Typhoon disaster assessment method based on wrf adaptive nesting and machine learning

Abstract

The application discloses a typhoon disaster evaluation method based on WRF adaptive nesting and machine learning, selects a typical typhoon example of a research area, collects and pre-processes meteorological, disaster loss and exposure data during the typhoon, evaluates multiple candidate thresholds under the constraint of an existing typhoon disaster wind speed threshold, optimally determines a disaster-causing threshold, constructs a multi-layer WRF nesting to downscale a wind field, identifies a region exceeding the threshold along a typhoon path according to the determined disaster-causing threshold, and performs high-resolution encryption solving in the region, re-selects the region exceeding the threshold and obtains a regional maximum wind speed, matches the region exceeding the threshold with exposure data space to form exposure, and takes the regional maximum wind speed and the exposure as input and historical loss rate as output, so as to establish an intelligent prediction model of loss rate of vulnerability quantization, and calculates a typhoon disaster risk by comprehensively considering disaster intensity, exposure and vulnerability. The application can effectively improve the accuracy of typhoon disaster risk evaluation, and provides a reliable basis for graded early warning and emergency decision-making.

Description

Technical Field

[0001] This invention relates to the field of natural disaster risk assessment, and combines adaptive nesting of numerical weather prediction models (WRF) with machine learning methods to carry out typhoon process wind field downscaling and disaster risk integrated assessment. Specifically, it relates to a typhoon disaster assessment method based on WRF adaptive nesting and machine learning. Background Technology

[0002] Typhoons are characterized by their sudden onset, mobility, and significant spatial variability, making them one of the most prominent meteorological disasters affecting coastal and some inland areas. The accompanying strong winds exhibit marked temporal and spatial heterogeneity, often leading to loss of life and property, transportation disruptions, and supply chain disruptions. Engineering and operational practices demonstrate that the disaster risk of typhoons is not solely determined by extreme wind speeds, but rather by the combined effects of disaster intensity, exposure level, and vulnerability. However, current risk assessments still primarily rely on experience and statistical assumptions, lacking a detailed characterization of disaster intensity, and utilize meteorological data with low spatial resolution, making it difficult to accurately represent the actual distribution characteristics of near-surface extreme winds.

[0003] Downscaling in numerical weather prediction (WRF) models has become a primary method for reconstructing near-surface wind fields during typhoon events, significantly improving the accuracy of strong wind impact areas. However, existing applications often employ fixed nesting or fixed-resolution schemes, lacking adaptive downscaling capabilities that automatically focus on the typhoon's path and impact range. This makes it difficult to achieve on-demand densification during critical periods and in key areas, resulting in insufficient specificity in intensity representation and a trade-off between computational efficiency and timeliness. Furthermore, the determination of wind speed thresholds lacks a systematic process based on historical evidence. While areas exceeding the threshold are key disaster concern areas, the identification criteria lack uniformity and traceability. Densified simulations of these areas can yield more accurate disaster intensity distributions, providing higher-resolution input for risk identification. However, intensity information alone is insufficient for risk assessment; exposure and vulnerability also play crucial roles. Exposure needs to be matched to the spatial distribution of population and economic activities, ensuring strict alignment with wind field results on the spatial grid to accurately reflect the distribution characteristics of affected objects. Vulnerability quantification typically relies on vulnerability curves to describe the relationship between wind speed and loss rate.

[0004] In characterizing vulnerability curves, conventional methods often rely on historical disaster damage samples or expert experience to fit curves, describing the correlation between wind speed and loss rate for rapid estimation. While simple to implement, this method is highly sensitive to sample size and representativeness; the curve shape and threshold selection are significantly influenced by human choice, making it difficult to objectively account for differences across regions and objects. To alleviate these limitations and improve reusability and transferability, it is necessary to introduce a machine learning modeling approach targeting historical loss rates at a unified resolution. This involves using extreme wind speeds and exposure features such as population and GDP as inputs, and establishing a relationship between these features and loss rates through machine learning to form an updatable vulnerability quantification. However, the above methods have not yet been systematically studied and applied in refined typhoon risk assessment. Therefore, there is an urgent need for a method based on WRF adaptive nesting and machine learning to conduct comprehensive assessment and risk identification research from the perspective of co-modeling intensity, exposure, and vulnerability. Summary of the Invention

[0005] To address the problems in existing typhoon risk assessments, such as reliance on fixed schemes for intensity characterization, lack of adaptive focusing that evolves with the process, subjective disaster thresholds, and reliance on static curves that make vulnerability difficult to transfer, this invention proposes a method based on WRF adaptive nesting and machine learning. Through threshold optimization and adaptive encryption, intensity and exposure matching, and machine learning training with loss rate as the target, an integrated assessment process is constructed to improve accuracy and stability and support early warning applications.

[0006] This invention is a typhoon disaster assessment method based on WRF adaptive nesting and machine learning, the method specifically includes the following steps:

[0007] S1. Select typical typhoon cases in the study area, collect meteorological data, disaster loss data and corresponding population / GDP exposure data of the study area during the typhoon, and perform unified preprocessing of coordinates and resolution to obtain historical samples.

[0008] S2, Under the constraint of the existing typhoon disaster wind speed threshold, use the historical samples collected in step S1 to conduct a systematic evaluation of multiple candidate wind speed thresholds and determine the disaster-causing wind speed threshold.

[0009] S3. Construct a multi-layered nested numerical weather prediction model WRF downscaled wind field. Identify the range exceeding the threshold along the typhoon path according to the disaster-causing wind speed threshold determined in step S2. Within this range, perform higher resolution finer grid densification solution. Based on this, reselect the range exceeding the threshold and obtain the maximum wind speed, which is recorded as the disaster intensity.

[0010] S4. Spatially match the over-threshold range obtained from step S3 with the population / GDP exposure data collected in step S1 to form the exposure level. Calculate the historical loss rate based on the disaster loss data and the exposure level. Construct a sample set with the maximum wind speed and exposure level as input and the historical loss rate as output. Divide the sample set into a training set and a test set. Use machine learning to train the loss rate prediction model. The predicted loss rate is the quantitative result of vulnerability.

[0011] S5. Based on the disaster intensity obtained in step S3, the exposure degree collected in step S1 and spatially matched in step S4, and the loss rate predicted in step S4, i.e. the vulnerability quantification result, the typhoon disaster risk is calculated comprehensively.

[0012] in,

[0013] In step S1, the meteorological data, disaster loss and exposure data of the study area during the typhoon are obtained based on publicly available datasets from meteorological and statistical agencies; the disaster loss data includes spatial information on population loss and GDP loss, and the exposure data includes spatial information on population and GDP; all the acquired data need to be spatially aligned and processed to the same spatial resolution and geographic coordinate system.

[0014] Step S2 specifically involves:

[0015] S21, based on existing strong wind business standards and historical experience, first determine the search range and step size for candidate thresholds:

[0016]

[0017]

[0018] In the formula, It is a set of wind speed thresholds, where wind speed refers to the extreme wind speed at a height of 10m. It is any candidate threshold in the set; These are the lower and upper limits of the threshold, respectively. This is the search step size; Non-negative integer index; This is the upper bound of the number of steps.

[0019] S22, for each historical event Let the maximum extreme wind speed at a height of 10m during this event be denoted as . Given any threshold To determine whether a disaster occurred in a historical event:

[0020]

[0021] in, For threshold-based Historical events The determination of whether a disaster has occurred will be made by comparing the assessment of whether a disaster has occurred with the actual situation, as shown in the table below:

[0022]

[0023] in, Indicates whether a disaster actually occurred. =1 indicates that a disaster actually occurred. =0 indicates that no disaster actually occurred. Based on threshold Historical events It was determined that a disaster had occurred, and indeed, a disaster had occurred. Based on threshold Historical events A disaster was presumed to have occurred, but in reality, no disaster did occur. Based on threshold Historical events The determination was made that no disaster occurred, but the actual situation indicated that a disaster had taken place. Based on threshold Historical events It was determined that no disaster occurred, and in reality, no disaster did occur.

[0024] The comprehensive performance index is used as the core evaluation criterion for threshold optimization, and its calculation formula is as follows:

[0025]

[0026] in and These represent the threshold values ​​respectively. The formulas for the accuracy and recall rate in determining whether a disaster has occurred are as follows:

[0027]

[0028]

[0029] exist Calculate one by one ,Pick The one with the largest value is optimal.

[0030]

[0031] in, This is the final calculated threshold wind speed that could cause disaster.

[0032] Step S3 is as follows:

[0033] S31. Establish a fixed three-layer nested structure within the study area, namely the outermost layer, the middle layer, and the inner layer. Set the horizontal resolution of the outermost layer, the middle layer, and the inner layer to 1:3:3, where the middle layer is a subdomain of the outermost layer and the inner layer is a subdomain of the middle layer. The boundary and initial field are provided by the meteorological data collected in step S1. The start and end time of the WRF calculation covers the entire process of the target typhoon.

[0034] S32, based on the disaster-causing wind speed threshold The extreme wind speed results at a height of 10m obtained from the inner layer simulation of S31 are used to determine the region exceeding the threshold; based on the inner layer wind field obtained from S31, at any output time... Grid The extreme wind speed at 10m is Where i is the index of the grid point in the east-west direction, j is the index of the grid point in the north-south direction, and t is any time. The region exceeding the threshold is expressed as follows:

[0035]

[0036] in, The set of super-threshold regions; to conduct higher resolution simulations in the super-threshold regions, the super-threshold grid point set obtained by equation (8) is converted into multiple rectangular computational domains aligned with the inner layer grid lines; taking each rectangular computational domain as the target, multiple corresponding inner layer nests are set up within the simulation range of the inner layer. The horizontal grid size of each inner layer nest is fixed at 100 grid points in the east-west direction and 100 grid points in the north-south direction. By adjusting the position of the inner layer nests, on the one hand, they fall entirely within the simulation range of the inner layer, and on the other hand, the corresponding super-threshold regions are completely wrapped within the computation range of the inner layer nests in physical space. The horizontal grid resolution ratio between the inner layer and the inner layer nests is 1:3; high-resolution densification is performed in the computational domain of the inner layer nests to obtain the wind speed field at a height of 10 m; subsequently, the threshold is re-determined for the inner layer nest simulation results, and the super-threshold regions are re-selected. Their spatial definition is:

[0037]

[0038] In the formula, This refers to the region exceeding the threshold obtained through multiple selections. Represents refined grid points In time The extreme wind speed at a height of 10m. The disaster-causing wind speed threshold determined in step S2; the total wind speed exceeding the threshold area obtained by multiple selection at each moment during the entire typhoon process is denoted as... ;

[0039] Finally, for each grid point within the selected threshold region... Find the extreme value of the time series to obtain the maximum wind speed at that grid point:

[0040]

[0041] Among them, the over-threshold region obtained by multiple selection only accounts for a part of the inner nesting. For ease of description, it will be... This represents the east-west index of the grid points within the selected threshold region. This is the index of the grid points in the selected region in the north-south direction.

[0042] Step S4 specifically involves:

[0043] S41, the total over-threshold region obtained by re-selection The maximum wind speed, population, and GDP data within the region are unified to the same spatial grid to form a data field with consistent resolution. At this point, the total population exposure and total GDP exposure of the region are denoted as follows: and The calculation formula is as follows:

[0044]

[0045] in, and These represent the population exposure and GDP exposure of a single grid cell within the threshold area, respectively.

[0046] S42, based on unified grid data, extracts the maximum wind speed of each unit within the selected threshold area. This data is then correlated with the corresponding population exposure and GDP exposure to form a sample unit set; the input and output of each sample unit are defined as follows:

[0047]

[0048] in, This represents the input feature vector, which includes the maximum wind speed and exposure level. The output vector is represented by the corresponding historical loss rate; and the historical loss rate consists of two parts: the population loss rate and the GDP loss rate, calculated by the following formula:

[0049]

[0050]

[0051] in, It concerns the historical rate of population loss. It is the region exceeding the threshold. In the data on population loss in internal disasters, the data is unified to the same spatial grid. It concerns the historical rate of GDP loss. It is the region exceeding the threshold. Data on GDP losses in internal disaster loss are unified into the same spatial grid data;

[0052] S43. The sample set is divided into a training set and a test set. A machine learning algorithm is used to train the loss rate prediction model. The model takes the maximum wind speed and exposure level as input and the historical loss rate as output, in the following form:

[0053]

[0054] in For machine learning models, For model parameters, The loss rate predicted by the machine learning model;

[0055] The training objective is to minimize the prediction loss function:

[0056]

[0057] After training, a loss rate prediction model is obtained, which predicts the corresponding loss rate given exposure level and wind speed, for the quantification of vulnerability. The predicted loss rate consists of two components: population loss rate and GDP loss rate.

[0058]

[0059] in, This represents the population loss rate output by the prediction model. This represents the GDP loss rate output by the prediction model.

[0060] In step S5, the typhoon disaster risk is calculated comprehensively based on the disaster intensity obtained in step S3, the exposure level collected in step S1 and spatially matched in step S4, and the loss rate predicted in step S4. Risk is defined as the combined effect of disaster intensity, exposure level, and vulnerability (loss rate). The risk is determined at any grid point within the selected threshold area (high-risk area). The disaster is expressed as follows:

[0061]

[0062]

[0063] In the formula, This represents the population risk value of a grid cell. The GDP risk value of each grid cell, summed over the entire study area, yields the overall risk.

[0064]

[0065]

[0066] in, Indicates overall population risk. This indicates the overall risk to GDP.

[0067] Compared with the prior art, the present invention has the following advantages:

[0068] (1) A wind speed threshold selection method is proposed. Candidate thresholds are evaluated and compared in combination with historical event data. The threshold that best represents the disaster-causing characteristics of typhoons is selected to improve the scientificity and consistency of risk identification.

[0069] (2) Establish an adaptive nesting mechanism to automatically focus on key areas based on the typhoon path and impact range, realize on-demand encryption and dynamic updates, and effectively improve the spatial resolution and simulation accuracy of the near-surface wind field;

[0070] (3) Introduce machine learning models for loss rate prediction, take extreme wind speed and population and GDP exposure as inputs, and take historical loss rate as the target to establish a data-driven vulnerability quantification model to improve the stability and generalization ability of the prediction results.

[0071] (4) Construct a unified grid-based assessment system for intensity, exposure and vulnerability, so as to achieve continuous connection from threshold determination, wind field simulation to risk calculation, and output intuitive and reusable spatialized risk results to support graded early warning and emergency decision-making. Attached Figure Description

[0072] Figure 1 A flowchart of a typhoon disaster assessment method based on WRF adaptive nesting and machine learning;

[0073] Figure 2 This is a schematic diagram of WRF downscaling nesting;

[0074] Figure 3 A schematic diagram of the refined simulation framework for the super-threshold region;

[0075] Figure 4This diagram illustrates how to train a machine learning program to predict the loss rate. Detailed Implementation

[0076] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0077] This application proposes a typhoon disaster assessment method based on WRF adaptive nesting and machine learning. The main technical process of the implementation scheme is as follows (see...). Figure 1 ):

[0078] S1. Select typical typhoon cases in the study area, collect meteorological data, disaster loss data and population / GDP exposure data in the corresponding area during the typhoon, and perform unified preprocessing of coordinates and resolution.

[0079] Specifically, the meteorological data, disaster loss and exposure data of the study area during the typhoon were mainly obtained from publicly available datasets from meteorological and statistical agencies. The disaster loss data included spatial information on population loss and GDP loss, and the exposure data included spatial information on population and GDP. All the acquired data needed to be spatially aligned and processed to the same spatial resolution and geographic coordinate system.

[0080] S2, Under the constraint of the existing typhoon disaster wind speed threshold, use the historical samples collected in step S1 to conduct a systematic evaluation of multiple candidate wind speed thresholds and determine the disaster-causing wind speed threshold.

[0081] Specifically, under the constraint of existing typhoon disaster wind speed thresholds, a systematic evaluation of multiple candidate wind speed thresholds is conducted using historical samples to select the optimal disaster-causing wind speed threshold. The specific steps are as follows:

[0082] S21, based on existing strong wind business standards and historical experience, first determine the search range and step size for candidate thresholds:

[0083]

[0084]

[0085] In the formula, It is a set of wind speed thresholds, where wind speed refers to the extreme wind speed at a height of 10m. This refers to any candidate threshold in the set; These are the lower and upper limits of the threshold, respectively. This is the search step size; Non-negative integer index; This is the upper bound of the number of steps.

[0086] S22, for each historical event Let the maximum extreme wind speed at a height of 10m during this event be denoted as . Given any threshold To determine whether a disaster occurred in a historical event:

[0087]

[0088] in, For threshold-based Historical events The determination of whether a disaster has occurred will be made by comparing the assessment of whether a disaster has occurred with the actual situation, as shown in the table below:

[0089]

[0090] in, Indicates whether a disaster actually occurred. =1 indicates that a disaster actually occurred. =0 indicates that no disaster actually occurred. Based on threshold Historical events It was determined that a disaster had occurred, and indeed, a disaster had occurred. Based on threshold Historical events A disaster was presumed to have occurred, but in reality, no disaster did occur. Based on threshold Historical events The determination was made that no disaster occurred, but the actual situation indicated that a disaster had taken place. Based on threshold Historical events It was determined that no disaster occurred, and in reality, no disaster did occur.

[0091] The comprehensive performance index is used as the core evaluation criterion for threshold optimization, and its calculation formula is as follows:

[0092]

[0093] in and These represent the threshold values ​​respectively. The formulas for the accuracy and recall rate in determining whether a disaster has occurred are as follows:

[0094]

[0095]

[0096] exist Calculate one by one ,Pick The one with the largest value is optimal.

[0097]

[0098] in, This is the final calculated threshold wind speed that could cause disaster.

[0099] S3. Construct a multi-layered nested WRF downscaling wind field. Identify the regions exceeding the threshold along the typhoon path according to the disaster-causing wind speed threshold determined in step S2. Then, perform higher resolution finer mesh refinement within the regions exceeding the threshold and select the regions exceeding the threshold to obtain the maximum wind speed, which is recorded as the disaster intensity.

[0100] Specifically, a multi-layered nested WRF downscaling wind field is constructed. Regions exceeding the threshold are identified along the typhoon path according to the threshold determined in step S2. Within this range, a higher resolution, finer mesh is used for solution. Based on this, regions exceeding the threshold are reselected, and the maximum wind speed is obtained. The specific steps are as follows:

[0101] S31. Establish a fixed three-layer nested structure within the study area: outermost d01, middle d02, and innermost d03. Set the horizontal resolution of the outermost d01, middle d02, and innermost d03 to the parent-child mesh ratio of 1:3:3. The middle d02 is a subdomain of the outermost d01, and the innermost d03 is a subdomain of the middle d02 (see [reference]). Figure 2 The boundary and initial field are provided by the meteorological data obtained in step S1; the start and end time of the WRF calculation covers the entire process of the target typhoon.

[0102] S32, based on the disaster-causing wind speed threshold The extreme wind speed at a height of 10m is simulated using the inner layer d03 obtained from S31 to determine candidate areas exceeding the threshold. Based on the inner layer d03 wind field obtained from S31, at any discrete output time... Grid The extreme wind speed at 10m is , This represents the east-west index of the grid point. This represents the north-south index of the grid point. For any given time, the region exceeding the threshold can be represented as follows:

[0103]

[0104] in, The set of super-threshold regions; to conduct higher resolution simulations in the super-threshold regions, the set of super-threshold grid points obtained by equation (8) is converted into multiple rectangular computational domains aligned with the inner d03 grid lines. Taking each rectangular computational domain as the target, multiple corresponding inner nested d04s are deployed within the simulation range of the inner d03 (see Figure 2 Each d04 has a fixed horizontal grid size of 100 grid points in the east-west direction and 100 grid points in the north-south direction. The position of d04 is adjusted so that it falls entirely within the simulation range of the inner d03 layer, and physically encompasses the corresponding overthreshold regions within the computational domain of the nested inner d04 layers. The horizontal grid resolution ratio of the inner d03 to the nested inner d04 layers is 1:3. High-resolution, finely detailed solutions are performed within the computational domain of the nested inner d04 layers to obtain the wind speed field at a height of 10 m. Subsequently, the simulation results of the nested inner d04 layers are re-evaluated for threshold determination, and overthreshold regions are reselected (see...). Figure 3 Its space is defined as:

[0105]

[0106] In the formula, The region exceeding the threshold obtained by multiple selection at any given time; Represents refined grid points In time The extreme wind speed at a height of 10 m, The disaster-causing wind speed threshold determined in step S2; the total exceedance area obtained by multiplexing at each moment during the entire typhoon process is ;

[0107] Finally, for each grid point within the selected threshold region... Find the extreme values ​​of the time series to obtain the maximum wind speed at that grid point:

[0108]

[0109] Among them, the super-threshold region obtained by re-selection only accounts for a part of d04. For ease of description, it will be... This represents the east-west index of the grid points within the selected threshold region. This is the index of the grid points in the selected region in the north-south direction;

[0110] S4. Spatially match the super-wind speed threshold area obtained from step S3 with the population / GDP exposure data collected in step S1 to form the exposure level. Calculate the historical loss rate based on the disaster loss data and the exposure level. Construct a sample set with the maximum wind speed and exposure level as input and the historical loss rate as output. Divide the sample set into a training set and a test set. Use machine learning to train the loss rate prediction model. The predicted loss rate is the quantitative result of vulnerability.

[0111] Specifically, the super-wind speed threshold areas obtained from step S3 are spatially matched with the population / GDP exposure grid collected in step S1 to form the exposure level. Historical loss rates are calculated based on disaster loss data and exposure levels. A sample set is constructed using maximum wind speed and exposure levels as inputs and historical loss rates as outputs. This sample set is then divided into a training set and a test set. A loss rate prediction model is trained using machine learning (see...). Figure 4 The specific steps are as follows:

[0112] S41, the total over-threshold region obtained by re-selection The maximum wind speed and population / GDP data within the region are unified to the same spatial grid, forming a data field with consistent resolution. At this point, the total population exposure and total GDP exposure for the region are denoted as follows: and The calculation formula is as follows:

[0113]

[0114] in, and These represent the population exposure and GDP exposure of a single grid cell within the threshold area, respectively.

[0115] S42, based on unified grid data, extracts the maximum wind speed of each unit within the selected threshold area. This data is then correlated with the corresponding population exposure and GDP exposure to form a sample unit set. The input and output of each sample unit are defined as follows:

[0116]

[0117] in, This represents the input feature vector, which includes the maximum wind speed and exposure level. The output vector is represented by the corresponding historical loss rate; and the historical loss rate consists of two parts: the population loss rate and the GDP loss rate, calculated by the following formula:

[0118]

[0119]

[0120] in, It concerns the historical rate of population loss. It is the total over-threshold region In the data on population loss in internal disasters, the data is unified to the same spatial grid. It concerns the historical rate of GDP loss. It is the total over-threshold region In the internal disaster loss data, GDP losses are unified to the same spatial grid. S43, the sample set is divided into training and testing sets, and a machine learning algorithm is used to train the loss rate prediction model. The model takes maximum wind speed and exposure as input and historical loss rate as output, in the following form:

[0121]

[0122] in For machine learning models, For model parameters, The loss rate predicted by the machine learning model.

[0123] The training objective is to minimize the prediction loss function:

[0124]

[0125] After training, a loss rate prediction model can be obtained, which can predict the corresponding loss rate given exposure level and wind speed, and is used for the quantification of vulnerability. The predicted loss rate consists of two components: population loss rate and GDP loss rate.

[0126]

[0127] in, This represents the population loss rate output by the prediction model. This represents the GDP loss rate output by the prediction model.

[0128] S5. Based on the disaster intensity (maximum wind speed) obtained in step S3, the exposure rate collected in step S1 and spatially matched in step S4, and the loss rate predicted in step S4 (i.e. vulnerability quantification results), the typhoon disaster risk is calculated comprehensively.

[0129] Specifically, risk is defined as the combined effect of disaster intensity, exposure, and vulnerability (i.e., loss rate). The disaster at any grid point within the threshold zone, i.e., the high-risk zone, is expressed as follows:

[0130]

[0131]

[0132] In the formula, This represents the population risk value of a grid cell. The GDP risk value of each grid cell, summed over the entire study area, yields the overall risk.

[0133]

[0134]

[0135] in, Indicates overall population risk. This indicates the overall risk to GDP.

[0136] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A typhoon disaster assessment method based on WRF adaptive nesting and machine learning, characterized in that, Includes the following steps: S1. Select typical typhoon cases in the study area, collect meteorological data, disaster loss data and corresponding population / GDP exposure data of the study area during the typhoon, and perform unified preprocessing of coordinates and resolution to obtain historical samples. S2, Under the constraint of the existing typhoon disaster wind speed threshold, use the historical samples collected in step S1 to conduct a systematic evaluation of multiple candidate wind speed thresholds and determine the disaster-causing wind speed threshold. S3. Construct a multi-layered nested numerical weather prediction model WRF downscaled wind field. Identify the range exceeding the threshold along the typhoon path according to the disaster-causing wind speed threshold determined in step S2. Within this range, perform higher resolution finer grid densification solution. Based on this, reselect the range exceeding the threshold and obtain the maximum wind speed, which is recorded as the disaster intensity. S4. Spatially match the over-threshold range obtained from step S3 with the population / GDP exposure data collected in step S1 to form the exposure level. Calculate the historical loss rate based on the disaster loss data and the exposure level. Construct a sample set with the maximum wind speed and exposure level as input and the historical loss rate as output. Divide the sample set into a training set and a test set. Use machine learning to train the loss rate prediction model. The predicted loss rate is the quantitative result of vulnerability. S5. Based on the disaster intensity obtained in step S3, the exposure degree collected in step S1 and spatially matched in step S4, and the loss rate predicted in step S4, i.e. the vulnerability quantification result, the typhoon disaster risk is calculated comprehensively. Specifically, step S3 includes the following steps: S31. Establish a fixed three-layer nested structure within the study area: outermost d01, middle d02, and innermost d03. Set the horizontal resolution of the outermost d01, middle d02, and innermost d03 to 1:3:3, where the middle d02 is a subdomain of the outermost d01, and the innermost d03 is a subdomain of the middle d02. The boundary and initial field are provided by the meteorological data collected in step S1. The start and end time of the WRF calculation covers the entire process of the target typhoon. S32, based on the disaster-causing wind speed threshold The extreme wind speed at a height of 10m is simulated using the inner layer d03 obtained from S31 to determine the region exceeding the threshold; based on the wind field of the inner layer d03 obtained from S31, at any output time... Grid The extreme wind speed at 10m is Where i is the index of the grid point in the east-west direction, j is the index of the grid point in the north-south direction, and t is any time. The region exceeding the threshold is expressed as follows: ； （1） in, The set of super-threshold regions; to conduct higher resolution simulations in the super-threshold regions, the super-threshold grid point set obtained by equation (1) is converted into multiple rectangular computational domains aligned with the inner d03 grid lines; taking each rectangular computational domain as the target, multiple corresponding inner nested d04s are deployed within the simulation range of the inner d03. The horizontal grid size of each inner nested d04 is fixed at 100 grid points in the east-west direction and 100 grid points in the north-south direction. By adjusting the position of the inner nested d04, it is made to fall within the simulation range of the inner d03 on the one hand, and to completely wrap the corresponding super-threshold regions within the computational range of the inner nested d04 in physical space on the other hand. The horizontal grid resolution ratio of the inner d03 and the inner nested d04 is 1:3; high-resolution densification is performed in the computational domain of the inner nested d04 to obtain the wind speed field at a height of 10 m; then, the threshold is re-determined for the simulation results of the inner nested d04, and the super-threshold regions are re-selected. Its spatial definition is: ； （2） In the formula, This refers to the region exceeding the threshold obtained through multiple selections. Represents refined grid points In time The extreme wind speed at a height of 10m. The disaster-causing wind speed threshold determined in step S2; the total wind speed exceeding the threshold area obtained by multiple selection at each moment during the entire typhoon process is denoted as... ; Finally, for each grid point within the selected threshold region... Find the extreme value of the time series to obtain the maximum wind speed at that grid point: ； （3） Among them, the over-threshold region obtained by multiple selection only accounts for a part of the inner nested d04. For ease of description, it will be... This represents the east-west index of the grid points within the selected threshold region. This is the index of the grid points in the selected region in the north-south direction.

2. The typhoon disaster assessment method based on WRF adaptive nesting and machine learning according to claim 1, characterized in that, In step S1, the meteorological data, disaster loss and exposure data of the study area during the typhoon are obtained based on publicly available datasets from meteorological and statistical agencies; the disaster loss data includes spatial information on population loss and GDP loss, and the exposure data includes spatial information on population and GDP; all the acquired data need to be spatially aligned and processed to the same spatial resolution and geographic coordinate system.

3. The typhoon disaster assessment method based on WRF adaptive nesting and machine learning according to claim 2, characterized in that, Step S2 specifically involves: S21, based on existing strong wind business standards and historical experience, first determine the search range and step size for candidate thresholds: ； （4） ； （5） In the formula, It is a set of wind speed thresholds, where wind speed refers to the extreme wind speed at a height of 10m. It is any candidate threshold in the set; These are the lower and upper limits of the threshold, respectively. This is the search step size; Non-negative integer index; The upper bound of the number of steps; S22, for each historical event Let the maximum extreme wind speed at a height of 10m during this event be denoted as . Given any threshold To determine whether a disaster occurred in a historical event: ； （6） in, For threshold-based Historical events The result of determining whether a disaster has occurred; Further with Indicates whether a disaster actually occurred, among which =1 indicates that a disaster actually occurred. =0 indicates that no disaster actually occurred, and the determination of whether a disaster occurred is compared with the actual situation, i.e., based on a threshold. Historical events The determination is that if a disaster is determined and the actual situation also corresponds to a disaster. If a disaster is determined to exist but the actual situation is not, it corresponds to... If it is determined that there is no disaster but the actual situation indicates a disaster, it corresponds to... If it is determined that there is no disaster and the actual situation also shows no disaster, then... ; The comprehensive performance index is used as the core evaluation criterion for threshold optimization, and its calculation formula is as follows: ； （7） in and These represent the threshold values ​​respectively. The formulas for the accuracy and recall rate in determining whether a disaster has occurred are as follows: ； （8） ； （9） exist Calculate one by one ,Pick The one with the largest value is the optimal one. ； （10） in, This refers to the final calculated threshold wind speed that could cause a disaster.

4. The typhoon disaster assessment method based on WRF adaptive nesting and machine learning according to claim 1, characterized in that, Step S4 specifically involves: S41, the total over-threshold region obtained by reselection The maximum wind speed, population, and GDP data within the region are unified to the same spatial grid to form a data field with consistent resolution. At this point, the total population exposure and total GDP exposure of the region are denoted as follows: and The calculation formula is as follows: ； （11） in, and These represent the population exposure and GDP exposure of a single grid cell within the threshold area, respectively. S42, based on unified grid data, extracts the maximum wind speed of each unit within the selected threshold area. This data is then correlated with the corresponding population exposure and GDP exposure to form a sample unit set; the input and output of each sample unit are defined as follows: ； in, This represents the input feature vector, which includes the maximum wind speed and exposure level. The output vector is represented by the corresponding historical loss rate; and the historical loss rate consists of two parts: the population loss rate and the GDP loss rate, calculated by the following formula: ； ； in, It concerns the historical rate of population loss. It is the total over-threshold region In the data on population loss in internal disasters, the data is unified to the same spatial grid. It concerns the historical rate of GDP loss. It is the total over-threshold region Data on GDP losses in internal disaster loss are unified to the same spatial grid data; S43. The sample set is divided into a training set and a test set. A machine learning algorithm is used to train the loss rate prediction model. The model takes the maximum wind speed and exposure level as input and the historical loss rate as output, in the following form: ； in For machine learning models, For model parameters, The loss rate predicted by the machine learning model; The training objective is to minimize the prediction loss function: ； After training, a loss rate prediction model is obtained, which predicts the corresponding loss rate given exposure level and wind speed, for the quantification of vulnerability. The predicted loss rate consists of two components: population loss rate and GDP loss rate. ； in, This represents the population loss rate output by the prediction model. This represents the GDP loss rate output by the prediction model.

5. The typhoon disaster assessment method based on WRF adaptive nesting and machine learning according to claim 1, characterized in that, In step S5, the typhoon disaster risk is calculated comprehensively based on the disaster intensity obtained in step S3, the exposure level collected in step S1 and spatially matched in step S4, and the loss rate predicted in step S4. Risk is defined as the combined effect of disaster intensity, exposure level, and vulnerability (loss rate). Any grid point within the selected high-risk area (the area exceeding the threshold) is considered a risk level. The disaster is expressed as follows: ； ； In the formula, This represents the population risk value of a grid cell. The GDP risk value of each grid cell is summed over the entire study area to obtain the overall risk. ； ； in, Indicates overall population risk. This indicates the overall risk to GDP.

6. A typhoon disaster assessment system based on WRF adaptive nesting and machine learning, characterized in that, The system is used to implement the steps of the method according to any one of claims 1 to 5, and the system comprises: The historical sample acquisition module is used to acquire meteorological data, loss data, and exposure data of the target study area during typhoons. The disaster-causing wind speed threshold determination module is used to determine the disaster-causing wind speed threshold based on historical samples and according to the typhoon disaster wind speed threshold constraint conditions. The disaster intensity acquisition module is used to filter out the range exceeding the disaster threshold through refined simulation based on the disaster-causing wind speed threshold, and determine the disaster intensity based on the maximum wind speed. The vulnerability quantification module is used to generate exposure and historical loss rate using historical sample data and threshold ranges, and to predict the loss rate. The typhoon disaster risk calculation module is used to comprehensively calculate the typhoon disaster risk based on disaster intensity, exposure level, and predicted loss rate.

7. The typhoon disaster assessment system based on WRF adaptive nesting and machine learning according to claim 6, characterized in that, The loss data includes spatial information on population loss and GDP loss, and the exposure data includes spatial information on population and GDP.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images