation results of the two steps is less than the specified tolerance limit and the convergence result is obtained.

7. The typhoon full-path simulation method based on conditional generative adversarial networks according to claim 1, characterized in that: In step S5, it is necessary to first determine whether the position of the next point in the typhoon's path is within the simulation radius centered on the simulation station. If the straight-line distance between the typhoon center and the center of the simulation station is less than 250km, then continue with step S5; if it is greater than 250km, then end the simulation. In step S6, based on the long-term typhoon wind speed process and annual extreme wind speed sequence obtained from the simulation process, the maximum likelihood estimation method is used to fit the annual extreme wind speed sample into multiple extreme probability distribution functions, and the AIC criterion is used to select the optimal annual extreme probability distribution, and then the typhoon wind speed of each station is calculated.

Images