Typhoon path prediction method and system fusing physical constraint and path mutation recognition

Abstract

The invention provides a typhoon path prediction method and system fusing physical constraint and path mutation recognition. The method comprises the steps that a multi-dimensional input feature tensor is obtained through a feature tensor generation module; performing time modeling through a multi-scale modeling module to obtain a deep feature tensor, and performing significance guidance through a significance guidance module to obtain a prediction path sequence; the bending event identification module identifies a bending angle to obtain a plurality of path abnormal scores, the adversarial disturbance analysis module introduces a plurality of preset disturbances to determine the path confidence, and the path prediction output module outputs a path prediction result. According to the technical scheme of the embodiment of the invention, refined modeling can be carried out on the typhoon path, the PINNs physical constraint module is utilized to ensure that the prediction result meets the physical conservation principle, the path mutation detection capability is enhanced through bending event recognition, uncertainty evaluation is realized through disturbance analysis, and the accuracy of the typhoon path detection is improved. And a typhoon path prediction result which is more accurate and credible and has physical consistency is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent weather prediction, and in particular to a typhoon path prediction method and system fusing physical constraints and path mutation recognition. BACKGROUND

[0002] With the intensification of global climate change, the uncertainty of typhoon intensity and path is increasing, and high-precision typhoon path prediction is a key technical problem in current meteorological services and disaster warning systems. Traditional path prediction methods mainly include statistical regression models (such as the CLIPER series), numerical weather prediction models (such as ECMWF, WRF, etc.), and intelligent prediction methods based on machine learning. Among them, although the numerical model has a physical basis, it still has problems such as large error and high computational resource consumption in the subtropical high control area and the path mutation segment; while the deep learning method, with its advantages in big data fitting and pattern recognition, has been gradually applied to path prediction, intensity estimation and other directions in recent years, showing good potential.

[0003] However, the existing typhoon path prediction methods lack physical consistency, cannot effectively identify typhoon path mutations, and the uncertainty of the prediction process is difficult to quantify, making it difficult to guarantee the reliability of typhoon prediction. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application proposes a typhoon path prediction method and system fusing physical constraints and path mutation recognition, which can balance physical consistency, mutation warning capability and uncertainty evaluation capability, and improve the reliability of typhoon path prediction.

[0005] In a first aspect, an embodiment of the present application provides a typhoon path prediction method fusing physical constraints and path mutation recognition, applied to a typhoon path prediction system, the typhoon path prediction system comprising a feature tensor generation module, a multi-scale modeling module, a saliency guiding module, a PINNs physical constraint module, a bending event recognition module, an adversarial perturbation analysis module and a path prediction output module, the PINNs physical constraint module being used to assist in supervising the training of the multi-scale modeling module and the saliency guiding module, and the method comprising: inputting typhoon information of a target typhoon into the feature tensor generation module to obtain a multi-dimensional input feature tensor; inputting the input feature tensor into the multi-scale modeling module for time modeling to obtain a deep feature tensor, and guiding the deep feature tensor through the saliency guiding module to obtain a predicted path sequence, wherein the predicted path sequence comprises predicted positions at multiple prediction times; Based on the predicted path sequence, a plurality of path anomaly scores are identified by the bending event identification module through the bending angle, a path confidence is determined by introducing a plurality of preset disturbances through the adversarial perturbation analysis module, and a path prediction result is output by the path prediction output module, wherein the path prediction result includes the path confidence of the predicted path sequence and the path anomaly score of each predicted time.

[0006] According to some embodiments of the present application, the typhoon information of the target typhoon is input into the feature tensor generation module to obtain a multi-dimensional input feature tensor, including: A plurality of typhoon variables are obtained from the typhoon information, wherein the typhoon variables include typhoon wind speed, typhoon pressure, typhoon temperature, typhoon potential height, and typhoon specific humidity; A preset collection time step, a preset resolution, and a plurality of preset atmospheric levels along the vertical direction are obtained, wherein the preset resolution is used to divide a plurality of spatial horizontal grids; The collection time step, the preset resolution, the preset atmospheric level, and the typhoon variable are input into the feature tensor generation module to obtain the input feature tensor.

[0007] According to some embodiments of the present application, the multi-scale modeling module includes a CNN network and a time modeling network, and the input feature tensor is input into the multi-scale modeling module for time modeling to obtain a deep feature tensor, including: The input feature tensor is input into the CNN network to extract a spatio-temporal feature sequence, wherein the spatio-temporal feature sequence includes spatio-temporal features corresponding to a plurality of collection time points, the spatio-temporal features include local spatial vortexes and wind field structures, and the collection time points are determined based on the collection time step; The spatio-temporal feature sequence is input into the time modeling network, and the deep feature tensor is output by the time modeling network based on global dependencies between a plurality of the spatio-temporal features.

[0008] According to some embodiments of the present application, a predicted path sequence is obtained by performing saliency guidance on the deep feature tensor through the saliency guidance module, including: A plurality of historical typhoon paths are obtained, and a saliency map is constructed after normalizing a plurality of the historical typhoon paths; The deep feature tensor and the saliency map are fused based on a preset saliency guidance factor to obtain a fused feature tensor, and the predicted path sequence is predicted based on the fused feature tensor; The calculation formula of the fused feature tensor is: wherein, is the fused feature tensor, is the saliency guidance factor, the deep feature tensor, the saliency map, characterizing point-by-point multiplication.

[0009] According to some embodiments of the present application, the PINNs physical constraint module is preset with a water vapor conservation constraint, a momentum conservation constraint and an energy conservation constraint, and the multi-scale modeling module and the saliency guiding module are obtained based on the PINNs physical constraint module for auxiliary supervision training, comprising: constructing a water vapor conservation loss function based on the water vapor conservation constraint, wherein the formula of the water vapor conservation loss function is , the water vapor conservation loss function, the water vapor budget calculated by the model at the path point, the true water vapor budget; constructing a momentum conservation loss function based on the momentum conservation constraint, wherein the formula of the momentum conservation loss function is , the momentum conservation loss function, the path point movement vector predicted by the model, the background environmental wind speed; constructing an energy conservation loss function based on the energy conservation constraint, wherein the formula of the energy conservation loss function is: , the energy conservation loss function, the total energy per unit atmospheric column predicted by the model, the actual total energy per unit atmospheric column, and satisfies , the total energy per unit atmospheric column, the constant-pressure specific heat capacity, the typhoon temperature, the gravitational acceleration, the typhoon geopotential height, the horizontal component of the typhoon wind speed in the east-west direction, the horizontal component of the typhoon wind speed in the north-south direction; weighting and summing the water vapor conservation loss function, the momentum conservation loss function and the energy conservation loss function to obtain a total loss function, and determining the total loss function as the loss function in the training process of the multi-scale modeling module and the saliency guiding module.

[0010] According to some embodiments of the present application, the bending event recognition module comprises a 3-layer fully connected network and a softmax layer, the softmax layer comprises a plurality of preset bending events, a plurality of path anomaly scores are obtained by recognizing the bending angle through the bending event recognition module, comprising: determine a plurality of consecutive predicted bending angles based on the plurality of consecutive predicted positions; when the plurality of consecutive predicted bending angles are all greater than a preset angle value, determine a path composed of corresponding predicted positions as a mutation path; input the mutation path into the softmax layer after sequentially passing through three layers of the full connection network to determine a bending event classification probability, normalize the bending event classification probability to obtain a path anomaly score, and associate the path anomaly score to each predicted position of the mutation path.

[0011] According to some embodiments of the present application, the preset disturbance includes air pressure disturbance and temperature disturbance, and the path confidence is determined by introducing a plurality of preset disturbances through the adversarial disturbance analysis module, including: determine initial meteorological field data based on the typhoon information, introduce a plurality of groups of air pressure disturbances and temperature disturbances in the initial meteorological field data, and generate a plurality of groups of disturbance samples; based on any disturbance sample, predict a plurality of disturbance paths according to a plurality of predicted positions, and determine the path confidence according to the mean and standard deviation of the plurality of disturbance paths.

[0012] According to some embodiments of the present application, the path prediction output module outputs a path prediction result, including: determine the predicted time corresponding to each path anomaly score based on the predicted position; associate the path confidence to the predicted path sequence; visually display the path prediction result, wherein the path anomaly score and the predicted position are displayed in order based on the predicted time.

[0013] In a second aspect, the embodiments of the present application provide a typhoon path prediction system fusing physical constraints and path mutation identification, including: a feature tensor generation module configured to generate a multi-dimensional input feature tensor based on typhoon information of a target typhoon; a multi-scale modeling module configured to perform time modeling based on the input feature tensor to obtain a deep feature tensor; a saliency guidance module configured to perform saliency guidance on the deep feature tensor to obtain a predicted path sequence, wherein the predicted path sequence includes predicted positions at a plurality of predicted times; a PINNs physical constraint module configured to assist in supervised training of the multi-scale modeling module and the saliency guidance module; a bending event identification module configured to identify a bending angle based on the predicted path sequence to obtain a plurality of path anomaly scores; an adversarial perturbation analysis module configured to determine path confidence by introducing a plurality of preset perturbations to the predicted path sequence; a path prediction output module configured to output a path prediction result according to the predicted path sequence, the path anomaly score and the path confidence, wherein the path prediction result comprises the path confidence of the predicted path sequence and the path anomaly score of each predicted time point.

[0014] In a third aspect, an embodiment of the present application provides a typhoon path prediction system fusing physical constraints and path mutation recognition, comprising at least one control processor and a memory connected in communication with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the typhoon path prediction method fusing physical constraints and path mutation recognition as described in the first aspect above.

[0015] The typhoon path prediction method fusing physical constraints and path mutation recognition according to the embodiment of the present application has at least the following beneficial effects: inputting typhoon information of a target typhoon into the feature tensor generation module to obtain a multi-dimensional input feature tensor; inputting the input feature tensor into the multi-scale modeling module to perform time modeling and obtain a deep feature tensor, performing saliency guidance on the deep feature tensor by the saliency guidance module to obtain a predicted path sequence, wherein the predicted path sequence comprises predicted positions of a plurality of predicted time points; based on the predicted path sequence, identifying a bending angle by the bending event recognition module to obtain a plurality of path anomaly scores, determining path confidence by the adversarial perturbation analysis module by introducing a plurality of preset perturbations, and outputting a path prediction result by the path prediction output module, wherein the path prediction result comprises the path confidence of the predicted path sequence and the path anomaly score of each predicted time point. According to the technical solution of the embodiment of the present application, the typhoon path can be finely modeled, the PINNs physical constraint module is used to ensure that the prediction result meets the physical conservation principle, the bending event recognition is used to enhance the path mutation detection capability, and the perturbation analysis is used to realize the uncertainty evaluation, so that a more accurate, reliable and physically consistent typhoon path prediction result is output. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a schematic diagram of a typhoon path prediction system provided by an embodiment of the present application; Figure 2 is a flowchart of a typhoon path prediction method fusing physical constraints and path mutation recognition provided by another embodiment of the present application; Figure 3is a functional schematic diagram of each module in the typhoon path prediction system provided by another embodiment of the present application; Figure 4 is a structural diagram of the typhoon path prediction system fusing physical constraints and path mutation identification provided by another embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary, only for explaining the present application, and cannot be understood as a limitation of the present application.

[0018] In the description of the present application, it is understood that the orientation description, such as the orientation or position relationship indicated by up, down, front, back, left, right, etc. is based on the orientation or position relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, and is not intended to indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0019] In the description of the present application, several meanings are one or more, and the meaning of multiple is two or more, greater than, less than, more than, etc. are understood as not including the number, above, below, etc. are understood as including the number. If the first, second is described, it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of indicated technical features.

[0020] In the description of the present application, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and the person skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical solution.

[0021] The embodiment of the application provides a typhoon path prediction method and system fusing physical constraints and path mutation identification, wherein the typhoon path prediction method fusing physical constraints and path mutation identification comprises the following steps: inputting typhoon information of a target typhoon into a feature tensor generation module to obtain a multidimensional input feature tensor; inputting the input feature tensor into a multiscale modeling module for time modeling to obtain a deep feature tensor, and performing significance guidance on the deep feature tensor through a significance guidance module to obtain a predicted path sequence, wherein the predicted path sequence comprises predicted positions at multiple prediction moments; based on the predicted path sequence, a bending event recognition module is used to recognize bending angles to obtain multiple path anomaly scores, a preset disturbance is introduced through an adversarial disturbance analysis module to determine path confidence, and a path prediction output module is used to output a path prediction result, wherein the path prediction result comprises the path confidence of the predicted path sequence and the path anomaly score of each prediction moment. According to the technical scheme of the embodiment of the application, the typhoon path can be finely modeled, the PINNs physical constraint module is used to ensure that the prediction result meets the physical conservation principle, the bending event recognition is used to enhance the path mutation detection capability, and the disturbance analysis is used to realize the uncertainty evaluation, so that a more accurate, reliable and physically consistent typhoon path prediction result is output.

[0022] First, the professional terms involved in the embodiment are explained as follows: A convolutional neural network (CNN) is a deep learning model specially designed for processing grid structure data (such as images and videos). Its core feature is to greatly reduce the number of parameters through local connection and weight sharing while preserving the spatial structure information of the data. It has become the core technology in the field of computer vision.

[0023] Physics-Informed Neural Networks (PINNs) is a machine learning framework that combines physical laws and deep learning. The core idea is to embed physical laws (such as partial differential equations and conservation laws) as constraints into neural network training, thereby realizing modeling and prediction of physical systems. Partial differential equations and other physical constraints are part of the neural network loss function, thereby improving the physical consistency and interpretability of the model.

[0024] BendNet is a path anomaly recognition model based on machine learning or weather rules, which is used to detect and classify mutation behaviors in typhoon paths, such as sharp turns, path rotations or double typhoon interactions, and provides support for mutation warning.

[0025] Adversarial perturbation analysis: refers to artificially adding small but structured perturbations (such as local wind field or pressure changes) in the initial field data of typhoon, to test the sensitivity of the prediction model to input uncertainty, and to evaluate its stability and robustness.

[0026] Significance guide map: a spatial guide map constructed according to the frequency of historical path, used to enhance the attention of the model to the common path area.

[0027] Momentum conservation: refers to the consistency of the model output with the change of the environmental wind field in the prediction of typhoon path. Momentum conservation can be used as a physical constraint condition to reduce the unreasonable deviation between the prediction and the real meteorological evolution.

[0028] Uncertainty quantification (UQ): refers to the description of the confidence level of the prediction results through probabilistic modeling. In the prediction of typhoon path, it can generate prediction interval and confidence probability distribution, which can be used in risk classification and disaster response strategy making.

[0029] Fujiwhara effect: refers to the interaction of two tropical cyclones about 1400 kilometers apart, which will rotate around the common center like a two-person dance, and finally may merge or weaken, proposed by Japanese meteorologist Fujiwhara Sakihachi.

[0030] First, refer to Figure 1 , Figure 1 The schematic diagram of the typhoon path prediction system provided by the embodiment of the present application, the prediction system of the embodiment includes a feature tensor generation module, a multi-scale modeling module, a significance guide module, a PINNs physical constraint module, a bending event identification module, an adversarial perturbation analysis module and a path prediction output module.

[0031] It should be noted that, as Figure 1As shown, the typhoon path prediction system of the embodiment in the application is connected in sequence from the feature tensor generation module, the multi-scale modeling module, the saliency guiding module, the PINNs physical constraint module, the bending event identification module, the adversarial perturbation analysis module, and the path prediction output module. The PINNs physical constraint module is in the middle of the overall training link, receives feature input from the multi-scale modeling module and the saliency guiding module, and uses the physical conservation equation (mass, water vapor, momentum, and energy) to construct a loss function for auxiliary supervision. Since the PINNs physical constraint module is used for training, the features output by the saliency guiding module to the bending event identification module and the adversarial perturbation analysis module are also affected by the physical constraint, so that the PINNs physical constraint module indirectly affects the bending event identification module and the adversarial perturbation analysis module, thereby outputting more accurate, reliable, and physically consistent typhoon path prediction results. The PINNs physical constraint module mainly provides supervision of the loss function during the training process, so it does not participate in data processing during actual application, and subsequent details are not repeated.

[0032] The following is based on the accompanying Figure 1 The prediction system as shown, the technical scheme of the embodiment of the application is further described.

[0033] Referring to Figure 2 , Figure 2 A flowchart of a typhoon path prediction method fusing physical constraints and path mutation identification is provided for the embodiment of the application. The typhoon path prediction method fusing physical constraints and path mutation identification includes but is not limited to steps S10, S20, and S30. The details are as follows: S10, input the typhoon information of the target typhoon into the feature tensor generation module to obtain a multi-dimensional input feature tensor.

[0034] It should be noted that the typhoon information of the embodiment can be obtained from reanalysis data (such as ERA5) or observation data of the target typhoon. The typhoon information can include wind speed, air pressure, temperature, and the like of the typhoon. The specific variables can be selected according to actual needs. It is worth noting that the typhoon information is continuous data for a period of time, so the sequence of information can be collected according to the set collection time step.

[0035] It should be noted that, as Figure 3 shown, the typhoon path prediction system of the embodiment in the application further includes a data input module and a data preprocessing module. The data input module is used to input the obtained typhoon information, and the data preprocessing module is used to preprocess the variables of the typhoon information. The data preprocessing is a technology well known to those skilled in the art, and will not be described in detail here.

[0036] It should be noted that after the feature tensor generation module obtains the typhoon information, a multi-dimensional input feature tensor is constructed based on multiple variables of the typhoon information, thereby providing a time-space multi-dimensional feature basis for the prediction of the typhoon path.

[0037] S20, input the input feature tensor into the multi-scale modeling module for time modeling to obtain a deep feature tensor, and perform saliency guidance on the deep feature tensor through the saliency guidance module to obtain a prediction path sequence, wherein the prediction path sequence includes predicted positions at multiple prediction times.

[0038] It should be noted that in the present embodiment, the multi-scale modeling module first extracts time-space features from the input feature tensor and performs time modeling. The typhoon information is collected based on the collection time step, and therefore the time modeling can capture the global dependence relationship across the time step. Finally, a deep feature representation corresponding to the time step of the multi-scale modeling module is output. In the present embodiment, the deep feature representation is taken as a deep feature tensor. The multi-scale modeling module can preliminarily model the typhoon path, and therefore the deep feature tensor is the initially predicted typhoon path.

[0039] It should be noted that the typhoon path has a significant statistical distribution feature, and therefore the present embodiment performs saliency guidance on the deep feature tensor through the saliency guidance module, so that the saliency features in history are fused with the deep feature tensor. The fused features are taken as the prediction path sequence obtained this time, and the prediction path sequence includes predicted positions at multiple prediction times.

[0040] It should be noted that the multi-scale modeling module and the saliency guidance module are obtained through supervised training according to the PINNs physical constraint module, which can help to avoid non-physical jumps or unreasonable deviations from the path, and improve the physical consistency of the prediction.

[0041] S30, based on the prediction path sequence, a plurality of path anomaly scores are identified by the bending event identification module through the bending angle, a path confidence is determined by introducing a plurality of preset disturbances through the adversarial perturbation analysis module, and a path prediction result is output through the path prediction output module, wherein the path prediction result includes the path confidence of the prediction path sequence and the path anomaly score at each prediction time.

[0042] It should be noted that the present embodiment adds the bending event identification module in the typhoon path prediction system. The bending angle of the typhoon path is identified through the bending event identification module. In the case where the bending angle exists, it is likely that there will be sudden turning, rotation, Fujiwhara effect and other mutation events in the path. The bending event identification can be realized through the bending event classifier BendNet, or other classifiers can be used. The bending angle can be identified based on the preset path sequence, which can effectively capture the sudden turning, rotation and other abnormal changes in the typhoon path.

[0043] It should be noted that the embodiment further adds an anti-disturbance analysis module. When the predicted path sequence has been obtained, the preset disturbance is introduced through the anti-disturbance analysis module, the meteorological field data of the typhoon is changed through the preset disturbance, and a disturbed path different from the original path is obtained. If the difference between the disturbed path and the original path is large, the uncertainty of the original path is high, and the path confidence is poor. If the difference between the disturbed path and the original path is small, the uncertainty of the original path is low, and the path confidence is high. The uncertainty evaluation mechanism is formed by applying disturbance to the initial field and guiding the path distribution. The output confidence interval provides quantitative support for the path confidence, which is convenient for the business system to process the forecast information in stages.

[0044] It should be noted that the embodiment integrates the predicted path sequence, the path anomaly score and the path confidence through the path prediction output module, so that the final output includes the predicted path sequence, the path anomaly score at each prediction time and the path confidence. The forecast map can be output in a visual form, so that the typhoon path prediction result can be intuitively displayed. In the case of improving the physical consistency, the identified bending event and the path confidence are further embodied in the path prediction result, and the reliability of the typhoon prediction is improved.

[0045] In addition, in an embodiment, step S10 specifically includes but is not limited to the following steps: S11, obtaining a plurality of typhoon variables from the typhoon information, wherein the typhoon variables include typhoon wind speed, typhoon pressure, typhoon temperature, typhoon potential height and typhoon specific humidity; S12, obtaining a preset collection time step, a preset resolution and a plurality of preset atmospheric levels along the vertical direction, wherein the preset resolution is used to divide a plurality of spatial horizontal grids; S13, inputting the collection time step, the preset resolution, the preset atmospheric level and the typhoon variable into the feature tensor generation module to obtain an input feature tensor.

[0046] It should be noted that the typhoon information of the embodiment can obtain multi-source typhoon variables from reanalysis data (such as ERA5) or observation data. The typhoon variables of the embodiment include typhoon wind speed, typhoon pressure p (hPa), typhoon temperature T (K), typhoon potential height Z (gpm) and typhoon specific humidity q, wherein the typhoon wind speed includes a horizontal velocity component along the east-west direction and a horizontal velocity component along the south-north direction .

[0047] It should be noted that the embodiment presets the collection time step T, for example, sets T=12, and then it can be determined that the collection is performed once per hour within the past 12 hours. The preset atmospheric level is a plurality of typical atmospheric levels in the vertical direction, for example, 5 typical atmospheric levels 850hPa, 700hPa, 500hPa, 300hPa and 200hPa are selected. The preset resolution is the resolution in the horizontal direction, for example, sets HxW=64x64, H is the grid length, and W is the grid width, and then it can be spatially divided into 64x64 horizontal grids, and the resolution is 1 degree.

[0048] It should be noted that, as shown in Figure 3 After the above parameters are determined, the input feature tensor is input to the feature tensor generation module, the shape of the input feature tensor is determined by the above parameters, for example, the input feature tensor is (T=12, L=5, H=64, W=64, F=6), wherein T is the collection time step, F is the number of typhoon variables, L is the number of preset atmospheric levels, H is the grid length, and W is the grid width. The tensor serves as the input of the subsequent multi-scale modeling module and provides a spatiotemporal multi-dimensional feature basis for the prediction of the typhoon path.

[0049] In addition, in an embodiment, in step S20, the multi-scale modeling module includes a CNN network and a time modeling network, and the input feature tensor is input to the multi-scale modeling module to obtain a deep feature tensor through time modeling, which specifically includes but is not limited to the following steps: S211, input the input feature tensor to the CNN network to extract a spatiotemporal feature sequence, wherein the spatiotemporal feature sequence includes a plurality of spatiotemporal features corresponding to a plurality of collection time points, the spatiotemporal features include local spatial vortexes and wind field structures, and the collection time points are determined based on the collection time step; S212, input the spatiotemporal feature sequence to the time modeling network, and output a deep feature tensor based on the global dependency between the plurality of spatiotemporal features through the time modeling network.

[0050] It should be noted that, as shown in Figure 3 The multi-scale modeling module includes a CNN network and a time modeling network, the CNN network can use a 3x3x3 convolution kernel to extract local spatial vortexes and wind field structures as a spatiotemporal feature sequence. The time modeling network can be an LSTM (hidden layer 128 units), a Transformer (4 layers, 8 attention heads), a GRU or a BiLSTM structure. After obtaining the high-dimensional input tensor, the model first extracts local spatial vortex features and wind field structures using a CNN neural network. Then, the extracted spatiotemporal feature sequence is input to the Transformer-based time modeling network (including 4 layers, 8 attention heads per layer) to capture the global dependency across time steps. Finally, the module outputs a deep feature representation with a prediction time step length of 24: , wherein represents an embedding feature dimension. The output result will be passed to the saliency guidance module and the PINNs physical constraint module for further processing.

[0051] In addition, in an embodiment, in step S20, the saliency guidance module is used to guide the deep feature tensor to obtain the predicted path sequence, which specifically includes but is not limited to the following steps: S221, obtaining a plurality of historical typhoon paths, and constructing a saliency map after normalizing the plurality of historical typhoon paths; S222, fusing the deep feature tensor and the saliency map according to a preset saliency guidance factor to obtain a fused feature tensor, and predicting the predicted path sequence based on the fused feature tensor; wherein, the calculation formula of the fused feature tensor is: , wherein, is the fused feature tensor, is the saliency guidance factor, is the deep feature tensor, is the saliency map, represents point-by-point multiplication.

[0052] It should be noted that the typhoon path has a significant statistical distribution characteristic. The typhoon path prediction system of the present embodiment is based on the historical typhoon path data in the past 30 years to construct a saliency map M, which is normalized to [0, 1]. The saliency map is fused with the deep feature tensor output by the multi-scale modeling through a weighting factor, and the fused feature tensor is , is the saliency guidance factor (usually 0.5-1.0), and the fused feature tensor can maintain the response ability to abnormal path regions while ensuring the model's attention to common path regions.

[0053] In addition, in an embodiment, the PINNs physical constraint module is preset with a water vapor conservation constraint, a momentum conservation constraint, and an energy conservation constraint, and the multi-scale modeling module and the saliency guidance module are obtained based on the PINNs physical constraint module for auxiliary supervision training, which specifically includes but is not limited to the following steps: S01, constructing a water vapor conservation loss function based on the water vapor conservation constraint, wherein the formula of the water vapor conservation loss function is , is the water vapor conservation loss function, is the water vapor budget calculated by the model at the path point, is the true water vapor budget; S02, constructing a momentum conservation loss function based on the momentum conservation constraint, wherein the formula of the momentum conservation loss function is , a momentum conservation loss function, a model predicted path point movement vector, a background environmental wind speed; S03, constructing an energy conservation loss function based on an energy conservation constraint, wherein a formula of the energy conservation loss function is: , an energy conservation loss function, a model predicted unit atmospheric column total energy, an actual unit atmospheric column total energy, and satisfying , a unit atmospheric column total energy, a constant pressure specific heat capacity, a typhoon temperature, a gravitational acceleration, a typhoon potential height, a horizontal component of a typhoon wind speed in an east-west direction, a horizontal component of a typhoon wind speed in a north-south direction; S04, obtaining a total loss function by weighted summation based on the water vapor conservation loss function, the momentum conservation loss function, and the energy conservation loss function, and determining the total loss function as a loss function in a training process of the multi-scale modeling module and the saliency guiding module.

[0054] It should be noted that in the water vapor conservation constraint, the water vapor flux is defined as: wherein is a water vapor flux, Q is air specific humidity, is a wind speed. The divergence operation is performed thereon to obtain a water vapor budget term. The water vapor conservation loss is: wherein, is a model predicted water vapor budget at a path point; : a true water vapor budget from reanalysis data.

[0055] It should be noted that in the momentum conservation constraint, it is required that a typhoon motion trend is consistent with a mid-layer environmental wind field, that is wherein is a model predicted path point movement vector, is a background environmental wind speed (taking 500 hPa or 700 hPa).

[0056] It should be noted that in the energy conservation constraint, the unit atmospheric column total energy is composed of three parts of thermal energy, potential energy, and kinetic energy, and an expression is wherein is a constant pressure specific heat capacity ( ), is a typhoon temperature, is a gravitational acceleration, a potential height of the typhoon, a horizontal component of the typhoon wind speed in the east-west direction, a horizontal component of the typhoon wind speed in the south-north direction, and an energy conservation loss defined as: The constraint is used to avoid unreasonable energy surges or drops in the prediction process.

[0057] It should be noted that after determining the above three constraints, the total loss function of the embodiment is a weighted sum of the above three loss functions, so that the trajectory coordinate error and the physical consistency error jointly constitute the training target, and the expression of the total loss function is , wherein is the mean square error (MSE) of each path in the predicted path sequence, are different preset hyperparameter weights, respectively.

[0058] It should be noted that, as shown in Figure 1 and Figure 3 , the PINNs physical constraint module is placed in the middle in Figure 1 , acting as a bridge between the front-end feature output and the subsequent path prediction, and is reversely constrained by the total loss function in the training process, rather than being a separate forward calculation module. In the middle of the overall training link, it receives feature inputs from the multi-scale modeling module and the saliency guiding module, and uses the physical conservation equation (mass, water vapor, momentum, energy) to construct a loss function for auxiliary supervision. At the same time, its result will affect the "bend event identification", "adversarial perturbation analysis" and the final "path prediction output". It is ensured that the output data and constraints are not fragmented, but are layer by layer progressive and interact with each other, so as to output more accurate, reliable and physically consistent typhoon path prediction results.

[0059] In addition, in an embodiment, the bend event identification module includes a 3-layer fully connected network and a softmax layer, the softmax layer includes a plurality of preset bend events, and in step S30, a plurality of path anomaly scores are obtained by identifying the bend angles through the bend event identification module, which specifically includes but is not limited to the following steps: S311, determining a plurality of continuous predicted bend angles based on a plurality of continuous predicted positions; S312, when the plurality of continuous predicted bend angles are all greater than a preset angle value, determining a path composed of the corresponding predicted positions as a sudden change path; S313, inputting the sudden change path into the 3-layer fully connected network in turn and then into the softmax layer to determine a bend event classification probability, normalizing the bend event classification probability to obtain a path anomaly score, and associating the path anomaly score to each predicted position of the sudden change path.

[0060] It should be noted that the bending angle of the embodiment is determined by at least three time-continuous predicted positions, and the predicted bending angle is defined as: , is the predicted position at time t. At the same time, in the typhoon prediction process, because it is time-continuous, new predicted bending angles are continuously generated as the value of t increases. If only one predicted bending angle is greater than the preset angle value (30 degrees), it may be a sporadic case, so the embodiment needs to ensure that the predicted bending angle lasts for a period of time before the path composed of these predicted positions is determined as a sudden change in the road surface, for example and lasts longer than 2 hours, it is marked as a path mutation. The embodiment determines the duration by the number of predicted bending angles, for example, if a predicted bending angle is determined every 10 minutes, then 12 consecutive predicted bending angles greater than the preset angle value are sufficient.

[0061] It should be noted that, as shown in Figure 3 , the embodiment uses a 3-layer fully connected neural network (64-32-2 nodes) in the bending event recognition module to output a path anomaly score with softmax, and the expression is The output probability value is used as the path anomaly score and input to the prediction output module, wherein is the first fully connected layer, is the second fully connected layer, is the third fully connected layer. The softmax layer can classify different bending events, including sharp turns, path rotations, or double typhoon interactions.

[0062] In addition, in an embodiment, the preset disturbance includes pressure disturbance and temperature disturbance, and in step S30, the path confidence is determined by introducing multiple preset disturbances through the anti-disturbance analysis module, which includes but is not limited to the following steps: S321, determine the initial meteorological field data based on the typhoon information, introduce multiple pressure disturbances and temperature disturbances in the initial meteorological field data, and generate multiple disturbance samples; S322, based on any disturbance sample, predict multiple disturbance paths according to multiple predicted positions, and determine the path confidence according to the mean and standard deviation of the multiple disturbance paths.

[0063] It should be noted that in the case of having typhoon information, determining the initial meteorological field data is a technique well known to those skilled in the art, and will not be described here.

[0064] It should be noted that, to quantify the uncertainty of the prediction results, this embodiment incorporates multiple sets of pressure and temperature disturbances into the initial meteorological field data. Pressure disturbances can be at the ±1 hPa or ±0.5 K level, and temperature disturbances can be at the ±0.5 K level. After introducing the disturbances, multiple disturbance samples are determined from the obtained field data. The inference process is repeated 20 times for each set of disturbance samples to obtain multiple disturbance samples. Each disturbance sample is then independently predicted to generate multiple disturbance paths, resulting in a path set. Finally, the mean and standard deviation of the path set are calculated to form a 95% confidence interval as the path confidence level.

[0065] In another embodiment, in step S30, the path prediction output module outputs the path prediction result, which specifically includes, but is not limited to, the following steps: S331, Determine the prediction time corresponding to the anomaly score of each path based on the predicted location; S332, associates the path confidence with the predicted path sequence; S333, visualizes the path prediction results, where path anomaly scores and predicted locations are displayed in the order of prediction times.

[0066] It should be noted that the expression for the predicted path sequence obtained in this embodiment is based on... For example, Let be the latitude at time t. Let t be the longitude and time step. (24 hours). When performing bend event recognition, it is based on the predicted position. Each predicted position can be used as a matching basis to determine the corresponding path anomaly score, thereby associating the path anomaly score with each prediction time. Similarly, the path confidence can also be associated with the corresponding predicted path sequence, so that the output of the predicted path output module not only records the predicted path sequence, but also includes the path anomaly score and path confidence at each prediction time.

[0067] In addition, one embodiment of the present invention provides a typhoon path prediction system that integrates physical constraints and path change identification, comprising: The feature tensor generation module is used to generate multi-dimensional input feature tensors based on the typhoon information of the target typhoon. A multi-scale modeling module is used to perform time modeling based on the input feature tensor to obtain a deep feature tensor; A saliency guidance module is used to saliency guide the deep feature tensor to obtain a predicted path sequence; The PINNs physical constraint module is used to assist in supervised training of the multi-scale modeling module and the saliency guidance module. A bend event recognition module is used to identify bend angles based on the predicted path sequence to obtain multiple path anomaly scores. An anti-disturbance analysis module is used to introduce multiple preset disturbances into the predicted path sequence to determine the path confidence. The path prediction output module is used to output a path prediction result based on the predicted path sequence, the path anomaly score, and the path confidence score, wherein the path prediction result includes the path confidence score of the predicted path sequence and the path anomaly score at each prediction time.

[0068] It should be noted that the principles of each of the above modules can be found in the description of the above method embodiments, and will not be repeated here.

[0069] like Figure 4 As shown, Figure 4 This is a structural diagram of a typhoon path prediction system that integrates physical constraints and path abrupt change identification, provided in one embodiment of the present invention. The present invention also provides a typhoon path prediction system that integrates physical constraints and path abrupt change identification, comprising: The processor 401 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 402 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 402 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 402 and called by the processor 401 to execute the typhoon path prediction method integrating physical constraints and path change identification of the embodiments of this application. Input / output interface 403 is used to implement information input and output; The communication interface 404 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 405 transmits information between various components of the device (e.g., processor 401, memory 402, input / output interface 403, and communication interface 404); The processor 401, the memory 402, the input / output interface 403 and the communication interface 404 are communicatively connected with each other through the bus 405.

[0070] The embodiment of the present application further provides an electronic device, which comprises the typhoon path prediction system fusing physical constraints and path mutation identification.

[0071] The embodiment of the present application further provides a storage medium, which is a computer readable storage medium, and stores a computer program. The computer program is executed by a processor to implement the typhoon path prediction method fusing physical constraints and path mutation identification.

[0072] The memory is a non-transitory computer readable storage medium, and can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory can include a high-speed random access memory, and can further include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and the remote memory can be connected to the processor through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The apparatus embodiment described above is merely illustrative, and units described as separate components can or can not be physically separated, and can be implemented in one place or distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment.

[0073] As will be appreciated by one of ordinary skill in the art, all or some of the steps, systems, etc. in the above-disclosed methods can be embodied in software, firmware, hardware, and / or suitable combinations thereof. Some or all of the physical components can be implemented with software executed by a processor, such as a central processing unit, a digital signal processor, or microprocessor, or can be implemented as hardware, or as an integrated circuit, such as an application- specific integrated circuit. Such software can be distributed on computer readable media, which can comprise computer storage media (or non-transitory media), and communication media (or transitory media). As is well known to those of ordinary skill in the art, computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Further, as is well known to those of ordinary skill in the art, communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal, such as carrier waves or other transport mechanisms, and includes any information delivery media.

[0074] The above description is that of the preferred embodiments of the application. Various modifications and changes can be made thereto without departing from the spirit and scope of the application, which is to be given the broadest interpretation of the laws.

Claims

1. A typhoon track prediction method fusing physical constraints and track mutation identification, characterized in that, The application is applied to a typhoon path prediction system, the typhoon path prediction system comprises a feature tensor generation module, a multi-scale modeling module, a saliency guiding module, a PINNs physical constraint module, a bending event identification module, an adversarial perturbation analysis module and a path prediction output module, the PINNs physical constraint module is used to assist in supervising training of the multi-scale modeling module and the saliency guiding module, and the method comprises: Typhoon information of a target typhoon is input into the feature tensor generation module to obtain an input feature tensor of multiple dimensions; The input feature tensor is input into the multi-scale modeling module to perform time modeling to obtain a deep feature tensor, and the deep feature tensor is guided by the saliency guiding module to obtain a predicted path sequence, wherein the predicted path sequence comprises predicted positions at multiple prediction times; Based on the predicted path sequence, a plurality of path anomaly scores are obtained by identifying bending angles through the bending event identification module, a path confidence is determined by introducing a plurality of preset perturbations through the adversarial perturbation analysis module, and a path prediction result is output through the path prediction output module, wherein the path prediction result comprises the path confidence of the predicted path sequence and the path anomaly score of each prediction time.

2. The typhoon track prediction method of claim 1, wherein, Typhoon information of a target typhoon is input into the feature tensor generation module to obtain an input feature tensor of multiple dimensions, comprising: A plurality of typhoon variables are obtained from the typhoon information, wherein the typhoon variables comprise typhoon wind speed, typhoon pressure, typhoon temperature, typhoon potential height and typhoon specific humidity; A preset acquisition time step, a preset resolution and a plurality of preset atmospheric levels along a vertical direction are obtained, wherein the preset resolution is used to divide a plurality of spatial horizontal grids; The acquisition time step, the preset resolution, the preset atmospheric level and the typhoon variable are input into the feature tensor generation module to obtain the input feature tensor.

3. The typhoon track prediction method of claim 2, wherein, The multi-scale modeling module comprises a CNN network and a time modeling network, and the input feature tensor is input into the multi-scale modeling module to perform time modeling to obtain a deep feature tensor, comprising: The input feature tensor is input into the CNN network to extract a spatiotemporal feature sequence, wherein the spatiotemporal feature sequence comprises spatiotemporal features corresponding to a plurality of acquisition times, the spatiotemporal features comprise local spatial vortexes and wind field structures, and the acquisition times are determined based on the acquisition time step; The spatiotemporal feature sequence is input into the time modeling network, and the deep feature tensor is output by the time modeling network based on global dependencies between a plurality of the spatiotemporal features.

4. The typhoon track prediction method of claim 3, wherein, The deep feature tensor is guided by the saliency guiding module to obtain a predicted path sequence, comprising: A plurality of historical typhoon paths are obtained, and a saliency map is constructed after the plurality of historical typhoon paths are normalized; The deep feature tensor and the saliency map are fused according to a preset saliency guiding factor to obtain a fusion feature tensor, and the predicted path sequence is predicted based on the fusion feature tensor; The calculation formula of the fusion feature tensor is: wherein, is the fusion feature tensor, is the saliency guide factor, is the deep feature tensor, is the saliency map, represents point-by-point multiplication.

5. The typhoon track prediction method of claim 1, wherein, The PINNs physical constraint module is preconfigured with water vapor conservation constraints, momentum conservation constraints, and energy conservation constraints, and the multi-scale modeling module and the saliency guiding module are obtained based on the PINNs physical constraint module for auxiliary supervision training, comprising: constructing a water vapor conservation loss function based on the water vapor conservation constraint, wherein a formula of the water vapor conservation loss function is , is the water vapor conservation loss function, is water vapor budget calculated by the model at the path point, is a true water vapor budget; construct a momentum conservation loss function based on the momentum conservation constraint, wherein a formula of the momentum conservation loss function is , is a momentum conservation loss function, is a model-predicted path point movement vector, is a background environmental wind speed; constructing an energy conservation loss function based on the energy conservation constraint, wherein a formula of the energy conservation loss function is: , is an energy conservation loss function, is a model-predicted unit-atmosphere-column total energy, is an actual unit-atmosphere-column total energy, and satisfies , is a unit-atmosphere-column total energy, is a constant-pressure specific heat capacity, is a typhoon temperature, is a gravitational acceleration, is a typhoon geopotential height, is a horizontal component of a typhoon wind speed in an east-west direction, is a horizontal component of a typhoon wind speed in a south-north direction; The total loss function is obtained by weighted sum of the water vapor conservation loss function, the momentum conservation loss function, and the energy conservation loss function, and the total loss function is determined as the loss function in the training process of the multi-scale modeling module and the saliency guiding module.

6. The typhoon track prediction method of claim 1, wherein, The bending event recognition module includes a 3-layer fully connected network and a softmax layer, the softmax layer includes a plurality of preset bending events, and a plurality of path anomaly scores are obtained by recognizing bending angles through the bending event recognition module, comprising: Based on a plurality of continuous prediction positions, a plurality of continuous prediction bending angles are determined; When a plurality of continuous prediction bending angles are greater than a preset angle value, the path composed of the corresponding prediction positions is determined as a mutation path; After the mutation path passes through the 3-layer fully connected network in turn, the bending event classification probability is determined by inputting the mutation path into the softmax layer, the path anomaly score is obtained by normalizing the bending event classification probability, and the path anomaly score is associated with each prediction position of the mutation path.

7. The typhoon track prediction method of claim 6, wherein, The preset perturbation includes air pressure perturbation and temperature perturbation, and the path confidence is determined by introducing a plurality of preset perturbations through the adversarial perturbation analysis module, comprising: Based on the typhoon information, initial meteorological field data are determined, a plurality of groups of air pressure perturbations and temperature perturbations are introduced into the initial meteorological field data, and a plurality of groups of perturbation samples are generated; Based on any perturbation sample, a plurality of perturbation paths are predicted according to a plurality of prediction positions, and the path confidence is determined according to the mean and standard deviation of a plurality of perturbation paths.

8. The typhoon track prediction method of claim 7, wherein, The path prediction output module outputs a path prediction result, comprising: Based on the prediction position, the prediction time corresponding to each path anomaly score is determined; The path confidence is associated with the prediction path sequence; The path prediction result is visually displayed, wherein the path anomaly score and the prediction position are displayed in order based on the prediction time. 9.A typhoon track prediction system fusing physical constraints and track mutation identification, characterized in that, Comprising: A feature tensor generation module for generating a multi-dimensional input feature tensor according to the typhoon information of a target typhoon; A multi-scale modeling module for time modeling based on the input feature tensor to obtain a deep feature tensor; A saliency guiding module for saliency guiding the deep feature tensor to obtain a prediction path sequence, wherein the prediction path sequence includes prediction positions at a plurality of prediction times; A PINNs physical constraint module for auxiliary supervision training of the multi-scale modeling module and the saliency guiding module; A bending event recognition module for recognizing bending angles based on the prediction path sequence to obtain a plurality of path anomaly scores; An adversarial perturbation analysis module for introducing a plurality of preset perturbations in the prediction path sequence to determine a path confidence. A path prediction output module is configured to output a path prediction result according to the predicted path sequence, the path anomaly score and the path confidence, wherein the path prediction result comprises the path confidence of the predicted path sequence and the path anomaly score of each predicted time point. 10.A typhoon track prediction system fusing physical constraints and track mutation identification, characterized in that, The method comprises the following steps: acquiring a plurality of historical typhoon path data; determining a plurality of historical typhoon path sequences according to the historical typhoon path data; determining a plurality of historical typhoon path anomaly scores according to the historical typhoon path data; determining a plurality of historical typhoon path confidences according to the historical typhoon path data; and determining a plurality of historical typhoon path sequences, a plurality of historical typhoon path anomaly scores and a plurality of historical typhoon path confidences according to the historical typhoon path data. The method comprises the following steps: acquiring a plurality of historical typhoon path data; determining a plurality of historical typhoon path sequences according to the historical typhoon path data; determining a plurality of historical typhoon path anomaly scores according to the historical typhoon path data; determining a plurality of historical typhoon path confidences according to the historical typhoon path data; and determining a plurality of historical typhoon path sequences, a plurality of historical typhoon path anomaly scores and a plurality of historical typhoon path confidences according to the historical typhoon path data.

Images