A typhoon path prediction method and system based on vorticity state field

Abstract

The application discloses a typhoon path prediction method and system based on a vorticity state field, and belongs to the technical field of typhoon path prediction. The method is as follows: combining real-time typhoon path information and acquired meteorological information based on a preset time axis, obtaining a wind field tensor related to a typhoon cyclone and constructing an initial vorticity state field; calculating a spatial variation rate of the initial vorticity state field in a preset spatial direction to analyze the influence of air movement on the vorticity variation and obtaining a vorticity local variation rate; optimizing the initial vorticity state field through the vorticity local variation rate to obtain a final vorticity state field, searching for a maximum value point of relative vorticity in the final vorticity state field and obtaining a vortex center; and predicting a future moving path of a typhoon to be measured according to the position coordinates of the vortex center. Therefore, by implementing the application, the problem that the prior art fails to accurately perceive typhoon path mutation and leads to a large error in path prediction results can be solved.

Description

Technical Field

[0001] This application belongs to the field of typhoon path prediction technology, specifically involving a typhoon path prediction method and system based on vorticity state field. Background Technology

[0002] Eddyness is a physical quantity that describes the intensity of fluid rotation. The eddy state field of a typhoon describes the distribution characteristics of atmospheric rotation within and around the typhoon, indicating whether the typhoon will change direction during its movement. Therefore, the eddy state field is not only a direct reflection of the typhoon's rotational characteristics, but also an important physical quantity for diagnosing its intensity changes and path movement.

[0003] Numerical models are the core tool for typhoon track forecasting, primarily determining typhoon tracks through large-scale circulation fields. Therefore, they can accurately simulate large-scale circulation fields to make reasonable predictions about typhoon tracks. However, existing numerical models still have some shortcomings. For example, typhoon track forecasting requires significant computational resources and time, is extremely sensitive to the quality of initial wind fields and wind data, and can lead to large errors in predictions if the typhoon track undergoes a sudden change. Summary of the Invention

[0004] This application proposes a typhoon track prediction method and system based on vorticity state field, which can solve the problem that the existing technology fails to accurately detect sudden changes in typhoon track, resulting in large errors in track prediction results.

[0005] The first aspect of this application provides a typhoon track prediction method based on vorticity state field, the method comprising: For each cyclone of the typhoon under test, the real-time typhoon path information and the acquired meteorological information are combined based on a preset time axis to obtain the wind field tensor; wherein, the wind field tensor includes several cyclone tensors; The nonlinear spatiotemporal characteristics of the wind field are extracted from the wind field tensor to construct the initial vorticity state field; Calculate the spatial rate of change of the initial vorticity state field in a preset spatial direction, and analyze the influence of air movement on vorticity change through the spatial rate of change to obtain the local vorticity rate of change; The initial vorticity state field is optimized by the local vorticity variation rate to obtain the final vorticity state field. The maximum value of relative vorticity is searched in the final vorticity state field to obtain the vortex center. Based on the coordinates of the vortex center, the future movement path of the typhoon under test can be predicted.

[0006] The above scheme acquires the latitude, longitude, and formation time of each cyclone within the typhoon under test at the current moment. It combines cyclone and meteorological information on a preset time axis to obtain a wind field tensor characterizing the features of each cyclone, providing richer feature content for subsequent typhoon path prediction. Then, it extracts the nonlinear spatiotemporal features of the wind field from the wind field tensor to reconstruct the vorticity state field, obtaining an initial vorticity state field constructed from the optimized wind field tensor. By calculating the spatial rate of change of the initial vorticity state field in multiple spatial directions, it obtains the variation of vorticity with latitude, longitude, and vertical direction, thus achieving comprehensive prediction of vorticity changes and providing accurate data support for analyzing typhoon movement. The obtained local vorticity rate of change can be used to calculate the future evolution of the vorticity state field of the typhoon under test. By optimizing the initial vorticity state field using the local vorticity rate of change, it accurately predicts the evolution speed of the vorticity state field in future periods, thereby predicting the location of the typhoon center in future periods, achieving accurate future typhoon movement paths even with abrupt changes in the typhoon path.

[0007] In one possible implementation of the first aspect, for each cyclone of the typhoon under test, the real-time typhoon path information and the acquired meteorological information are combined based on a preset time axis to obtain the wind field tensor, specifically: Cyclone information for each cyclone is extracted from the real-time typhoon path information. The cyclone information includes the latitude and longitude coordinates and formation time of the cyclone. The cyclone information is collected based on the time axis. The air pressure and relative vorticity of the nearshore surface layer within the time axis are extracted from the meteorological information to obtain meteorological element variables. A square region is formed with the latitude and longitude coordinates of the cyclone as the center. Information on meteorological element variables is extracted through the square region, and the wind field tensor is constructed based on the extracted multi-dimensional information.

[0008] The above scheme integrates the real-time typhoon track information and meteorological information obtained on the time axis, and combines the most commonly used and representative indicators for analyzing and predicting typhoon movement tracks to obtain a multi-dimensional wind field tensor. This clearly reflects the operation of the typhoon under test over time, providing richer feature content for subsequent typhoon track prediction.

[0009] In one possible implementation of the first aspect, the first dimension of the wind field tensor represents the meteorological element variable, the second dimension represents the step size of the time axis, and the third and fourth dimensions both represent the number of grid points in the square region.

[0010] In one possible implementation of the first aspect, the nonlinear spatiotemporal characteristics of the wind field are extracted from the wind field tensor to construct an initial vorticity state field, specifically as follows: The wind field tensor is input into a preset feature reconstruction model, and multi-layer convolution and pooling are performed on the wind field tensor to obtain the first tensor; After upsampling and convolving the first tensor, the nonlinear spatiotemporal features are obtained; In the feature reshaping layer of the feature reconstruction model, the nonlinear spatiotemporal features are decomposed into dimensions to construct an initial vortex state field; wherein the initial vortex state field has the same dimension as the wind field tensor.

[0011] The above scheme performs multiple rounds of convolution and pooling in the feature reconstruction model to extract the feature data that best characterizes the vortex state field from the wind field tensor, and improves data quality through upsampling to construct an initial vortex state field. Compared with the initial data, the initial vortex state field has a highly nonlinear spatial structure and can provide clearer and more accurate data.

[0012] In one possible implementation of the first aspect, the spatial rate of change of the initial vorticity state field in a preset spatial direction is calculated, and the influence of air movement on vorticity variation is analyzed through the spatial rate of change to obtain the local vorticity rate of change, specifically: The initial vorticity state field is processed by the central difference method, and the first spatial rate of change of the initial vorticity state field in the longitude direction, the second spatial rate of change in the latitude direction, and the third spatial rate of change in the vertical direction are calculated respectively. Based on the first spatial rate of change and the second spatial rate of change, the influence of horizontal air movement on the vorticity time rate of change is analyzed to obtain the horizontal advection term of vorticity. Based on the third spatial rate of change, the influence of vertical air movement on the vorticity time rate of change is analyzed to obtain the vorticity vertical transport term. By integrating the horizontal advection term of vorticity, the vertical transport term of vorticity, and the preset planetary vorticity advection term, the local vorticity variation rate is obtained; wherein, the planetary vorticity advection term reflects the relative vorticity change caused by the change of the absolute vorticity of the air parcel due to the variation of the Earth's rotation with latitude.

[0013] The above scheme calculates the spatial rate of change of the initial vorticity state field from multiple directions, thereby obtaining the variation of vorticity with latitude, longitude, and vertical direction. The calculated horizontal advection term and vertical transport term of vorticity characterize the vorticity changes of the typhoon under test caused by horizontal and vertical air movements of different vorticities, respectively, reflecting the main dynamic mechanism of typhoon movement. The planetary vorticity advection term can reflect the polar deflection of the typhoon, realizing a comprehensive construction of the changing trend of the vorticity state field.

[0014] In one possible implementation of the first aspect, the initial vortex state field is calculated by calculating the first spatial rate of change in the longitude direction, the second spatial rate of change in the latitude direction, and the third spatial rate of change in the vertical direction, specifically as follows: Based on the square region where the initial vortex state field is located, the first spatial rate of change is obtained by taking the partial derivative of the initial vortex state field along the longitude direction according to the longitude difference between each grid point and the Earth's radius; Based on the square region, the second spatial rate of change is obtained by taking the partial derivative of the initial vortex state field along the latitudinal direction according to the latitudinal difference between each grid point and the Earth's radius; The third spatial rate of change is calculated based on the preset vertical change benchmark value and vortex correction factor.

[0015] In one possible implementation of the first aspect, the calculation process for the vertical variation reference value and the vorticity correction factor is as follows: The relative vorticity under the climatic mean is obtained by using vorticity data over historical time periods. Based on the relative vorticity, the pressure difference between adjacent pressure layers in the vertical direction and the standard pressure layer, the reference value of the rate of change of vorticity in the vertical direction is calculated to obtain the vertical variation reference value. Calculate the degree of deviation of the vorticity under the current state relative to the relative vorticity, and obtain the vorticity correction factor based on the degree of deviation and the preset climatic standard deviation of the relative vorticity.

[0016] In one possible implementation of the first aspect, the initial vorticity state field is optimized using the local vorticity variation rate to obtain the final vorticity state field, specifically as follows: The local vorticity variation rate is discretized to obtain a discrete function; wherein, the discrete function is used to characterize the instantaneous velocity of the vorticity state field evolution over time; Based on a preset time step, the slope of the discrete function is iteratively updated to obtain a set of slope estimates. The slope estimates are weighted and averaged, and the weighted average result is introduced into the initial vortex state field to obtain the final vortex state field.

[0017] The above scheme obtains a set of slope estimates by continuously updating the slope of the discrete function. These estimates reflect the different timing of predicting and correcting the future trend of the typhoon under test. Moreover, the predicted typhoon path gets closer and closer to the actual trajectory, thus greatly reducing the prediction error.

[0018] One possible implementation of the first aspect also includes: The position coordinates of the vortex center are used as the center position of the typhoon to be measured at a future time. The center position is compared with the preset reference typhoon center to obtain the prediction error. The typhoon path prediction process is iteratively updated by using prediction errors.

[0019] The above scheme provides a feedback mechanism to perform closed-loop optimization of the typhoon path prediction process, thereby obtaining high-precision path prediction results.

[0020] The second aspect of this application provides a typhoon path prediction system based on vorticity state field, the system comprising: a wind field tensor calculation module, an initial vorticity state field construction module, a rate of change calculation module, a vortex center determination module, and a path prediction module; The wind field tensor calculation module is used to combine real-time typhoon path information and acquired meteorological information based on a preset time axis for each cyclone of the typhoon under test to obtain the wind field tensor; wherein the wind field tensor includes several cyclone tensors. The initial vorticity state field construction module is used to extract the nonlinear spatiotemporal characteristics of the wind field from the wind field tensor and construct the initial vorticity state field. The rate of change calculation module is used to calculate the spatial rate of change of the initial vorticity state field in a preset spatial direction. The influence of air movement on vorticity change is analyzed through the spatial rate of change to obtain the local vorticity rate of change. The vortex center determination module is used to optimize the initial vortex state field by the local vortex variation rate to obtain the final vortex state field. The maximum value point of relative vortex is searched in the final vortex state field to obtain the vortex center. The path prediction module is used to predict the future path of the typhoon under test based on the location coordinates of the vortex center. Attached Figure Description

[0021] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a schematic flowchart of a typhoon track prediction method based on vorticity state field provided in an embodiment of this application; Figure 2 This is a structural diagram of a typhoon path prediction system based on vortex state field provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0025] First Embodiment Traditional numerical weather prediction models are sensitive to initial field errors and have large errors when predicting typhoons with abrupt changes in track. However, the embodiments of this application can more accurately capture the dynamic process of typhoon development on this traditional method, thereby significantly improving the accuracy of forecasts for abnormal tracks, near-shore typhoons and landfall points, and achieving accurate prediction of typhoon movement paths.

[0026] like Figure 1 As shown, to address the problem in existing technologies that fail to accurately detect sudden changes in typhoon paths, leading to significant errors in path prediction results, the first embodiment of this application provides a detailed flowchart of a typhoon path prediction method based on vorticity state fields. This embodiment's typhoon path prediction method based on vorticity state fields includes steps S1 to S5, detailed below: Step S1: For each cyclone of the typhoon to be tested, the real-time typhoon path information and the acquired meteorological information are combined based on a preset time axis to obtain the wind field tensor.

[0027] This application embodiment selects to obtain real-time typhoon path information from information released by meteorological stations, including the latitude and longitude and formation time of each cyclone within the typhoon under test. In fact, each complete typhoon contains multiple cyclones, so multiple cyclone information can be obtained for each cyclone, with each cyclone information recorded one hour apart.

[0028] Optionally, this application embodiment uses a 12-hour timeline for information extraction and collection.

[0029] The nearshore surface layer pressure and relative vorticity within the time axis are obtained from the available meteorological information, thus yielding two meteorological variable elements.

[0030] Optionally, considering that the nearshore surface layer at 850 hPa and a height of about 1500 m is located in the core energy layer of the typhoon, it can most clearly reflect its rotational circulation and key water vapor transport, and is less affected by ground friction, so the data quality is reliable. Therefore, meteorological information on the nearshore surface layer is mainly acquired to provide the most commonly used and representative indicator data for typhoon track prediction.

[0031] A square region is formed centered on the latitude and longitude coordinates of the cyclone. Information on meteorological variables is extracted using this square region, and a wind field tensor is constructed based on the extracted multi-dimensional information.

[0032] For example, suppose the square region is 1024×1024. Air pressure and relative vorticity within a range of 1024×1024 are extracted from meteorological variable data. These extracted data are then combined to obtain a wind field tensor with dimensions (2, 12, 1024, 1024). Here, the first dimension 2 represents the two meteorological variable data: air pressure and relative vorticity; the second dimension 12 represents the time step of the time axis, which is 12 hours; and the third and fourth dimensions 1024 both represent the number of grid points in the square region.

[0033] Therefore, the wind field tensors at 12 time points are obtained through the above steps.

[0034] As an improvement to the above scheme, the first and second dimensions of the wind field tensor in the example can be merged, as shown in the following expression: ; in, X f The combined wind field tensor X This is the original wind field tensor; reshape For merging operations.

[0035] Merging dimensions is to facilitate subsequent two-dimensional convolution operations, enabling the two-dimensional convolution operations to simultaneously extract the spatial structure and temporal dynamics of the data, thereby providing richer features for typhoon path prediction.

[0036] Step S2: Extract the nonlinear spatiotemporal characteristics of the wind field from the wind field tensor and construct the initial vorticity state field.

[0037] This application provides a trained feature reconstruction model for reconstructing the features of the wind field tensor, extracting the nonlinear spatiotemporal features of the wind field, and realizing the reconstruction of the vorticity state field.

[0038] The feature reconstruction model includes convolutional layers, pooling layers, upsampling layers, and a feature reconstruction layer. The wind field tensor is input into the feature reconstruction model, and multiple convolutional and pooling layers are performed first to extract the first tensor. Then, upsampling and convolution are performed on the first tensor to obtain the nonlinear spatiotemporal features. Finally, the feature reshaping layer decomposes the nonlinear spatiotemporal features into dimensions to construct the initial vortex state field.

[0039] For example, suppose the feature reconstruction model includes 6 convolutional layers, 3 pooling layers, 2 sampling layers, and 1 feature reconstruction layer, and the dimension of the wind field tensor is . .

[0040] The obtained wind field tensor The input is fed into convolutional layer 1 for convolution, and the specific formula is as follows: ; in, For convolution operations, Representing 48 items of size The convolution kernel has a stride of 1 and padding of 1. This is a bias term. This represents the activation function, i.e., the result of the convolution. as well as Perform a linear output to obtain a tensor. Tensor The dimensions are: ; Therefore, tensor to tensor Its dimensions are from Change to .

[0041] tensor The input to pooling layer 1 is given by the following formula: ; in, Here is the pooling operation function, and the pooling kernel size is... The step size is 2. Tensor to tensor Its dimensions are from Change to .

[0042] tensor The input to convolutional layer 2 is expressed as follows: ; in, Representing 96 items of size The convolution kernel has a stride of 1 and padding of 1. This is the bias term. Tensor to tensor Its dimensions are from Change to .

[0043] tensor The input to pooling layer 2 is as follows: ; Among them, the pooling kernel size is The step size is 2. Tensor to tensor Its dimensions are from Change to .

[0044] tensor The input to convolutional layer 3 is expressed as follows: ; in, Representing 192 items of size The convolution kernel has a stride of 1 and padding of 1. This is the bias term. Tensor to tensor Its dimensions are from Change to .

[0045] tensor The input to pooling layer 3 is as follows: ; Among them, the pooling kernel size is The step size is 2. Tensor to tensor Its dimensions are from Change to In this layer, the features are highly compressed, and the final output tensor is... It is actually the first tensor.

[0046] Then tensor The input to upsampling layer 1 is expressed as follows: ; in, 192 of which are of size The transposed convolution kernel has a stride of 2 and padding of 1. This is the bias term for upsampling layer 1. The output tensor The dimensions are: ; Therefore, tensor to tensor Its dimensions are from Change to .

[0047] tensor The input to convolutional layer 4 is expressed as follows: ; in, Representing 96 items of size The convolution kernel has a stride of 1 and padding of 1. This is the bias term. Tensor to tensor Its dimensions are from Change to .

[0048] tensor The input to upsampling layer 2 is expressed as follows: ; in, 96 of which are of size The transposed convolution kernel has a stride of 2 and padding of 1. This is the bias term for upsampling layer 2. Tensor to tensor Its dimensions are from Change to .

[0049] tensor The input to convolutional layer 5 is expressed as follows: ; in, Representing 48 items of size The convolution kernel has a stride of 1 and padding of 1. This is the bias term. Tensor to tensor Its dimensions are from Change to .

[0050] tensor The input to convolutional layer 6 is expressed as follows: ; This layer does not use an activation function to ensure the linear range of the output tensor. Representing 24 sizes The convolution kernel has a stride of 1 and padding of 1. This is the bias term. Tensor to tensor Its dimensions are from Change to .

[0051] Therefore, the tensor obtained here This is actually the aforementioned nonlinear spatiotemporal characteristic.

[0052] tensor The formula input to the "feature reshaping layer" is: ; Also, tensor The first dimension, 24, is decomposed into 2 channels and 12 time steps. It can be seen that the tensor... The wind field tensor before dimension merging has the same dimensions, both being... .

[0053] Based on the output of the feature reshaping layer, an initial vorticity state field is constructed, and the initial vorticity state field has the same dimension as the wind field tensor.

[0054] The initial vortex state field is actually an integrated data carrier of all the "nonlinear spatiotemporal features" learned in the model. These features can characterize the following complex patterns and properties of the initial vortex state field itself after network optimization compared to the original vortex state field: highly nonlinear spatial structure, such as clearer vortex eyewalls, asymmetric spiral rainband shapes, and small-scale vortex disturbances.

[0055] By combining basic features step by step using nonlinear activation functions (such as ReLU), implicit time evolution information is extracted, and time series information is fused in specific channels, which can characterize dynamic features such as the trend of vortex advection and precursors of system intensity changes.

[0056] Step S3: Calculate the spatial variation rate of the initial vorticity state field in the preset spatial direction, and analyze the influence of air movement on vorticity variation through the spatial variation rate to obtain the local variation rate of vorticity.

[0057] To analyze the spatial variation rate of typhoon vorticity in latitude, longitude, and vertical directions, the central difference method was used to process the initial vorticity state field. The first spatial variation rate of the initial vorticity state field in the longitude direction, the second spatial variation rate in the latitude direction, and the third spatial variation rate in the vertical direction were calculated sequentially.

[0058] First, based on the square region where the initial vortex state field is located, the square region is regarded as a latitude and longitude network, and the spatial variation rate of the initial vortex state field along the latitude and longitude direction and the vertical direction is calculated by the grid index.

[0059] The first spatial rate of change of the initial vorticity state field in the longitude direction is calculated using the following formula: ; in, For tensor Along the longitude direction (i.e., along the coordinate axis) (Direction) Find the partial derivative. Represents the two-dimensional horizontal grid points in latitude ( Index number in the direction, This indicates a grid point adjacent to the north (or adding a direction to the grid index). This represents a grid point adjacent to it to the south. Represents the longitude of two-dimensional horizontal grid points ( Index number in the direction, This indicates a grid point adjacent to the east (or adding a direction to the grid index). This represents a grid point adjacent to the west. This represents the actual distance between two adjacent grid points in the longitude direction (unit: In this latitude and longitude network, It is usually calculated based on the longitude difference between each grid point and the Earth's radius.

[0060] The formula for calculating the second spatial rate of change of the initial vorticity state field in the latitudinal direction is as follows: ; in, Tensor Along the latitudinal direction (i.e., along the coordinate axis) (Direction) Find the partial derivative. Tensor At grid points The value at the grid point adjacent to the north side of the value. Tensor At grid points The value at the grid point adjacent to the south side of the value. This represents the actual distance between two adjacent grid points in the latitudinal direction (unit: In this latitude and longitude network, It is usually calculated based on the latitude difference between each grid point and the Earth's radius.

[0061] The third spatial rate of change in the vertical direction calculated in the embodiments of this application is actually the vertical rate of change of vorticity, which is used to reflect the vertical change amplitude of typhoon vorticity.

[0062] To calculate the third spatial rate of change, we first calculate the baseline value of the vertical change of relative vorticity in the vertical direction under the climatic mean state, using the following formula: ; in, Indicates at horizontal grid points and barosphere The baseline value of vertical climatological eddy covariance (unit: ). This indicates that based on ERA5 reanalysis data, at horizontal grid points... and barosphere The relative vorticity calculated under the 30-year climate average (obtained from vorticity data for historical periods). and These represent the vertical direction relative to the target pressure layer. The air pressure between adjacent upper and lower layers. Indicates standard atmospheric pressure layer and Pressure difference between (unit: ).

[0063] By calculating the aforementioned vertical change baseline value, a physically reasonable and calculable estimate can be provided for the subsequent calculation of the third spatial change rate in the absence of instantaneous vertical layer data.

[0064] Furthermore, in this embodiment, the calculation of the vertical variation benchmark value of relative vorticity under climatic mean conditions is to obtain a physically reliable background field to compensate for the lack of instantaneous vertical data. Under the condition of limited data resources, this maximizes the physical authenticity of the dynamic equation calculation and the accuracy of the forecast, thereby improving the accuracy of typhoon path prediction.

[0065] Then, the deviation of the vorticity in the current state from the relative vorticity is calculated, and the vorticity correction factor is obtained based on the deviation and the preset climatic standard deviation of the relative vorticity.

[0066] The formula for calculating the vorticity correction factor is as follows: ; The vorticity correction factor is a dimensionless correction factor that reflects the degree of deviation of the vorticity intensity of the current typhoon system from the average climate state. Indicates at grid points The vorticity correction factor at that point is also a dimensionless scaling factor. For tensor That is, the value of the relative vorticity of the initial vorticity state field output by the feature reconstruction layer at the current moment on the target pressure layer (e.g., 1000 hPa). Indicates at grid points At this point, the climatological relative vorticity value (single-layer value, i.e., the value of relative vorticity in ERA5 data) is... (Same grid point, same pressure layer). Indicates at grid points The climatic standard deviation of relative vorticity calculated based on multi-year ERA5 data at the corresponding pressure layer, in units of . This represents a preset scaling factor or a scaling factor that can be optimized during model training. In this embodiment, the value is 0.2 to prevent excessive correction and ensure the physical rationality of the calculation.

[0067] Then, based on the obtained vertical variation baseline value and vortex correction factor, the third spatial variation rate is calculated, and the specific calculation formula is as follows: ; in, Indicates a network point At, tensor An estimate of the rate of spatial variation (i.e., the vertical rate of vorticity) along the direction of the pressure layer (i.e., the vertical direction). This shows that for... The calculation is based on the vertical structure of the climate average. Multiply by a factor that reflects the current system strength anomaly. This allows the magnitude of vertical variation in regions with strong vorticity systems such as typhoons to be dynamically adjusted according to the system intensity, providing a more physically grounded approach than directly using climate averages or simple interpolation.

[0068] Based on the first spatial rate of change and the second spatial rate of change, the influence of horizontal air movement on the vorticity time rate of change is analyzed, and the horizontal advection term of vorticity is obtained. The calculation formula is as follows: ; in: Represents the horizontal grid points The vorticity horizontal advection term at that point, i.e., the tensor. Eddy advection term (unit: ), which represents the rate of change of vorticity over time caused by horizontal winds transporting air of different vorticities to the local area. and , respectively, represent the horizontal wind field components corresponding to the spatiotemporal field of the vortex state field, with units of . .

[0069] Based on the aforementioned third spatial rate of change, the influence of vertical air movement on the vorticity time rate of change is analyzed, yielding the vorticity vertical transport term, calculated using the following formula: ; in, Represents the horizontal grid points The vertical transport term of vorticity at that point, i.e., the tensor Vertical transport of vorticity (unit: Vortex represents the rate of change of vorticity over time caused by vertical motion (ascent or descent) transporting air with different vorticities from other pressure layers to this layer. Vertical velocity of the air parcel (unit: ).

[0070] In addition, this application also introduces a planetary vorticity advection term to reflect the relative vorticity change caused by the conservation of absolute vorticity of air parcels due to the variation of the Earth's rotation effect with latitude.

[0071] The expression for the planetary vortex advection term is: ; in, The Rossby parameter characterizes the rate of change of the Coriolis parameter with latitude. In the embodiments of this application, Take a fixed constant . This represents the latitude index as On the grid rows Effect term, unit: .

[0072] By integrating the horizontal advection term of vorticity, the vertical transport term of vorticity, and the planetary advection term of vorticity, the local rate of vorticity variation is obtained, and the calculation formula is as follows: ; The above formula is also known as the "atmospheric vorticity tendency equation," which is the "physical constraint" followed by the embodiments of this application. Indicates at horizontal grid points The local rate of change of vorticity at a given location, also known as vorticity tendency, is measured in units of . .

[0073] The horizontal advection term of vorticity physically represents the change in vorticity caused by the wind field transporting air of different vorticities to the local area, which is the main driving mechanism for typhoon movement. The vertical transport term of vorticity physically represents the change in vorticity caused by vertical motion transporting air of different vorticities from other layers to the local layer, which is crucial for changes in typhoon intensity and the maintenance of the warm core structure.

[0074] Step S4: Optimize the initial vorticity state field by the local vorticity variation rate to obtain the final vorticity state field. Search for the maximum value point of relative vorticity in the final vorticity state field to obtain the vortex center.

[0075] The obtained local vorticity variation rate can be regarded as a function of time and vorticity state field, specifically: ; This function defines the instantaneous velocity of the vorticity state field as it evolves over time. By discretizing this function, the "vorticity solution problem" is transformed into an "initial value problem of ordinary differential equations," that is, for the vorticity state field... Given the current vorticity value Then the future moment vorticity state field ,in This is the forecast time step.

[0076] This application employs a fourth-order Runge-Kutta method, which solves the problem by iterating over the slope of the discretized function based on a preset time step. .

[0077] The initial slope of the function is calculated in the first iteration. The calculation formula is: ; in, The calculation can be understood as: at the starting point At this point, the "instantaneous rate of change" of the wind field system is measured. Then assume the system throughout the next step By maintaining this constant speed throughout, the change in state at the end of the time period can be predicted.

[0078] The second slope of the function is calculated in the second iteration. The calculation formula is: ; in, This indicates the point at half the time step. , using by Updated vorticity state field The calculated slope. This represents the instantaneous rate of change estimated based on the updated vorticity state at the "midpoint" of the time step. Essentially, it is a "sampling" of the dynamic trend of the wind field system as it evolves, used to correct for errors that may arise from linear extrapolation from the starting point.

[0079] The third slope of the function is calculated in the third iteration. The calculation formula is: ; Calculated It is at the "midpoint" of the time step, based on the slope optimized in the previous round. The more accurate instantaneous rate of change is estimated. It is essentially calculated using... Alternative Subsequently, an iterative correction was performed on the "midpoint" trend, and its estimated value was higher than... It is closer to the average evolution trend of the system over the entire period.

[0080] Finally, calculate the slope at the terminal point: ; in, Indicates the point at the end of the step. ,use Updated vorticity state field The calculated slope.

[0081] The above results , , , As a set of slope estimates, the essential difference between these four values ​​lies in the timing at which they predict and correct for the future trend of the wind field system, and the fact that the state estimates they are based on increasingly approximate the actual trajectory. The effect they achieve is that by "sampling" and weighting the trends of four different virtual state points within a single time step, they can approximate the actual continuous change process with high accuracy, thereby greatly reducing the error in extrapolating the vorticity state field.

[0082] Finally, the slope estimates are weighted and averaged, and the weighted average result is introduced into the initial vortex state field to obtain the final vortex state field. The formula for calculating the final vortex state field is as follows: ; in, This indicates that by using four slope estimates... The weights are used to calculate a weighted average, resulting in the final high-precision final vortex state field.

[0083] In the final vorticity state field In the search, the point with the maximum relative vorticity is identified, which is considered the location of the strongest typhoon circulation, i.e., the vortex center. The latitude and longitude coordinates of the vortex center are recorded as the future center location of the typhoon under test.

[0084] Step S5: Based on the position coordinates of the vortex center, predict the future movement path of the typhoon to be tested.

[0085] Based on the position coordinates of the vortex center, i.e. the center position of the typhoon to be tested at a future time, the future movement path of the typhoon to be tested can be predicted.

[0086] As an improvement to the above scheme, after obtaining the prediction results, the prediction results are compared with the authoritative typhoon observation data at the corresponding time, and the prediction error is calculated.

[0087] Specifically, a reference typhoon center is obtained using authoritative typhoon observation data at the corresponding time. The distance between the predicted result and the reference typhoon center is calculated as the prediction error for this forecast. E The specific formula is as follows: Among them, prediction error The unit is Earth's average radius . The reference typhoon center, This refers to the central location.

[0088] Prediction error As a loss function Calculate the loss function All trainable parameters in the typhoon path prediction model gradient Its formula is: ; Where, gradient It indicates the direction and magnitude in which model parameters should be adjusted to make typhoon center predictions more accurate.

[0089] Gradient descent algorithm ( (Optimizer) updates model parameters based on gradients: ; in, These are the model parameters before the update. The learning rate controls the step size for updating parameters. The updated model parameters are used. Therefore, this embodiment provides a closed-loop optimization loop that continues until the model training reaches a stable state. The updated model parameters are then used. The typhoon's path is re-predicted until a new predicted vorticity state field is obtained, and the center location and error are calculated again. This process is repeated cyclically.

[0090] Optionally, when the prediction error of multiple iterations... E When the convergence threshold is lower than the preset threshold, or when the error no longer decreases significantly, the model is considered to have been trained to a stable state, and the iteration stops.

[0091] Implementing the embodiments of this application has the following beneficial effects: This application's embodiments acquire the latitude, longitude, and formation time of each cyclone within the typhoon under test at the current moment. The cyclone information and meteorological information are combined on a preset time axis to obtain a wind field tensor that characterizes the features of each cyclone, providing richer feature content for subsequent typhoon path prediction. Then, the nonlinear spatiotemporal features of the wind field are extracted from the wind field tensor to reconstruct the vorticity state field, resulting in an initial vorticity state field constructed from the optimized wind field tensor. By calculating the spatial rate of change of the initial vorticity state field in multiple spatial directions, the variation of vorticity with latitude, longitude, and vertical direction is obtained, thus achieving comprehensive prediction of vorticity changes and providing accurate data support for analyzing typhoon movement. The obtained local vorticity rate of change can be used to calculate the future evolution of the vorticity state field of the typhoon under test. By optimizing the initial vorticity state field using the local vorticity rate of change, the evolution speed of the vorticity state field in future periods is accurately predicted, thereby predicting the location of the typhoon center in future periods, achieving accurate future typhoon movement paths even with abrupt changes in the typhoon path.

[0092] Second Embodiment Furthermore, in order to implement the typhoon path prediction system based on vorticity state field corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, Figure 2 A structural diagram of a typhoon track prediction system based on vortex state field is provided. For ease of explanation, only the parts relevant to this embodiment are shown. The typhoon track prediction system based on vortex state field provided in this application embodiment includes: The wind field tensor calculation module 201 is used to combine real-time typhoon path information and acquired meteorological information based on a preset time axis for each cyclone of the typhoon to be measured to obtain the wind field tensor; wherein, the initial wind field information includes several cyclone tensors.

[0093] In this embodiment of the application, cyclone information of each cyclone is extracted from the real-time typhoon path information. The cyclone information includes the latitude and longitude coordinates and formation time of the cyclone; wherein, the cyclone information is collected based on the time axis. The air pressure and relative vorticity of the nearshore surface layer within the time axis are extracted from the meteorological information to obtain meteorological element variables. A square region is formed with the latitude and longitude coordinates of the cyclone as the center. Information on meteorological element variables is extracted through the square region, and the wind field tensor is constructed based on the extracted multi-dimensional information.

[0094] The first dimension of the wind field tensor represents the meteorological element variable, the second dimension represents the step size of the time axis, and the third and fourth dimensions both represent the number of grid points in the square region.

[0095] The initial vorticity state field construction module 202 is used to extract the nonlinear spatiotemporal characteristics of the wind field from the wind field tensor and construct the initial vorticity state field.

[0096] In this embodiment of the application, the wind field tensor is input into a preset feature reconstruction model, and the wind field tensor is subjected to multi-layer convolution and pooling to obtain a first tensor; After upsampling and convolving the first tensor, the nonlinear spatiotemporal features are obtained; In the feature reshaping layer of the feature reconstruction model, the nonlinear spatiotemporal features are decomposed into dimensions to construct an initial vortex state field; wherein the initial vortex state field has the same dimension as the wind field tensor.

[0097] The rate of change calculation module 203 is used to calculate the spatial rate of change of the initial vorticity state field in a preset spatial direction, and to analyze the influence of air movement on vorticity change through the spatial rate of change to obtain the local vorticity rate of change.

[0098] In this embodiment of the application, the initial vorticity state field is processed by the central difference method, and the first spatial rate of change of the initial vorticity state field in the longitude direction, the second spatial rate of change in the latitude direction, and the third spatial rate of change in the vertical direction are calculated respectively. Based on the first spatial rate of change and the second spatial rate of change, the influence of horizontal air movement on the vorticity time rate of change is analyzed to obtain the horizontal advection term of vorticity. Based on the third spatial rate of change, the influence of vertical air movement on the vorticity time rate of change is analyzed to obtain the vorticity vertical transport term. By integrating the horizontal advection term of vorticity, the vertical transport term of vorticity, and the preset planetary vorticity advection term, the local vorticity variation rate is obtained; wherein, the planetary vorticity advection term reflects the relative vorticity change caused by the change of the absolute vorticity of the air parcel due to the variation of the Earth's rotation with latitude.

[0099] The vortex center determination module 204 is used to optimize the initial vortex state field by the local vortex variation rate to obtain the final vortex state field, and search for the maximum value point of relative vortex in the final vortex state field to obtain the vortex center.

[0100] In this embodiment of the application, the local vorticity variation rate is discretized to obtain a discrete function; wherein, the discrete function is used to characterize the instantaneous velocity of the vorticity state field evolution over time; Based on a preset time step, the slope of the discrete function is iteratively updated to obtain a set of slope estimates. The slope estimates are weighted and averaged, and the weighted average result is introduced into the initial vortex state field to obtain the final vortex state field.

[0101] The path prediction module 205 is used to predict the future path of the typhoon under test based on the position coordinates of the vortex center.

[0102] In this embodiment of the application, the position coordinates of the vortex center are used as the center position of the typhoon to be measured at a future time, and the center position is compared with the preset reference typhoon center to obtain the prediction error; The typhoon path prediction process is iteratively updated by using prediction errors.

[0103] The rate of change calculation module 203 further includes: Based on the square region where the initial vortex state field is located, the first spatial rate of change is obtained by taking the partial derivative of the initial vortex state field along the longitude direction according to the longitude difference between each grid point and the Earth's radius; Based on the square region, the second spatial rate of change is obtained by taking the partial derivative of the initial vortex state field along the latitudinal direction according to the latitudinal difference between each grid point and the Earth's radius; The third spatial rate of change is calculated based on the preset vertical change benchmark value and vortex correction factor.

[0104] The calculation process for the vertical variation benchmark value and vorticity correction factor is as follows: The relative vorticity under climatic mean is obtained using vorticity data from historical time periods; based on the relative vorticity, the pressure difference between adjacent pressure layers and the standard pressure layer in the vertical direction, the benchmark value of the rate of change of vorticity in the vertical direction is calculated to obtain the vertical variation benchmark value; the degree of deviation of the vorticity in the current state relative to the relative vorticity is calculated; and the vorticity correction factor is obtained based on the degree of deviation and the preset climatic standard deviation of the relative vorticity.

[0105] Implementing the embodiments of this application has the following beneficial effects: This application's embodiments acquire the latitude, longitude, and formation time of each cyclone within the typhoon under test at the current moment. The cyclone information and meteorological information are combined on a preset time axis to obtain a wind field tensor that characterizes the features of each cyclone, providing richer feature content for subsequent typhoon path prediction. Then, the nonlinear spatiotemporal features of the wind field are extracted from the wind field tensor to reconstruct the vorticity state field, resulting in an initial vorticity state field constructed from the optimized wind field tensor. By calculating the spatial rate of change of the initial vorticity state field in multiple spatial directions, the variation of vorticity with latitude, longitude, and vertical direction is obtained, thus achieving comprehensive prediction of vorticity changes and providing accurate data support for analyzing typhoon movement. The obtained local vorticity rate of change can be used to calculate the future evolution of the vorticity state field of the typhoon under test. By optimizing the initial vorticity state field using the local vorticity rate of change, the evolution speed of the vorticity state field in future periods is accurately predicted, thereby predicting the location of the typhoon center in future periods, achieving accurate future typhoon movement paths even with abrupt changes in the typhoon path.

[0106] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, or improvements made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A typhoon track prediction method based on vorticity state field, characterized in that, include: For each cyclone of the typhoon under test, the real-time typhoon path information and the acquired meteorological information are combined based on a preset time axis to obtain the wind field tensor; wherein, the wind field tensor includes several cyclone tensors; The nonlinear spatiotemporal characteristics of the wind field are extracted from the wind field tensor to construct the initial vorticity state field; Calculate the spatial rate of change of the initial vorticity state field in a preset spatial direction, and analyze the influence of air movement on vorticity change through the spatial rate of change to obtain the local vorticity rate of change; The initial vorticity state field is optimized by the local vorticity variation rate to obtain the final vorticity state field. The maximum value of relative vorticity is searched in the final vorticity state field to obtain the vortex center. Based on the coordinates of the vortex center, the future movement path of the typhoon under test can be predicted.

2. The typhoon path prediction method based on vorticity state field according to claim 1, characterized in that, For each cyclone of the typhoon under test, the real-time typhoon path information and the acquired meteorological information are combined based on a preset time axis to obtain the wind field tensor, specifically: Cyclone information for each cyclone is extracted from the real-time typhoon path information. The cyclone information includes the latitude and longitude coordinates and formation time of the cyclone. The cyclone information is collected based on the time axis. The air pressure and relative vorticity of the nearshore surface layer within the time axis are extracted from the meteorological information to obtain meteorological element variables. A square region is formed with the latitude and longitude coordinates of the cyclone as the center. Information on meteorological element variables is extracted through the square region, and the wind field tensor is constructed based on the extracted multi-dimensional information.

3. The typhoon path prediction method based on vorticity state field according to claim 2, characterized in that, The first dimension of the wind field tensor represents the meteorological element variable, the second dimension represents the step size of the time axis, and the third and fourth dimensions both represent the number of grid points in the square region.

4. The typhoon path prediction method based on vorticity state field according to claim 1, characterized in that, The extraction of nonlinear spatiotemporal features of the wind field from the wind field tensor to construct the initial vorticity state field specifically involves: The wind field tensor is input into a preset feature reconstruction model, and multi-layer convolution and pooling are performed on the wind field tensor to obtain the first tensor; After upsampling and convolving the first tensor, the nonlinear spatiotemporal features are obtained; In the feature reshaping layer of the feature reconstruction model, the nonlinear spatiotemporal features are decomposed into dimensions to construct an initial vortex state field; wherein the initial vortex state field has the same dimension as the wind field tensor.

5. The typhoon path prediction method based on vorticity state field according to claim 1, characterized in that, The calculation of the initial vorticity state field's spatial rate of change in a preset spatial direction, and the analysis of the influence of air movement on vorticity variation using this spatial rate of change, yields the local vorticity rate of change. Specifically: The initial vorticity state field is processed by the central difference method, and the first spatial rate of change of the initial vorticity state field in the longitude direction, the second spatial rate of change in the latitude direction, and the third spatial rate of change in the vertical direction are calculated respectively. Based on the first spatial rate of change and the second spatial rate of change, the influence of horizontal air movement on the vorticity time rate of change is analyzed to obtain the horizontal advection term of vorticity. Based on the third spatial rate of change, the influence of vertical air movement on the vorticity time rate of change is analyzed to obtain the vorticity vertical transport term. By integrating the horizontal advection term of vorticity, the vertical transport term of vorticity, and the preset planetary vorticity advection term, the local vorticity variation rate is obtained; wherein, the planetary vorticity advection term reflects the relative vorticity change caused by the change of the absolute vorticity of the air parcel due to the variation of the Earth's rotation with latitude.

6. The typhoon path prediction method based on vorticity state field according to claim 5, characterized in that, The calculation of the initial vortex state field's first spatial rate of change in the longitude direction, the second spatial rate of change in the latitude direction, and the third spatial rate of change in the vertical direction are specifically as follows: Based on the square region where the initial vortex state field is located, the first spatial rate of change is obtained by taking the partial derivative of the initial vortex state field along the longitude direction according to the longitude difference between each grid point and the Earth's radius; Based on the square region, the second spatial rate of change is obtained by taking the partial derivative of the initial vortex state field along the latitudinal direction according to the latitudinal difference between each grid point and the Earth's radius; The third spatial rate of change is calculated based on the preset vertical change benchmark value and vortex correction factor.

7. The typhoon path prediction method based on vorticity state field according to claim 6, characterized in that, The calculation process for the vertical variation reference value and vortex correction factor is as follows: The relative vorticity under the climatic mean is obtained by using vorticity data over historical time periods. Based on the relative vorticity, the pressure difference between adjacent pressure layers in the vertical direction and the standard pressure layer, the reference value of the rate of change of vorticity in the vertical direction is calculated to obtain the vertical variation reference value. Calculate the degree of deviation of the vorticity under the current state relative to the relative vorticity, and obtain the vorticity correction factor based on the degree of deviation and the preset climatic standard deviation of the relative vorticity.

8. The typhoon path prediction method based on vorticity state field according to claim 1, characterized in that, The optimization of the initial vorticity state field using the local vorticity variation rate yields the final vorticity state field, specifically as follows: The local vorticity variation rate is discretized to obtain a discrete function; wherein, the discrete function is used to characterize the instantaneous velocity of the vorticity state field evolution over time; Based on a preset time step, the slope of the discrete function is iteratively updated to obtain a set of slope estimates. The slope estimates are weighted and averaged, and the weighted average result is introduced into the initial vortex state field to obtain the final vortex state field.

9. The typhoon path prediction method based on vorticity state field according to claim 1, characterized in that, Also includes: The position coordinates of the vortex center are used as the center position of the typhoon to be measured at a future time. The center position is compared with the preset reference typhoon center to obtain the prediction error. The typhoon path prediction process is iteratively updated by using prediction errors.

10. A typhoon track prediction system based on vorticity state field, characterized in that, include: The module includes wind field tensor calculation, initial vorticity state field construction, rate of change calculation, vortex center determination, and movement path prediction. The wind field tensor calculation module is used to combine real-time typhoon path information and acquired meteorological information based on a preset time axis for each cyclone of the typhoon under test to obtain the wind field tensor; wherein the wind field tensor includes several cyclone tensors. The initial vorticity state field construction module is used to extract the nonlinear spatiotemporal characteristics of the wind field from the wind field tensor and construct the initial vorticity state field. The rate of change calculation module is used to calculate the spatial rate of change of the initial vorticity state field in a preset spatial direction. The influence of air movement on vorticity change is analyzed through the spatial rate of change to obtain the local vorticity rate of change. The vortex center determination module is used to optimize the initial vortex state field by the local vortex variation rate to obtain the final vortex state field. The maximum value point of relative vortex is searched in the final vortex state field to obtain the vortex center. The path prediction module is used to predict the future path of the typhoon under test based on the location coordinates of the vortex center.

Images