Method for evaluating influence of climate change on typhoon extreme wind climate based on AI

Abstract

The invention belongs to the technical field of artificial intelligence application, and discloses an AI-based method for evaluating the influence of climate change on typhoon extreme wind climate, and the method comprises the following steps: obtaining a historical meteorological data set and a predicted meteorological data set, and carrying out the data preprocessing; constructing and training a typhoon starting point information generation model according to the historical meteorological data set; constructing and training a typhoon path and intensity generation model according to the historical meteorological data set; driving the starting point information generation model and the typhoon path and intensity generation model by using the predicted meteorological data set, and generating and outputting a simulated typhoon sample within the age limit of the predicted meteorological data set; and according to the simulated typhoon sample data output repeatedly for multiple times, evaluating the extreme wind climate of the to-be-tested area. According to the model constructed by the invention, more accurate and more efficient long-period typhoon extreme wind climate characteristic evaluation is realized by learning and extracting implicit relationships between various meteorological elements of multiple barometric strata and typhoon movement and intensity changes.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence application technology, and specifically relates to an AI-based method for evaluating the impact of climate change on typhoon extreme wind climate. Background Art

[0002] Typhoon is a kind of disastrous weather under the mutual influence of multi-scale systems, which will cause direct disasters such as strong winds, heavy rains, storm surges, and is often accompanied by secondary disasters such as floods, mudslides, and landslides. This disastrous weather not only threatens people's lives and property safety, but also causes huge economic losses in the areas along the typhoon. Therefore, in order to avoid the destruction of engineering buildings along the way by the typhoon, the primary issue in the design of wind resistance of engineering structures is to reasonably evaluate the climate characteristics of typhoon extreme winds.

[0003] The assessment methods of extreme wind climate can be divided into probability statistics methods based on measured data and typhoon extreme wind speed estimation methods based on numerical simulation. Among them, the probability statistics method uses the extreme value distribution function to count and fit the measured data, and obtains the parameter values ​​of the distribution model and infers the extreme wind speed corresponding to different recurrence periods. However, if the occurrence of a typhoon passing through a certain place is a low-probability event, then the existing number of measured data will be difficult to support the use of probability statistics methods to obtain more reliable extreme wind speeds.

[0004] In order to make full use of limited measured data, researchers have developed a numerical estimation method based on Monte Carlo sampling simulation technology. Among them, Vickery et al. proposed a typhoon full-path simulation method. This method grids the entire target area and uses the input historical meteorological data of the target area to statistically summarize the historical typhoon information of each grid to construct an empirical regression model for each grid. It first simulates the initial position of the tropical cyclone. Then, based on its moving speed and direction, it calculates its position and intensity at the next moment. In this way, the entire life cycle of a tropical cyclone can be simulated.

[0005] However, each grid does not always provide enough historical typhoon data. When there is a lack of valid historical typhoon data in the grid, the accuracy of typhoon path simulation will be significantly affected. In addition, the linear regression model proposed by Vickery et al. requires a lot of parameter adjustments to ensure that the simulation results do not deviate significantly from historical meteorological data. In essence, this parameter adjustment method is relatively subjective, lacks universality and standardization, and is therefore difficult to further promote and apply. In addition, the model is unable to capture the implicit relationship between environmental variables and typhoon movement and intensity changes, which further limits its effectiveness in practical applications. Summary of the invention

[0006] In order to solve the above problems, the present invention provides an AI-based method for assessing the impact of climate change on typhoon extreme wind climate. The method can learn and extract the implicit relationship between multiple meteorological elements in multiple pressure layers and typhoon movement and intensity changes, so as to achieve more accurate and efficient long-period typhoon extreme wind climate characteristic assessment.

[0007] The embodiment of the present invention provides the following technical solution:

[0008] An AI-based method for assessing the impact of climate change on typhoon extreme wind climate includes the following steps:

[0009] Obtain historical meteorological data sets and forecast meteorological data sets and perform data preprocessing;

[0010] Constructing and training a typhoon starting point information generation model based on the historical meteorological data set;

[0011] Constructing and training a typhoon path and intensity generation model based on the historical meteorological data set;

[0012] Using the forecast meteorological data set to drive the starting point information generation model and the typhoon path and intensity generation model, generate and output simulated typhoon samples within the forecast meteorological data set period;

[0013] Based on the simulated typhoon sample data output repeatedly, the extreme wind climate of the test area is evaluated.

[0014] The historical meteorological data set includes: meteorological element data and tropical cyclone optimal path data recorded according to historical climate cycles; the predicted meteorological data set includes: atmospheric and ocean environmental data obtained by predicting future climate cycles according to global climate models.

[0015] The data preprocessing includes: gridding the study area based on the resolution requirements of the typhoon generation starting point and the predicted path, and performing corresponding gridding preprocessing on the meteorological data according to the gridded study area.

[0016] The typhoon starting point information generation model is constructed through DCNN, and the historical meteorological data set is input for model training.

[0017] The typhoon starting point information generation model includes two sub-models: a quantity generation model and a quantity distribution model; the quantity generation model and the quantity distribution model have similar structures, and each sub-model includes: an input layer, two feature extraction layers, two Dense layers and a Flatten layer.

[0018] The input layers of the sub-models are used to import preprocessed gridded meteorological data; the feature extraction layers of the sub-models extract the features of each grid from the input samples, and each feature extraction layer contains a Conv layer and a Dropout layer; the two Dense layers and one Flatten layer of the sub-models form a "Dense layer-Flatten layer-Dense layer" structure, converting the features extracted by the feature extraction layer into the final output; the two Dense layers are the first Dense layer and the second Dense layer of each sub-model respectively.

[0019] The first Dense layer of the sub-model performs nonlinear transformation on the features extracted by the feature extraction layer and sends the results to the Flatten layer. The Flatten layer flattens the multi-dimensional feature vector into a one-dimensional vector and sends the one-dimensional vector to the second Dense layer. The second Dense layer of the quantity generation model is an output layer using a linear activation function, which outputs the number of typhoons simulated and generated each month. The second Dense layer of the quantity distribution model is an output layer using Sigmoid as an activation function, which outputs the generation probability of the simulated typhoon at each grid position each month, forming a grid array composed of probabilities. The output of the quantity distribution model adopts the mean value of the output results of the same month within a certain period of time.

[0020] ConvLSTM is used to construct the typhoon path and intensity generation model, and the historical meteorological dataset is input for model training.

[0021] The typhoon path and intensity generation model includes a ConvLSTM1D layer, a Normalization layer, a Dropout layer, a Conv layer, a Flatten layer and a Dense layer.

[0022] The ConvLSTM1D layer extracts the spatiotemporal characteristics of multi-dimensional and multi-meteorological elements during the development of the typhoon.

[0023] The Normalization layer is used to normalize the extracted features of the ConvLSTM1D layer.

[0024] The Dropout layer is used to process the normalized features using a regularization technique.

[0025] The Conv layer further extracts the normalized and regularized features, and flattens the extracted features through the Flatten layer.

[0026] The Dense layer is an output layer using a linear activation function, and is used to output information about the next location of the typhoon;

[0027] The input of the typhoon path and intensity generation model is the longitude, latitude, central air pressure and maximum wind speed information of the typhoon starting point, as well as various meteorological elements of the meteorological data of the grid where the starting point is located.

[0028] The output of the typhoon path and intensity generation model is the longitude, latitude, central air pressure and maximum wind speed information of the typhoon at the next location.

[0029] Using longitude, latitude and time tags as indexes to obtain a variety of meteorological data of the grid where the model output point is located, combining the output information of the typhoon path and intensity generation model with the meteorological data obtained by the index as the operation input of the next round of typhoon path and intensity generation model, and iterating multiple rounds until the typhoon full path simulation is completed. In some possible implementations, the extreme wind climate of the area to be tested is evaluated based on the simulated typhoon sample data repeatedly output, including:

[0030] The extreme wind speed probability statistical method is used to process the wind speed data of simulated typhoon samples affecting the test area, and the extreme wind speeds corresponding to different return periods in the test area are determined in combination with historical typhoon data; in multiple simulations, for each return period, multiple extreme wind speeds are obtained, and the average value of these extreme wind speeds is taken as the long-period extreme wind climate assessment result for that return period.

[0031] The present invention can learn and extract the implicit relationship between multiple meteorological elements in multiple pressure layers and typhoon movement and intensity changes, and achieve more accurate and efficient assessment of long-period typhoon extreme wind climate characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0033] Figure 1 It is a schematic flow chart of a method for assessing the impact of climate change on typhoon extreme wind climate based on AI provided by an embodiment of the present invention;

[0034] Figure 2 It is a structural diagram of a DCNN model provided by an embodiment of the present invention;

[0035] Figure 3 It is a structural diagram of the ConvLSTM model provided by an embodiment of the present invention;

[0036] Figure 4-1 is a location distribution map of the simulated typhoon starting point provided by an embodiment of the present invention;

[0037] Figure 4-2 is a location distribution map of historical typhoon starting points provided by an embodiment of the present invention;

[0038] Figure 5-1 is a schematic diagram of a simulated typhoon trajectory provided by an embodiment of the present invention;

[0039] Figure 5-2 It is a schematic diagram of historical typhoon tracks provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs. Although any methods similar or equivalent to those described herein can also be used in the practice or testing of the present disclosure, only exemplary methods are now described.

[0042] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0043] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0044] See also Figure 1 , the invention embodiment provides an AI-based method for assessing the impact of climate change on typhoon extreme wind climate, which includes the following steps:

[0045] The first step is to obtain historical meteorological data sets and forecast meteorological data sets and perform data preprocessing.

[0046] Specifically, the historical meteorological data set includes: meteorological element data and tropical cyclone optimal path data recorded according to historical climate cycles. In an embodiment of the present application, the northwest Pacific Ocean (100°E-180°E, 0°N-35°N) from 1979 to 2019 is taken as the study area, and the COBE-SST data set, NCEP / NCAR reanalysis data set and ERA-interim reanalysis data set of the study area from 1979 to 2019 are obtained, and one of the data sets is used as the meteorological element data. Among them, the meteorological elements that need to be considered in the data set include the relative vorticity of the 850hPa pressure layer, SST, the relative humidity of 8 pressure layers, the atmospheric temperature of 17 pressure layers, the meridional wind speed of 17 pressure layers, and the zonal wind speed of 17 pressure layers. The embodiment of the present application simultaneously obtains the tropical cyclone optimal path data corresponding to the historical climate cycle.

[0047] For example, if the historical climate period is 1979-2019, and the northwest Pacific Ocean (i.e., the geographical area with longitude of 100°E-180°E and latitude of 0°N-35°N) is taken as the study area, the historical meteorological data set described in the embodiment of the present application includes the meteorological element data of the geographical area from 1979 to 2019, and the tropical cyclone optimal path data of the geographical area from 1979 to 2019.

[0048] Specifically, the predicted meteorological data set includes: atmospheric and ocean environmental data obtained by predicting future climate cycles based on global climate models.

[0049] In the embodiment of the present invention, the forecast meteorological data set obtained is a CMIP6 data set. The CMIP6 data set refers to the atmospheric and oceanic environmental data obtained by predicting future climate cycles (e.g., 2031 to 2090) based on four climate models: ACCESS-CM2, BCC-CSM2-MR, KIOST-ESM, and MIROC6. The above-mentioned multiple climate models are all published climate models, and the forecast data output by different climate models represent the prediction of future climate change under different scenarios, such as different emission scenarios.

[0050] Further, based on the resolution requirements of the typhoon generation starting point and the predicted path, the study area is gridded respectively, and the meteorological data is pre-processed by corresponding gridding according to the gridded study area. The various meteorological data downloaded can be pre-processed by resampling, interpolation, etc. using Python's open source computing package Xarray. The typhoon starting point data of the tropical cyclone optimal path data set is used to make a monthly typhoon generation placeholder grid (2.5°×2.5° resolution), and 1 is used on the placeholder grid to indicate that a typhoon has occurred at the grid point this month, and 0 is used to indicate that no typhoon has occurred at the grid point this month. The typhoon starting point information generation model established in this embodiment uses meteorological data with a grid resolution of 2.5°×2.5°, while the typhoon path and intensity model uses meteorological data with a grid resolution of 5°×5°.

[0051] The second step is to build and train a typhoon starting point information generation model based on the historical meteorological data set. Figure 2 In this embodiment, DCNN is specifically used to construct the typhoon starting point information generation model, and the historical meteorological data set is input for model training.

[0052] Specifically, the grid resolution of the study area of ​​the typhoon starting point information generation model described in this embodiment is 2.5°×2.5°, and the input shape of the input layer of the model is (batch_size, Channels, Height, Width). Among them, batch_size represents the number of input samples, and setting it to None means accepting batch_size of any size; Channels represents the number of input channels, which represents the number of meteorological elements (61 types) and typhoon generation placeholder grids at each grid point in the model, and is set to 62; Height represents the height of the grid, which corresponds to the number of rows in the geographical area, and is set to 33 according to the grid division; Width represents the width of the grid, which corresponds to the number of columns in the geographical area, and is set to 15 according to the grid division. Therefore, each training sample of the typhoon starting point information generation model described in this embodiment is a three-dimensional tensor composed of 62 33×15 grids stacked together, which contains the typhoon generation placeholder grids and the mean grids of each meteorological element in the study area within one month.

[0053] Furthermore, the typhoon starting point information generation model includes two sub-models: a quantity generation model and a quantity distribution model. The quantity generation model and the quantity distribution model have similar structures, and each sub-model includes: an input layer, two feature extraction layers, two Dense layers and a Flatten layer.

[0054] Specifically, the input layer of the sub-model is used to import the pre-processed gridded meteorological data; the feature extraction layer of the sub-model extracts the features of each grid from the input sample, and each feature extraction layer includes a Conv layer and a Dropout layer. In this embodiment, the first feature extraction layer uses a Conv layer with a convolution kernel of (3×3), an activation function of ReLU, and a Dropout layer with a dropout rate of 20%; the second extraction layer uses a Conv layer with a convolution kernel of (2×2), an activation function of ReLU, and a Dropout layer with a dropout rate of 20%. Among them, the Conv layer is used to extract features, which extracts feature information at different positions on the input data by sliding the convolution kernel. The Dropout layer is a regularization technique used to reduce overfitting of neural networks. It randomly discards the output of some neurons with a certain probability during the training process to prevent it from over-relying on certain features and improve the generalization ability of the model.

[0055] The starting point information generation model includes two Dense layers connected by a Flatten layer, that is, the two Dense layers and one Flatten layer of the sub-model form a "Dense layer-Flatten layer-Dense layer" structure, which can convert the features extracted by the feature extraction layer into the final output. The two Dense layers are the first Dense layer and the second Dense layer of each sub-model.

[0056] The first Dense layer of both sub-models is used to receive the output of the feature extraction layer, and uses ReLU as the activation function to further perform nonlinear transformation on the features to generate a multi-dimensional feature tensor. Then, the multi-dimensional feature tensor is flattened into a one-dimensional vector through the Flatten layer, and the one-dimensional vector is input into the second Dense layer.

[0057] The second Dense layer is the output layer, and it has different settings according to the function of the sub-model: the second Dense layer in the typhoon quantity generation model uses a linear activation function, the output shape is 1, and the output content is the number of typhoons that occurred in the month of the input data. The second Dense layer in the typhoon quantity distribution model uses the Sigmoid function as the activation function, the output shape is 33×15, and the output content is the probability of a typhoon occurring in each grid in the month of the input data. The typhoon starting point information generation model generates typhoon starting point information by combining the outputs of the two sub-models and using probability statistics methods. Please refer to Figure 4, Figure 4-1 is a position distribution diagram of a simulated typhoon starting point provided according to an embodiment of the present invention; Figure 4-2It is a location distribution map of the historical typhoon starting points provided according to an embodiment of the present invention. After inputting the predicted meteorological data, for the monthly probability grid output by the typhoon quantity distribution model, the output of the same month every 8 years is averaged to obtain the location distribution probability grid of the typhoons occurring every month from January to December, which is an average of 8 years. The value output by the typhoon quantity generation model is rounded down, and the Monte Carlo sampling method is used to extract a corresponding number of grid points from the location distribution probability grid of the typhoon occurring in the corresponding month. These grid points are the grid points for simulating the occurrence of typhoons. The latitude and longitude information of the simulated typhoon starting point is inverted using the grid point information, and then the central low pressure and maximum wind speed of the simulated typhoon starting point are assigned using the sampling method based on the historical typhoon starting point data, so as to obtain the simulated typhoon starting point information.

[0058] The third step is to construct and train a typhoon path and intensity generation model based on the historical meteorological dataset.

[0059] Specifically, see Figure 3 , Figure 3 is a structure of a ConvLSTM model. In this embodiment, Figure 3 The ConvLSTM shown constructs the typhoon path and intensity generation model, and inputs the historical meteorological data set for model training. The grid resolution of the study area of ​​the typhoon path and intensity generation model is 5°×5°. The model also specifically includes ConvLSTM1D layer, Normalization layer, Dropout layer, Conv layer, Flatten layer and Dense layer. The training sample of the model is the historical typhoon path point information, which contains typhoon intensity information and the meteorological elements of the grid point. The first input of the typhoon path and intensity generation model prediction is the simulated typhoon starting point information and the meteorological elements of the grid point obtained in the second step.

[0060] The ConvLSTM1D layer is used to extract the spatiotemporal characteristics of multi-dimensional meteorological elements during the development of a typhoon. The convolution kernel size of the ConvLSTM layer is 4, ReLU is used as the activation function, and the input shape is (batch_size, time_steps, Channels). The meanings of the parameters are as follows:

[0061] Batch_size indicates the number of samples input each time. Since the model is recursively implemented by inputting the information of the current position point to infer the information of the next position point, the model only inputs the information of one position point at a time and is set to 1.

[0062] time_steps represents the input step. According to this model, one time position is input each time and it is set to 1.

[0063] Channels represents the number of input channels. According to the model input, there are 63 elements, including the intensity information of the typhoon at that moment and the meteorological information of the grid point at the location, which is set to 63.

[0064] The Normalization layer is used to normalize the extracted features of the ConvLSTM1D layer;

[0065] The Dropout layer is used to process the normalized features using regularization techniques;

[0066] The Conv layer further extracts the normalized and regularized features, and flattens the extracted features through the Flatten layer;

[0067] The Dense layer is an output layer using a linear activation function, and is used to output information about the next location point of the typhoon.

[0068] After the ConvLSTM layer, the typhoon path and intensity generation model uses normalization and regularization techniques to process the features extracted by the ConvLSTM layer, and then inputs them into the Conv layer for further feature extraction. The features extracted by the Conv layer are flattened in the Flatten layer and input into the Dense layer. The Dense layer uses a linear activation function to output a result of shape 4, which includes the longitude, latitude, central low pressure, and maximum wind speed of the next location of the typhoon. Using the longitude and latitude information of the next location of the typhoon output by the model as the index, the typhoon intensity information is combined with the meteorological information of the corresponding grid point to obtain the complete information of the next location of the typhoon and input it into the model. This iteration is repeated until the full path simulation of the typhoon is completed. Figure 5-1 is a simulated typhoon track diagram provided according to an embodiment of the present invention; Figure 5-2 It is a historical typhoon track map provided according to an embodiment of the present invention.

[0069] Preferably, in order to enable the model to capture rare strong typhoon information, the training data of the logged strong typhoon information can be screened out to perform secondary training on the typhoon path and intensity generation model.

[0070] Preferably, in view of the extremely uneven intensity and sample distribution of tropical cyclones, undersampling processing can be performed on low-intensity typhoon tracks.

[0071] The fourth step is to use the predicted meteorological data set to drive the starting point generation model and the typhoon path and intensity generation model to generate and output simulated typhoon samples within the forecast meteorological data set period.

[0072] In this embodiment, the predicted meteorological data set is used to drive the starting point information generation model and the typhoon path and intensity generation model to generate simulated typhoon full path information for each year and month from 2031 to 2090. In order to study the extreme wind climate in the coastal areas of China in the northwest Pacific, the study area is divided into 0.5°×0.5° grids, and the areas of interest are marked as target grids. The typhoon data of the target grid points affected by the typhoon trajectory grid points in the statistical simulation of typhoon data are calculated using the extreme value type I distribution (annual extreme value method) to calculate the extreme wind speed of each target grid point with different recurrence periods.

[0073] The fifth step is to evaluate the extreme wind climate of the test area based on the simulated typhoon sample data output repeatedly.

[0074] In an embodiment of the present application, a probability statistical method of extreme wind speed is used to process the wind speed data of simulated typhoon samples that affect the area to be tested, and the extreme wind speeds corresponding to different recurrence periods of the area to be tested are determined in combination with historical typhoon data. The average of the extreme wind speeds obtained multiple times is calculated as the long-period extreme wind climate assessment result.

[0075] Exemplarily, after the fourth step is repeated 10,000 times, in multiple simulations, for each return period, multiple extreme wind speeds are obtained, and the average value of these extreme wind speeds is taken as the long-period extreme wind climate assessment result for the return period.

[0076] The AI ​​model constructed by the above-mentioned embodiments of the present invention realizes a more accurate and efficient assessment of the long-period typhoon extreme wind climate characteristics by learning and extracting the implicit relationship between multiple meteorological elements in multiple pressure layers and typhoon movement and intensity changes.

[0077] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. An AI-based method for assessing the impact of climate change on typhoon extreme wind climate, characterized in that: The following steps are involved: Obtain historical meteorological data sets and forecast meteorological data sets and perform data preprocessing; Constructing and training a typhoon starting point information generation model based on the historical meteorological data set; Constructing and training a typhoon path and intensity generation model based on the historical meteorological data set; Using the forecast meteorological data set to drive the starting point information generation model and the typhoon path and intensity generation model, generate and output simulated typhoon samples within the forecast meteorological data set period; Based on the simulated typhoon sample data output repeatedly, the extreme wind climate of the test area is evaluated.

2. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 1 is characterized in that: The historical meteorological data set includes: meteorological element data and tropical cyclone optimal path data recorded according to historical climate cycles; the predicted meteorological data set includes: atmospheric and ocean environmental data obtained by predicting future climate cycles according to global climate models.

3. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 1 is characterized in that: The data preprocessing includes: Based on the resolution requirements of the typhoon generation starting point and the predicted path, the study area is gridded respectively, and the meteorological data is preprocessed according to the gridded study area.

4. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 1 is characterized in that: The typhoon starting point information generation model is constructed through DCNN, and the historical meteorological data set is input for model training.

5. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 4 is characterized in that: The typhoon starting point information generation model includes two sub-models: a quantity generation model and a quantity distribution model; The quantity generation model and the quantity distribution model have similar structures, and each sub-model includes: an input layer, two feature extraction layers, two Dense layers and a Flatten layer.

6. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 5 is characterized in that: The input layers of the sub-models are all used to import pre-processed gridded meteorological data; The feature extraction layer of the sub-model extracts the features of each grid from the input samples, and each feature extraction layer contains a Conv layer and a Dropout layer; The two Dense layers and one Flatten layer of the sub-model form a "Dense layer-Flatten layer-Dense layer" structure, converting the features extracted by the feature extraction layer into the final output; The two Dense layers are the first Dense layer and the second Dense layer of each sub-model.

7. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 6 is characterized in that: The first Dense layer of the sub-model performs nonlinear transformation on the features extracted by the feature extraction layer and sends them to the Flatten layer. The Flatten layer flattens the multi-dimensional feature vector into a one-dimensional vector and sends the one-dimensional vector to the second Dense layer. The second Dense layer of the quantity generation model is an output layer using a linear activation function, which outputs the number of typhoons simulated and generated each month; The second Dense layer of the quantity distribution model is an output layer with Sigmoid as the activation function, which outputs the generation probability of the simulated typhoon at each grid position every month to form a grid array composed of probabilities; The output of the quantity distribution model adopts the average value of the output results of the same month within a certain period of time.

8. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 1 is characterized in that: ConvLSTM is used to construct the typhoon path and intensity generation model, and the historical meteorological dataset is input for model training.

9. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 8, characterized in that: The typhoon path and intensity generation model includes ConvLSTM1D layer, Normalization layer, Dropout layer, Conv layer, Flatten layer and Dense layer; The ConvLSTM1D layer extracts the spatiotemporal characteristics of multi-dimensional and multi-meteorological elements during the development of the typhoon; The Normalization layer is used to normalize the extracted features of the ConvLSTM1D layer; The Dropout layer is used to process the normalized features using regularization techniques; The Conv layer further extracts the normalized and regularized features, and flattens the extracted features through the Flatten layer; The Dense layer is an output layer using a linear activation function, and is used to output information about the next location of the typhoon; The input of the typhoon path and intensity generation model is the longitude, latitude, central air pressure and maximum wind speed information of the typhoon starting point, as well as various meteorological elements of the meteorological data of the grid where the starting point is located; The output of the typhoon path and intensity generation model is the longitude, latitude, central air pressure and maximum wind speed information of the typhoon at the next location; Using longitude, latitude and time tags as indexes, a variety of meteorological data of the grid where the model output point is located is obtained, and the output information of the typhoon path and intensity generation model and the meteorological data obtained by the index are combined as the operation input of the next round of typhoon path and intensity generation model. Multiple rounds of iterations are repeated until the full typhoon path simulation is completed.

10. The AI-based method for assessing the impact of climate change on typhoon extreme wind climate according to claim 1 is characterized in that: The method of evaluating the extreme wind climate of the area to be tested based on the simulated typhoon sample data outputted repeatedly for multiple times includes: The extreme wind speed probability statistical method is used to process the wind speed data of simulated typhoon samples affecting the test area, and the extreme wind speeds corresponding to different return periods in the test area are determined in combination with historical typhoon data; in multiple simulations, for each return period, multiple extreme wind speeds are obtained, and the average value of these extreme wind speeds is taken as the long-period extreme wind climate assessment result for that return period.

Images