Typhoon effect evaluation method and device, electronic equipment and readable storage medium

Abstract

The invention provides a typhoon effect evaluation method and device, electronic equipment and a readable storage medium. The method comprises the following steps: capturing unobserved time-varying confounding factors by analyzing typhoon data and geographic area covariants; learning representation of time-varying hybrid factors by using machine learning technologies such as a recurrent neural network and a graph convolutional network in combination with expanded observable data and regional covariable data of a previous time step; based on the confounding factors and observable data of the current time step, the conditional density is calculated to estimate the expected value of the human mobility index, and an average dose-reaction curve is drawn to evaluate the causal effect of the continuous weather handling level on the human mobility. Through the method, accurate prediction of human mobility influence and accurate evaluation of the causal relationship are realized, and the problems of insufficient accuracy of human mobility prediction and causal effect evaluation under extreme weather conditions are solved.

Description

Technical Field

[0001] The present application relates to the technical field of meteorological data analysis and causal inference, and specifically to a typhoon effect assessment method, device, electronic device and readable storage medium. Background Art

[0002] Under extreme weather conditions such as typhoons, researchers and government officials are committed to analyzing human mobility behavior in order to play a key role in disaster emergency management and urban operations. A variety of methods are used, including social media data to assess the impact of typhoons on specific groups such as tourists, and causal inference techniques to explore the potential impact of extreme weather events on population mobility. These techniques have been widely used in many fields such as education, medicine, and economics, among which machine learning methods, such as propensity score matching and neural networks, are used to estimate the causal effects of disasters.

[0003] Although existing technologies have made progress in analyzing the impact of typhoons on human mobility, some key issues remain. The main problem lies in unobserved confounding factors, such as regional characteristics, which may be correlated with typhoon intensity and affect human mobility, leading to biases in causal analysis. In addition, existing methods are difficult to adapt to the situation where confounding factors change over time and typhoon dynamics, which affects the accurate estimation of causal effects. Therefore, more advanced methods are needed to take into account these dynamic changes to improve the accuracy and reliability of the analysis. Summary of the invention

[0004] In view of the above problems, the present application provides a typhoon effect assessment method, device, electronic device and readable storage medium.

[0005] In order to solve the above technical problems, a technical solution adopted by the present invention is to provide a typhoon effect assessment method, the method comprising: Analyze the target typhoon data and incorporate covariates of residents in different geographical areas to capture unobserved time-varying confounding factors; Capturing and learning representations of time-varying confounders through machine learning based on extended observable data as well as regional covariate data and unobserved confounders at the previous time step; Conditional density is calculated based on the confounding factors and observable data used for analysis at the current time step, the expected value of the human mobility index is estimated, an objective function incorporating the estimates is constructed, and average dose-response curves are plotted to assess the causal effects of successive weather treatment levels on human mobility.

[0006] Among them, the target typhoon data is analyzed and combined with covariates of residents in different geographical areas to capture unobserved time-varying confounding factors, including: Analyze the target typhoon data to capture basic observable data, select and quantify variables representing unobserved confounding factors based on the basic observable data to define them as proxy variables, and convert the observable data into feature vectors; wherein the basic observable data include typhoon weather data, resident alertness data, and resident basic flow pattern data; the proxy variables include resident alertness data volume and resident basic flow pattern data; the proxy variables are as described in formula (1): (1) In formula (1), t represents the time step, n Indicates the number of selected geographic regions; Conduct time series analysis on the basic mobility pattern data of residents to capture their time dependence; The mobility tensor of different travel modes is decomposed into a basic life mode tensor by formula (3); wherein the basic life mode tensor includes a regional intensity change mode tensor, a time period mode tensor and a day mode tensor; (3) In formula (3), represents the liquidity tensor, The intensity matrix representing the basic life patterns, coefficient matrix representing the temporal pattern; A time-space embedding is formed based on the region and time information of the residents' search for typhoon keywords, and the time-space embedding is input into the causal network structure to capture unobserved confounding factors; wherein the confounding factors are as described in formula (2): (2)

[0007] Among them, according to the extended observable data and the regional covariate data and unobserved confounders of the previous time step, the representation of time-varying confounders is captured and learned through machine learning, including: Capturing the extended observable data; wherein the extended observable data includes extreme weather data , Human Mobility Index , mobile streaming network ; The mobile streaming network As described in formula (4): (4) In formula (4), R represents the set of real numbers; According to the extended observable data, causal effects are estimated by constructing a causal network structure and calculating the average treatment effect; Using a recurrent neural network to process historical covariate data through formula (8) to capture dynamic features in time series data, and using a graph convolutional network to process a mobile flow network to capture spatial dependencies, and combining the dynamic features in the time series data captured by the recurrent neural network and the spatial dependencies captured by the graph convolutional network through formula (9) to predict time-varying confounding factors; (8) In formula (8), Indicates that at time step t -1 hidden vector, capturing the time steps up to t -1 historical information; Indicates that at time step t The hidden vector of -2 is the initial state of the recurrent neural network; Indicates that at time step t -1 unobserved confounder; Indicates that at time step t -1 covariate; Indicates that at time step t -1 extreme weather data; Indicates that at time step t The potential result of the human mobility index of -1; ⊕ represents the concatenation of different vectors; (9) In formula (9), Indicates that at time step t unobserved confounders; Indicates that at time step t The mobile flow network matrix; ReLU Represents the rectified linear unit activation function, which is used to introduce nonlinearity and accelerate the learning of the model; and They respectively represent the two-layer model parameters of the graph convolutional network.

[0008] Among them, according to the extended observable data, the causal effect is estimated by constructing a causal network structure and calculating the average treatment effect, specifically including: According to the extended observable data, a causal network structure is constructed based on the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur; wherein the causal network structure is as described in formulas (5) and (6): (5) (6) In formulas (5) and (6), and Respectively represent the time step tpotential outcomes of the human mobility index for typhoon occurrence and typhoon non-occurrence; Indicates that at time step t within, no. n potential results of the human mobility index for each region during a typhoon; Indicates that at time step t within, no. n potential results of the human mobility index for each region when a typhoon does not occur; The time step is calculated by formula (7): t The potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur are averaged across all regions to obtain the average treatment effect for causal effect estimation; (7) In formula (7), Indicates that at time step t , for the region i , potential results of the human mobility index during typhoons; Indicates that at time step t , for the region i , potential results of the human mobility index when the typhoon did not occur; Indicates that at time step t ,area i unobserved confounders; Indicates that at time step t ,area i extreme weather data; E(⋅) represents the expected value, which is used to estimate the potential results under given confounding factors and weather data conditions; Represents the time step t The average treatment effect.

[0009] Among them, the conditional density is calculated based on the confounding factors and observable data for analysis at the current time step, the expected value of the human mobility index is estimated, an objective function containing the estimation is constructed, and the average dose-response curve is plotted to evaluate the causal effect of continuous weather treatment levels on human mobility, including: Based on the learned confounding factors of the current time step and observable data for analysis, conditional density is calculated using probability density estimation; wherein the observable data for analysis includes typhoon impact data and human mobility data; Based on the learned confounding factors of the current time step and the human mobility index, the expected value of the human mobility index is calculated using the machine learning model; Constructing an objective function based on the conditional density and the result estimation; wherein the objective function is as described in formula (17); (17) In formula (17), represents the objective function, n represents the total number of observed samples; Indicates i Actual human mobility index for a sample of observations; Indicates i The predicted value of the human mobility index for a sample of observations; Indicates i Extreme weather data for observation samples; Indicates i Unobserved confounders of the observed sample; Indicates that given the unobserved confounder Under the conditions of extreme weather data The probability density of α is a regularization hyperparameter used to balance the weight of prediction error and conditional density estimation loss; The average dose-response curve is plotted by using the confounding factors and extreme weather data of the current time step obtained based on learning, and the causal effect of the continuous weather treatment level on the human mobility index is obtained; wherein the confounding factors and extreme weather data of the current time step are as described in formula (18); the average dose-response curve is as described in formula (19); (18) In formula (18), Indicates that at time step t , No. i The result of whether a typhoon occurs in the observation sample; Indicates that at time step t , No. i unobserved confounders learned from the observed sample; Indicates that at time step t , No. i Extreme weather data in observation samples; (19) In formula (19), E (⋅) indicates expected value; Indicates that at time step t Human Mobility Index; and Indicates that at time step t Other covariates or network structure data; Indicates that at time step t The level of weather treatment; A specific value representing a given weather treatment level.

[0010] Among them, based on the confounding factors of the current time step obtained by learning and the observable data for analysis, the conditional density is calculated using probability density estimation, specifically including: Apply the activation function to estimate the conditional density under given confounding factors at multiple grid points through formula (10) to obtain the probability distribution; (10) In formula (10), Confounding factors The eigenvector of The result is J +1-dimensional real vector; Indicates that given the confounding factors Under the conditions of J + Conditional density at 1 grid point; , The dimension is The weight matrix of Confounding factors Dimensions; The conditional density is adjusted by ensuring that the estimated conditional density is continuous for any given value at a number of grid points from 0 to 1 and satisfies the valid density conditions that all probability values ​​are non-negative and that the sum of the entire probability distribution is 1; The specific value of the conditional density at each grid point is calculated based on formula (11) by linear interpolation to obtain the detailed probability distribution of different weather treatment levels under given confounding factors; (11) In formula (11), Indicates the corresponding weather data processing level , in j The conditional density estimate at grid points; where j ∈{0, 1, ..., J}, J The total number of grids minus one; By performing linear interpolation based on formula (12) at predefined grid points, the conditional density of any specific weather treatment level is accurately estimated to obtain a function that continuously and accurately reflects the probability distribution of the corresponding level; (12) In formula (12), Indicates that given the confounding factors Under the conditions of weather treatment level The conditional density estimate of ; and Represents the grid points and The conditional density estimate at ; and , which means less than or equal to The largest integer sum greater than or equal to The smallest integer of ; Represents the total number of grid points minus one.

[0011] Among them, based on the confounding factors of the current time step and the human mobility index obtained by learning, the expected value of the human mobility index is calculated using a machine learning model, specifically including: The human mobility index is defined under given extreme weather data and confounding factors by formula (13) to obtain the expected value of the human mobility index; (13) In formula (13), represents the expected value of the human mobility index; The continuous extreme weather data and confounding factors are used as input to train the multilayer perceptron network, and the extreme weather data and confounding factors are processed by integrating the variable coefficient prediction head through formula (14) to obtain the expected value of the human mobility index; where, (14) In formula (14), Represents the parameters of the variable coefficient prediction head; Indicates that it has parameters Neural network function; represents the confounding factors as input vector The eigenvector of By applying the activation function through formula (15) and the spline odd function modeling through formula (16) in the multilayer perceptron network, nonlinear changes and continuity modeling of the input data are performed to obtain a predicted value of the human mobility index with accuracy and flexibility; (15) In formula (15), represents the degree of the truncated basis, Indicates i The parameter vector of neurons depends on extreme weather data ; Represents the parameter vector The transpose of the input vector Perform dot product operation; (16) In formula (16), represents the linear combination of spline basis functions, represents the spline basis, Indicates i Neurons and l The coefficients between the spline basis functions; Indicates l spline basis functions.

[0012] In order to solve the above technical problems, another technical solution adopted by the present invention is to provide a typhoon effect assessment device, the device comprising: The typhoon effect analysis unit is used to analyze the target typhoon data and incorporate covariates of residents in different geographical areas to capture unobserved time-varying confounding factors; A time-varying confounding factor learning unit, used to capture and learn the representation of time-varying confounding factors through machine learning based on the extended observable data and the regional covariate data and unobserved confounding factors of the previous time step generated by the typhoon effect analysis unit; A causal effect evaluation unit is used to calculate the conditional density based on the confounders of the current time step generated by the time-varying confounder learning unit and the observable data for analysis, estimate the expected value of the human mobility index, construct an objective function containing the estimation, and draw an average dose-response curve to evaluate the causal effect of continuous weather treatment levels on human mobility.

[0013] Wherein, the typhoon effect analysis unit includes: A typhoon data capture module, used to analyze target typhoon data to capture basic observable data; wherein the basic observable data includes typhoon weather data, resident alertness data, and resident basic flow pattern data; The proxy variable definition module is used to select and quantify variables representing unobserved confounding factors based on the basic observable data captured by the typhoon data capture module and define them as proxy variables, and convert the observable data into feature vectors; wherein the proxy variables include the amount of residents' alertness data and the basic flow pattern data of residents; the proxy variables are as described in formula (1): (1) In formula (1), t represents the time step, n Indicates the number of selected geographic regions; A time series analysis module, used for performing time series analysis on the basic flow pattern data of residents captured by the typhoon data capture module to capture its time dependency; A non-negative matrix decomposition module is used to decompose the mobility tensors of different travel modes into basic life pattern tensors through formula (3); wherein the basic life pattern tensor includes a regional intensity change pattern tensor, a time period pattern tensor and a day pattern tensor; (3) In formula (3), represents the liquidity tensor, The intensity matrix representing the basic life patterns, coefficient matrix representing the temporal pattern; A time-space embedding generation module, used to form a time-space embedding according to the area and time information of residents searching for typhoon keywords captured by the typhoon data capture module; A causal network processing module is used to input the time-space embedding generated by the time-space embedding generation module into the causal network structure to capture unobserved confounding factors; wherein the confounding factors are as described in formula (2): (2).

[0014] Wherein, the time-varying confounding factor learning unit includes: An extended observable data capture module is used to capture the extended observable data; wherein the extended observable data includes extreme weather data , Human Mobility Index , mobile streaming network ; The mobile streaming network As described in formula (4): (4) In formula (4), R represents the set of real numbers; A causal network construction module, used to construct a causal network structure according to the extended observable data captured by the extended observable data capture module; A causal effect estimation module, used to calculate the average treatment effect through the causal network structure constructed by the causal network construction module, and perform causal effect estimation; A recurrent neural network processing module, used for processing historical covariate data using a recurrent neural network through formula (8) to capture dynamic features in time series data; (8) In formula (8), Indicates that at time step t -1 hidden vector, capturing the time steps up to t -1 historical information; Indicates that at time step t The hidden vector of -2 is the initial state of the recurrent neural network; Indicates that at time step t -1 unobserved confounder; Indicates that at time step t -1 covariate; Indicates that at time step t -1 extreme weather data; Indicates that at time step t The potential result of the human mobility index of -1; ⊕ represents the concatenation of different vectors; A graph convolutional network processing module, which is used to process the mobile flow network using a graph convolutional network to capture spatial dependencies; A time-varying confounding factor prediction module, used to predict time-varying confounding factors by combining the capture results of the recurrent neural network processing module and the graph convolution network processing module through formula (9); (9) In formula (9), Indicates that at time step t unobserved confounders; Indicates that at time step t The mobile flow network matrix; ReLU Represents the rectified linear unit activation function, which is used to introduce nonlinearity and accelerate the learning of the model; and They respectively represent the two-layer model parameters of the graph convolutional network.

[0015] The causal network construction module is used to construct a causal network structure based on the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur according to the extended observable data; wherein the causal network structure is as described in formulas (5) and (6): (5) (6) In formulas (5) and (6), and Respectively represent the time step t potential outcomes of the human mobility index for typhoon occurrence and typhoon non-occurrence; Indicates that at time step t within, no. n potential results of the human mobility index for each region during a typhoon; Indicates that at time step t within, no. n potential results of the human mobility index for each region when a typhoon does not occur; The causal effect estimation module is used to calculate the time step t The potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur are averaged across all regions to obtain the average treatment effect for causal effect estimation; (7) In formula (7), Indicates that at time step t , for the region i , potential results of the human mobility index during typhoons; Indicates that at time step t , for the region i , potential results of the human mobility index when the typhoon did not occur; Indicates that at time step t ,area i unobserved confounders; Indicates that at time step t ,area i extreme weather data; E(⋅) represents the expected value, which is used to estimate the potential results under given confounding factors and weather data conditions; Represents the time step t The average treatment effect.

[0016] Wherein, the causal effect assessment unit comprises: A conditional density calculation module, used to calculate the conditional density using probability density estimation based on the confounding factors of the current time step learned by the time-varying confounding factor prediction module and observable data for analysis; wherein the observable data for analysis include typhoon impact data and human mobility data; A human mobility index estimation module, configured to calculate an expected value of the human mobility index using a machine learning model based on the confounding factors of the current time step learned by the time-varying confounding factor prediction module and the human mobility index; An objective function construction module is used to construct an objective function based on the conditional density calculated by the conditional density calculation module and the causal effect estimation result obtained by the causal effect estimation module; wherein the objective function is as described in formula (17); (17) In formula (17), represents the objective function, n represents the total number of observed samples; Indicates i Actual human mobility index for a sample of observations; Indicates i The predicted value of the human mobility index for a sample of observations; Indicates i Extreme weather data for observation samples; Indicates i Unobserved confounders of the observed sample; Indicates that given the unobserved confounder Under the conditions of extreme weather data The probability density of α is a regularization hyperparameter used to balance the weight of prediction error and conditional density estimation loss; The average dose-response curve drawing module is used to draw the average dose-response curve using the confounding factors of the current time step learned by the time-varying confounding factor prediction module and the extreme weather data captured by the extended observable data capture module, so as to obtain the causal effect of the continuous weather treatment level on the human mobility index; wherein the confounding factors of the current time step and the extreme weather data are as described in formula (18); and the average dose-response curve is as described in formula (19); (18) In formula (18), Indicates that at time step t , No. i The result of whether a typhoon occurs in the observation sample; Indicates that at time step t , No. i unobserved confounders learned from the observed sample; Indicates that at time step t , No. i Extreme weather data in observation samples; (19) In formula (19), E (⋅) indicates expected value; Indicates that at time step t Human Mobility Index; and Indicates that at time step t Other covariates or network structure data; Indicates that at time step t The level of weather treatment; A specific value representing a given weather treatment level.

[0017] Wherein, the conditional density calculation module is used to: Apply the activation function to estimate the conditional density under given confounding factors at multiple grid points through formula (10) to obtain the probability distribution; (10) In formula (10), Confounding factors The eigenvector of The result is J +1-dimensional real vector; Indicates that given the confounding factors Under the conditions of J + Conditional density at 1 grid point; , The dimension is The weight matrix of Confounding factors Dimensions; The conditional density is adjusted by ensuring that the estimated conditional density is continuous for any given value at a number of grid points from 0 to 1 and satisfies the valid density conditions that all probability values ​​are non-negative and that the sum of the entire probability distribution is 1; The specific value of the conditional density at each grid point is calculated based on formula (11) by linear interpolation to obtain the detailed probability distribution of different weather treatment levels under given confounding factors; (11) In formula (11), Indicates the corresponding weather data processing level , in j The conditional density estimate at grid points; where j ∈{0, 1, ..., J}, J The total number of grids minus one; By performing linear interpolation based on formula (12) at predefined grid points, the conditional density of any specific weather treatment level is accurately estimated to obtain a function that continuously and accurately reflects the probability distribution of the corresponding level; (12) In formula (12), Indicates that given the confounding factors Under the conditions of weather treatment level The conditional density estimate of ; and Represents the grid points and The conditional density estimate at ; and , which means less than or equal to The largest integer sum greater than or equal to The smallest integer of ; Represents the total number of grid points minus one.

[0018] Wherein, the human mobility index estimation module is used to: The human mobility index is defined under given extreme weather data and confounding factors by formula (13) to obtain the expected value of the human mobility index; (13) In formula (13), represents the expected value of the human mobility index; The continuous extreme weather data and confounding factors are used as input to train the multilayer perceptron network, and the extreme weather data and confounding factors are processed by integrating the variable coefficient prediction head through formula (14) to obtain the expected value of the human mobility index; where, (14) In formula (14), Represents the parameters of the variable coefficient prediction head; Indicates that it has parameters Neural network function; represents the confounding factors as input vector The eigenvector of By applying the activation function through formula (15) and the spline odd function modeling through formula (16) in the multilayer perceptron network, nonlinear changes and continuity modeling of the input data are performed to obtain a predicted value of the human mobility index with accuracy and flexibility; (15) In formula (15), represents the degree of the truncated basis, Indicates i The parameter vector of neurons depends on extreme weather data ; Represents the parameter vector The transpose of the input vector Perform dot product operation; (16) In formula (16), represents the linear combination of spline basis functions, represents the spline basis, Indicates i Neurons and l The coefficients between the spline basis functions; Indicates l spline basis functions.

[0019] To solve the above technical problems, another technical solution adopted by the present invention is: to provide an electronic device, including: a processor and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, when the processor executes the computer instructions, the electronic device executes the steps of the typhoon effect assessment method as described above.

[0020] In order to solve the above technical problems, another technical solution adopted by the present invention is: providing a readable storage medium, in which a computer program is stored. The computer program includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes the steps of the typhoon effect assessment method as described above.

[0021] The present invention provides a typhoon effect assessment method, device, electronic device and readable storage medium. By using extreme weather data, human mobility index, mobile flow network data and other covariate information, the recurrent neural network and graph convolutional network techniques are used to learn time-varying confounding factors and estimate continuous causal effects, thereby achieving accurate prediction of the impact on human mobility and accurate evaluation of causal relationships, thereby solving the technical problems of insufficient accuracy in predicting human mobility under extreme weather conditions and the difficulty in evaluating causal effects. In addition, by integrating the vigilance data of residents in different regions, a certain degree of personalized analysis is provided for different regions, and by using proxy variables and non-negative matrix decomposition and other technologies, the accuracy of human mobility prediction in different scenarios is improved while indirectly achieving privacy protection, thereby solving the technical problems of ignoring unobserved confounding factors, being difficult to adapt to the temporal variability of confounding factors, and balancing privacy protection and data utilization in existing studies. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of a typhoon effect assessment method provided by an embodiment of the present invention; Figure 2 yes Figure 1 A schematic diagram of a method flow chart of an embodiment of step S10; Figure 3 yes Figure 1 A schematic diagram of a method flow chart of an embodiment of step S11; Figure 4 yes Figure 3 A schematic diagram of a method flow of an embodiment of step S111; Figure 5 yes Figure 1 A method flow diagram of an embodiment of step S12; Figure 6 yes Figure 5 A method flow diagram of an embodiment of step S120; Figure 7 It is a schematic diagram of the module structure of a typhoon effect assessment device provided by an embodiment of the present invention; Figure 8 yes Figure 7 A schematic diagram of a module structure of an embodiment of a typhoon effect analysis unit; Fig. 9 yes Figure 7A schematic diagram of a module structure of an embodiment of a time-varying confounding factor learning unit; Fig.10 yes Figure 7 A module structure diagram of an embodiment of a causal effect evaluation unit in FIG. Fig.11 It is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention.

[0023] The reference numerals in the above drawings are described as follows: 20. Typhoon effect assessment device; 21. Typhoon effect analysis unit; 210. Typhoon data capture module; 211. Proxy variable definition module; 212. Time series analysis module; 213. Non-negative matrix decomposition module; 214. Time-space embedding generation module; 215. Causal network processing module; 22. Time-varying confounding factor learning unit; 220. Extended observable data capture module; 221. Causal network construction module; 222. Causal effect estimation module; 223. Recurrent neural network processing module; 224. Graph convolutional network processing module; 225. Time-varying confounding factor prediction module; 23. Causal effect evaluation unit; 230. Conditional density calculation module; 231. Human mobility index estimation module; 232. Objective function construction module; 233. Average dose-response curve drawing module; 3. Electronic device; 31. Processor; 32. Memory; 33. Input device; 34. Output device; 35. Computer program. DETAILED DESCRIPTION

[0024] In order to explain in detail the possible application scenarios, technical principles, specific schemes that can be implemented, and the purposes and effects that can be achieved, the following is a detailed description of the specific embodiments listed in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0025] Reference to "embodiment" herein means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present application. The term "embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or association with other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the various technical features mentioned in the embodiments can be combined in any way to form a corresponding implementable technical solution.

[0026] Unless otherwise defined, the technical terms used in this document have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms in this document is only for describing specific embodiments and is not intended to limit this application.

[0027] In the description of this application, the term "and / or" is an expression used to describe the logical relationship between objects, indicating that three relationships may exist, for example, A and / or B, which means: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this article generally indicates that the objects before and after are in an "or" logical relationship.

[0028] In the present application, terms such as “first” and “second” are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship of quantity, priority or sequence between these entities or operations.

[0029] Without further limitations, in this application, the words "include", "comprises", "has" or other similar open-ended expressions used in sentences are intended to cover non-exclusive inclusion. These expressions do not exclude the presence of additional elements in the process, method or product including the elements, so that the process, method or product including a series of elements may include not only those limited elements, but also other elements not explicitly listed, or also include elements inherent to such process, method or product.

[0030] Similar to the understanding in the Examination Guidelines, in this application, expressions such as "greater than", "less than", "exceed" and the like are understood to exclude the number itself; expressions such as "above", "below", "within" and the like are understood to include the number itself. In addition, in the description of the embodiments of this application, "multiple" means more than two (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups", "multiple times", etc., unless otherwise clearly and specifically limited.

[0031] In the description of the embodiments of the present application, space-related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or position relationship based on the orientation or position relationship shown in the specific embodiments or drawings, and are only for the convenience of describing the specific embodiments of the present application or facilitating the reader's understanding, and do not indicate or imply that the referred device or component must have a specific position, a specific orientation, or be constructed or operated in a specific orientation, and therefore cannot be understood as a limitation on the embodiments of the present application.

[0032] Typhoons are extremely destructive weather phenomena characterized by heavy rains and strong winds, which often lead to poor road conditions, reduce traffic operation efficiency, severely disrupt commercial activities and transportation, and have a significant impact on human mobility and traffic patterns. In the field of traffic data analysis and prediction technology, researchers are increasingly paying attention to traffic patterns under extreme weather conditions, which are of great significance for protecting the safety of life and property, maintaining urban operations, and disaster emergency management. However, existing studies often ignore unobserved confounding factors such as regional characteristics when analyzing the impact of typhoons on human mobility, such as topography, building density, etc. These factors may be associated with typhoon intensity and affect human mobility during typhoons, which may lead to deviations in the causal analysis process.

[0033] Based on this, the present application proposes a typhoon effect assessment method, which is based on the continuous causal effect estimation of neural networks. By combining recurrent neural networks and graph convolutional network technologies, the method learns the representation of time-varying confounding factors, reduces the impact of unobserved confounding factors, and evaluates the continuous causal response of extreme weather to human mobility, thereby improving the accuracy and reliability of human mobility prediction during typhoons. This application is used to solve the technical problem of ignoring unobserved confounding factors such as regional characteristics in the typhoon impact analysis research of the prior art. These confounding factors may be associated with typhoon intensity and affect human mobility during typhoons, resulting in deviations in the causal analysis process.

[0034] See also Figure 1 , is a flow chart of a typhoon effect assessment method in an embodiment of the present invention. The method comprises: Step S10, analyze the target typhoon data and combine the covariates of residents in different geographical areas to capture unobserved time-varying confounding factors.

[0035] Among them, the covariates include but are not limited to residents' alertness and basic flow patterns.

[0036] Please also see Figure 2 , step S10, i.e., analyzing the target typhoon data and combining the covariates of residents in different geographical areas to capture unobserved time-varying confounding factors, specifically including: Step S100, analyzing the target typhoon data to capture basic observable data, selecting and quantifying variables representing unobserved confounding factors based on the basic observable data to define as proxy variables, and converting the observable data into feature vectors.

[0037] Among them, the basic observable data include typhoon weather data, residents' alertness data and residents' basic flow pattern data.

[0038] Specifically, typhoon-related extreme weather data are obtained directly from the Meteorological Agency to obtain typhoon weather data. By selecting keywords related to the target typhoon from the search engine and assigning points based on their search share, the residents' alertness to typhoons is quantified to generate resident alertness data. The urban dynamics in the basic life pattern space of residents in different geographical areas (such as administrative districts) are simulated through the tensor decomposition method to obtain the basic flow pattern data of residents.

[0039] For example, the big data of two major typhoons in Japan (Haibis and Faxai) were selected as the target typhoon data. By analyzing these typhoon data and observing the changes in the mobility patterns of daily human travel before and after the typhoon, it was found that the traffic flow rate decreased significantly, thus proving the significant impact of unobserved confounding factors on human mobility between different geographical regions. The search engine-related keyword search index was used to reveal the impact of typhoons on residents' alertness, especially the significant increase in keyword searches for "Japanese Railways" within three days after the typhoon. The recovery process after a typhoon is often long and not completely immediate, which emphasizes the importance of estimating the causal effect of disasters through covariates such as residents' alertness.

[0040] Proxy variables here refer to those variables that can represent or indicate other variables that are not easy to directly observe or measure (such as unobserved time-varying confounding factors). In this embodiment, the proxy variables include the resident alertness data and the resident basic flow pattern data, because they can indicate or affect human mobility and are associated with the impact of typhoons.

[0041] Specifically, the proxy variable is expressed as formula (1): (1) in, t represents the time step, n Indicates the number of selected geographic regions.

[0042] Furthermore, the collected basic observable data (resident alertness and basic mobility patterns) are converted into feature vectors, which can be used as inputs to the model. Feature vectors refer to the conversion of raw data (such as search index, mobility pattern data) into a format suitable for machine learning model processing, which usually involves preprocessing steps such as data standardization, normalization or encoding, which will not be elaborated here.

[0043] Step S101, performing time series analysis on the basic flow pattern data of residents to capture its time dependency.

[0044] This step is to identify the temporal characteristics of travel patterns such as periodicity, trend and seasonality, which are crucial for the subsequent non-negative matrix factorization.

[0045] Step S102, decomposing the mobility tensors of different travel modes into basic life mode tensors through non-negative matrix decomposition.

[0046] Among them, the basic life pattern tensor includes the regional intensity change pattern tensor, the time period pattern tensor and the day pattern tensor.

[0047] Specifically, through non-negative tensor decomposition, formula (3) is used to decompose various mobility tensors of different travel modes into basic life pattern tensors described by three basic aspects: intensity change between regions, time period, and day.

[0048] (3) in, represents the liquidity tensor, The intensity matrix representing the basic life patterns, Coefficient matrix representing temporal patterns.

[0049] In this way, the understanding of complex liquidity data is simplified by extracting the main pattern tensors. Each pattern tensor represents a basic pattern or trend in the liquidity data. For example, the regional intensity change pattern tensor may reveal the intensity changes of liquidity between different regions, the period pattern tensor may indicate the liquidity changes at different time periods in a day, and the day pattern tensor may show the liquidity patterns of different days.

[0050] Step S103, forming a time-space embedding according to the region and time information of the residents' search for typhoon keywords, and inputting the time-space embedding into the causal network structure to capture unobserved confounding factors.

[0051] The confounding factors are defined by formula (2), as follows: (2) in, t is the time step, n is the number of selected geographic regions. These confounders represent unobserved time-varying confounders and are obtained by adding proxy variables obtained through analysis and learning.

[0052] Specifically, by embedding time-space data, combining the region (spatial information) and time information of residents' search for typhoon keywords, a comprehensive time-space data structure can be formed. The time-space embedding is used as the input of the causal network structure to control these unobserved confounding factors in the analysis, thereby reducing their impact on the causal effect estimation. This structure can more comprehensively describe and analyze the changes in human mobility patterns under the influence of typhoons.

[0053] In this way, the formation of time-space embedding and the process of inputting it into the causal network structure are actually trying to capture and control those unobserved time-varying confounding factors that may affect the impact of typhoons on human mobility. In this way, researchers can more accurately estimate the causal effect of typhoons on human mobility while reducing the estimation bias caused by unobserved confounding factors. This analytical method helps to improve the accuracy and reliability of causal inference, especially when dealing with complex and dynamically changing natural environments and socio-economic phenomena.

[0054] Step S11, capturing and learning the representation of time-varying confounders through machine learning based on the extended observable data, as well as the regional covariate data and unobserved confounders at the previous time step.

[0055] In this embodiment, the machine learning includes recurrent neural network and graph convolutional network technology.

[0056] Specifically, the recurrent neural network (RNN) is used to process time series data, capture dynamic features in historical covariate data, and extract historical information to generate hidden vectors, providing a basis for subsequent causal effect analysis. The graph convolutional network (GCN) is used to process mobile flow network data, capture the spatial dependencies between different regions, and combine this spatial information with the temporal information extracted by the RNN to learn the confounding factor representation of the current time step.

[0057] Please also see Figure 3 , step S11, i.e., capturing and learning the representation of time-varying confounders through machine learning based on the extended observable data, as well as the regional covariate data and unobserved confounders at the previous time step.

[0058] Step S110: capturing the extended observable data.

[0059] The extended observable data include extreme weather data , Human Mobility Index , mobile streaming network wait.

[0060] Capturing the extended observable data, specifically including: By collecting typhoon-related weather data from the Japan Meteorological Agency , including the rainfall and wind speed of the stage, the time interval data set is set to 10 minutes; Analyze mobility changes based on travel mode selection based on tagged GPS data and obtain the human mobility index , implement a random forest classifier on the remaining dataset based on these labeled training data samples; and Construct a mobile flow network between regions, simulate the potential relationship between unobserved confounding factors in the current time step, and obtain the mobile flow network through renormalization processing .

[0061] For example, it can be generated based on the human GPS trajectory big data provided by Blogwatcher. Relying on the self-developed visualization system, a small part of mobile GPS data is divided into several segments based on the map matching algorithm, and the real travel mode selection of each GPS trajectory is manually annotated according to the speed, distance and matching results corresponding to the public transportation network (such as subway, highway, sidewalk).

[0062] The mobile flow network is a graph in which nodes represent different regions (such as counties) and edges represent the flow between these regions. It is used to simulate the movement patterns of humans during typhoons. By collecting traffic volume data between different regions during typhoons, which can come from GPS tracking, public transportation records, or other traffic monitoring systems. The mobile flow network is time-varying, and at each time step t , the network represents the traffic flow status at that specific moment. Each edge in the network represents the traffic volume between a pair of regions within the typhoon time range, which can be the number of vehicles, passengers, or other appropriate flow volume measures. The mobile flow network is renormalized to adjust the weights of the edges in the network so that it reflects the actual flow pattern, highlights the most important flow paths in the network, and reduces noise or unimportant flows. After renormalization, the variable is a n × n The matrix of n is the number of selected regions. Each element of a ij Indicates that at time step t Time, Region i and Region j The standardized flow between is as shown in formula (4): (4) in, R Represents the set of real numbers.

[0063] Step S111, based on the extended observable data, causal effect estimation is performed by constructing a causal network structure and calculating the average treatment effect.

[0064] Among them, through confounding factors Implementing a causal network structure for assessing extreme weather How to affect each time period t Different human mobility indices .

[0065] Please also see Figure 4 Step S111, i.e., estimating the causal effect by constructing a causal network structure and calculating the average treatment effect according to the extended observable data, specifically includes: Step S1110, constructing a causal network structure based on the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur according to the extended observable data; Specifically, use and Respectively represent at a certain time step t The potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur are used to construct a causal network structure, as shown in formulas (5) and (6), respectively: (5) (6) In formulas (5) and (6), and Respectively represent the time step t potential outcomes of the human mobility index for typhoon occurrence and typhoon non-occurrence; Indicates that at time step t within, no. n potential results of the human mobility index for each region during a typhoon; Indicates that at time step t within, no. n Potential results of the human mobility index for each region when a typhoon does not occur.

[0066] Step S1111, calculate the time step t The average treatment effect is obtained by averaging the difference between the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur across all regions. To estimate causal effects.

[0067] Specifically, the time step is calculated by formula (7): t The average treatment effect is obtained by averaging the difference between the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur across all regions. To estimate causal effects.

[0068] (7) in, Indicates that at time step t , for the region i , potential results of the human mobility index during typhoons; Indicates that at time step t , for the region i, potential results of the human mobility index when the typhoon did not occur; Indicates that at time step t ,area i unobserved confounders; Indicates that at time step t ,area i extreme weather data; E(⋅) represents the expected value, which is used to estimate the potential results under given confounding factors and weather data conditions; Represents the time step t The average treatment effect (ATE).

[0069] Step S112, using a recurrent neural network to process historical covariate data to capture dynamic features in time series data, and using a graph convolutional network to process the mobile flow network to capture spatial dependencies, so as to predict time-varying confounding factors.

[0070] Specifically, a recurrent neural network (RNN) is used to extract historical information from the changing variables (i.e., confounding factors, treatments, and outcomes) through formula (8) to capture the dynamic characteristics in time series data. The historical information is represented by the hidden vector It is expressed as shown in formula (8): (8) in, Indicates that at time step t -1, which captures the time steps up to t -1 historical information; RNN stands for recurrent neural network, a neural network that can process sequence data and capture time dependencies; Indicates that at time step t A hidden vector of -2, which serves as the initial state of the RNN; Indicates that at time step t -1 unobserved confounder; Indicates that at time step t -1 covariate; Indicates that at time step t -1 extreme weather data; Indicates that at time step t The human mobility result of -1; ⊕ represents the concatenation operation, which connects different vectors to form a longer vector.

[0071] Regions with large population flows usually have similar confounding factors. Formula (9) is combined with a graph convolutional network (GCN) to model the relationship information between all regions, and two layers of GCNs are used to obtain the confounding factors of the current time period from the current flow network. , which includes the covariate and hidden vector .

[0072] Combining RNNs and GCNs to learn time-varying confounding factors through formula (9) for further causal analysis of the network.

[0073] (9) in, Indicates that at time step t unobserved confounders; Indicates that at time step t The mobile flow network matrix; ReLU represents the rectified linear unit activation function, which is used to introduce nonlinearity and accelerate the learning of the model; and Respectively represent the two-layer model parameters of GCN; Indicates that at time step t -1 covariate; Indicates that at time step t -1 hidden vector; ⊕ represents a concatenation operation, which connects different vectors to form a longer vector.

[0074] Step S12, calculate the conditional density based on the confounding factors and observable data for analysis at the current time step, estimate the expected value of the human mobility index, construct an objective function containing the estimate, and plot the average dose-response curve to evaluate the causal effect of continuous weather treatment levels on human mobility.

[0075] Please also see Figure 5 , step S12, i.e., calculating the conditional density based on the confounding factors and observable data for analysis at the current time step, estimating the expected value of the human mobility index, constructing an objective function containing the estimation, and drawing an average dose-response curve to evaluate the causal effect of the continuous weather treatment level on human mobility, specifically includes the following steps: Step S120, based on the learned confounding factors of the current time step and the observable data for analysis, the conditional density is calculated using probability density estimation.

[0076] The probability density estimation may be a technique such as kernel density estimation or maximum likelihood estimation. The observable data for analysis include typhoon impact data and human mobility data.

[0077] Please also see Figure 6 , step S120, that is, based on the confounding factors of the current time step obtained by learning and the observable data for analysis, the conditional density is calculated using probability density estimation, specifically including: Step S1200, applying an activation function to estimate the conditional density under a given confounding factor condition at a plurality of grid points to obtain a probability distribution; Specifically, the activation function is applied to estimate the conditional density under given confounding factors at multiple grid points through formula (10) to obtain the probability distribution.

[0078] Normalize the continuous Divide evenly into J grid, and a simple network using linear interpolation To estimate the conditional density on the grid points. Apply the softmax activation function to the given confounding factor through formula (10) = z Conditional density under conditions exist J +1 grid points to obtain the probability distribution, as shown in formula (10): (10) in, , The dimension is The weight matrix of Confounding factors The softmax function converts the input vector into a probability distribution so that the sum of all output values ​​is 1. The result is J +1-dimensional real vector.

[0079] Step S1201, adjusting the conditional density by ensuring that the estimated conditional density is continuous for any given value at multiple grid points from 0 to 1 and satisfies the valid density conditions that all probability values ​​are non-negative and the sum of the entire probability distribution is 1; Step S1202, calculating the specific value of the conditional density at each grid point by linear interpolation to obtain the detailed probability distribution of different weather treatment levels under given confounding factors; Specifically, the specific value of the conditional density at each grid point is calculated based on formula (11) by linear interpolation to obtain the detailed probability distribution of different weather treatment levels under given confounding factors.

[0080] Estimating conditional density at the softmax layer exist J +1 On the grid point interval [0,1], for any given , the estimated is always continuous, so it needs to be rescaled to a valid density condition, that is, and , and calculate the conditional density by linear interpolation exist J The specific value at +1 grid point is as shown in formula (11): (11) in, Indicates the corresponding weather data processing level , in j The conditional density estimate at grid points; where j ∈{0, 1, ..., J}, J The total number of grid cells minus one.

[0081] Step S1203, by performing linear interpolation processing on predefined grid points, the conditional density of any specific weather treatment level is accurately estimated to obtain a function that continuously and accurately reflects the probability distribution at the corresponding level; Specifically, by performing linear interpolation processing based on formula (12) at predefined grid points, the conditional density of any specific weather treatment level is accurately estimated to obtain a function that continuously and accurately reflects the probability distribution at the corresponding level; For any weather processing level The conditional density The estimate, through J +1 grid points are linearly interpolated to obtain the value, as shown in formula (12): (12) in, and Respectively means less than or equal to The largest integer sum greater than or equal to The smallest integer of ; and Respectively, at the grid points and The conditional density estimate at ; Indicates that given the confounding factors Under the conditions of weather treatment level The conditional density estimate of ; and Represents the grid points and The conditional density estimate at ; Represents the total number of grid points minus one.

[0082] Step S121, based on the learned confounding factors of the current time step and the human mobility index, the expected value of the human mobility index is calculated using a machine learning model.

[0083] Specifically, given extreme weather data, and confounding factors Under the condition of Expected value The goal of the resulting estimator, expressed by equation (13), is to predict the expected value of human mobility under specific weather conditions and confounding factors.

[0084] (13) In order to avoid the problem of discontinuous average dose-response curve caused by discretizing extreme weather data, the present invention does not discretize the weather processing of the result estimator, but directly converts the continuous extreme weather data into and confounding factors As input, we can directly train a simple network such as multi-layer perceptrons (MLPs) to As input, to simplify the operation and output the result. However, due to the way this method is handled in the network model Similar, resulting in a probability loss related To solve the problem of loss caused by direct input To solve the problem of information, a variable coefficient prediction head is introduced, the parameters of which depend on the processing data. w , as shown in formula (14): (14) in, represents a neural network with parameters Depends on processing data Equation (14) allows the model to adjust its predictions of human mobility according to different weather conditions.

[0085] Neural Networks right z Apply the ReLU activation function, as shown in formula (15): (15) in, represents the degree of the truncated basis, is dependent on w The network parameters are as follows. Formula (15) is z and The dot product result is passed through the ReLU activation function to generate a nonlinear prediction.

[0086] In order to flexibly and continuously model the processing effect w , using the spline odd function w It is modeled as a linear combination of basis functions, as shown in formula (16): (16) in, represents the spline basis, Formula (16) allows the model to depend on the treatment effect in a flexible and continuous way. w , By changing the parameters to effectively influence the human mobility index.

[0087] Step S122, constructing an objective function based on the conditional density and the result estimation.

[0088] Specifically, using the conditional density and results , and perform the resulting estimator and the conditional density estimator The prediction loss is calculated, and the parameter α is used to balance these losses to obtain the objective function that comprehensively considers the prediction accuracy and probability density fit .

[0089] In the above, we get the conditional density and results Then, use these values ​​to construct the objective function , as shown in formula (17): (17) in, The resulting estimator is calculated The prediction loss of and, that is, the actual observation value and expected value The difference between. Based on the negative log-likelihood, the conditional density estimator is calculated The prediction loss, that is, the conditional density The logarithm of . α is a regularization hyperparameter used to balance the weight of prediction error and conditional density estimation loss; by manually setting the hyperparameter α , you can adjust α The value of controls the emphasis of the model on prediction accuracy (the first term) and conditional density estimate accuracy (the second term).

[0090] Using the learned time-varying confounding factors, density estimators and outcome estimators are combined for causal effect estimation, and average dose-response curves are introduced to evaluate the changes in human mobility under different weather treatment levels in consecutive time steps.

[0091] Step S123, using the confounding factors of the current time step obtained based on learning and the extreme weather data, an average dose-response curve is plotted to obtain the causal effect of the continuous weather treatment level on the human mobility index.

[0092] Among them, the relationship curve between continuous weather data and mobility index was drawn according to the "causal effect of continuous weather treatment level on human mobility index", avoiding reliance on the assumption of potential causal relationship.

[0093] Specifically, based on the aforementioned step S111 "According to the extended observable data, causal effect estimation is performed by constructing a causal network structure and calculating the average treatment effect" the causal effect of certain binary weather treatments on human mobility can be captured. Since weather intensity is not a constant, it is assumed that all potential confounding factors are independent results when the treatment distribution adjustment is performed. In order to handle this continuity, an evaluation dose-response curve (ADRF) is introduced, based on the observed sample, including the results of whether the typhoon occurs , the learned unobserved confounders and extreme weather data , as shown in formula (18); through these data, a curve between continuous weather data and liquidity index is drawn, as shown in formula (19): (18) in, t is the time step, Indicates that at time step t , No. i The result of whether a typhoon occurs in the observation sample; Indicates that at time step t , No. i unobserved confounders learned from the observed sample; Indicates that at time step t , No. i Extreme weather data in observation samples; n is the total number of observed samples.

[0094] (19)

[0095] in, E (⋅) indicates expected value; represents the human mobility index; and may represent other covariates or network structure that are related to the level of weather treatment together, for predicting human mobility; Indicates a given level of weather treatment.

[0096] The expected value of human mobility under specific weather conditions is calculated by formula (19), which is part of the average dose-response curve (ADRF) and is used to evaluate the impact of weather conditions on human mobility. Through this formula, we can understand the changing trend of human mobility under different weather intensity levels and avoid relying on observation samples and different assumptions about potential causal relationships.

[0097] As mentioned above, by using extreme weather data, human mobility index, mobile flow network data and other covariate information, using recurrent neural network and graph convolutional network technology to learn time-varying confounding factors and estimate continuous causal effects, we can achieve accurate prediction of the impact on human mobility and accurate evaluation of causal relationships, thereby solving the technical problems of insufficient accuracy in predicting human mobility under extreme weather conditions and the difficulty in evaluating causal effects. In addition, by integrating the vigilance data of residents in different regions, a certain degree of personalized analysis is provided for different regions, and by using proxy variables and non-negative matrix decomposition and other technologies, the accuracy of human mobility prediction in different scenarios is improved while indirectly achieving privacy protection, thereby solving the technical problems of ignoring unobserved confounding factors, difficulty in adapting to the temporal variability of confounding factors, and the balance between privacy protection and data utilization in existing studies.

[0098] The present invention is not only applicable to the assessment of the impact of typhoons on human mobility, but can also be widely used in multiple fields that need to assess and predict the impact of extreme events on human activities, providing scientific data support and decision-making basis. For example, assessing the impact of floods and rainstorms on human mobility to provide decision support for emergency evacuation and rescue operations; analyzing the changes in human mobility after an earthquake to optimize the allocation of rescue resources and post-disaster reconstruction plans; in urban planning, predicting the impact of extreme weather on traffic flow and optimizing the design and layout of transportation infrastructure; analyzing the impact of extreme weather on agricultural production to guide the prevention and response measures for agricultural disasters.

[0099] Please also see Figure 7 , is a schematic diagram of the module structure of a typhoon effect assessment device provided by an embodiment of the present invention. Corresponding to the above-mentioned typhoon effect assessment method, a typhoon effect assessment device 20 provided by an embodiment of the present invention includes: a typhoon effect analysis unit 21, a time-varying confounding factor learning unit 22, and a causal effect assessment unit 23.

[0100] The typhoon effect analysis unit 21 is used to analyze target typhoon data and capture unobserved time-varying confounding factors in combination with covariates of residents in different geographical areas.

[0101] Among them, the covariates include but are not limited to residents' alertness and basic flow patterns.

[0102] The time-varying confounding factor learning unit 22 is used to capture and learn the representation of time-varying confounding factors through machine learning based on the extended observable data and the regional covariate data and unobserved confounding factors of the previous time step generated by the typhoon effect analysis unit 21.

[0103] In this embodiment, the machine learning includes recurrent neural network and graph convolutional network technology.

[0104] The causal effect evaluation unit 23 is used to calculate the conditional density based on the confounding factors of the current time step generated by the time-varying confounding factor learning unit 22 and the observable data for analysis, estimate the expected value of the human mobility index, construct an objective function containing the estimation, and draw an average dose-response curve to evaluate the causal effect of continuous weather treatment levels on human mobility.

[0105] See also Figure 8 The typhoon effect analysis unit 21 includes: a typhoon data capture module 210, a proxy variable definition module 211, a time series analysis module 212, a non-negative matrix decomposition module 213, a time-space embedding generation module 214, and a causal network processing module 215.

[0106] The typhoon data capturing module 210 is used to analyze target typhoon data to capture basic observable data.

[0107] Among them, the basic observable data include typhoon weather data, residents' alertness data and residents' basic flow pattern data.

[0108] Specifically, the typhoon data capture module 210 directly obtains typhoon-related extreme weather data from the Meteorological Bureau to obtain typhoon weather data; quantifies residents' alertness to typhoons by selecting keywords related to the target typhoon from the search engine and assigning scores according to their search share to generate resident alertness data; and simulates the urban dynamics in the basic life pattern space of residents in different geographical areas (such as administrative districts) through the tensor decomposition method to obtain residents' basic flow pattern data.

[0109] The proxy variable definition module 211 is used to select and quantify variables representing unobserved confounding factors as proxy variables based on the basic observable data captured by the typhoon data capture module 210, and convert the observable data into feature vectors.

[0110] Proxy variables are variables that can represent or indicate other variables that are not easily observed or measured directly (such as unobserved time-varying confounding factors). In this example, residents’ alertness and basic mobility patterns are selected as proxy variables because they can indicate or affect human mobility and are associated with the impact of typhoons.

[0111] Specifically, the proxy variable is expressed as formula (1): (1) in, t represents the time step, n Indicates the number of selected geographic regions.

[0112] Furthermore, the proxy variable definition module 211 converts the basic observable data (resident alertness and basic flow patterns) collected by the typhoon data capture module 210 into feature vectors, which can be used as inputs to the model. The feature vector refers to the conversion of raw data (such as search index, flow pattern data) into a format suitable for machine learning model processing, which usually involves pre-processing steps such as data standardization, normalization or encoding, which will not be elaborated here.

[0113] The time series analysis module 212 is used to perform time series analysis on the basic flow pattern data of residents captured by the typhoon data capture module 210 to capture its time dependency.

[0114] The non-negative matrix decomposition module 213 is used to decompose the mobility tensors of different travel modes into basic life mode tensors through non-negative matrix decomposition.

[0115] Among them, the basic life pattern tensor includes the regional intensity change pattern tensor, the time period pattern tensor and the day pattern tensor.

[0116] Specifically, the non-negative matrix decomposition module 213 implements non-negative tensor decomposition through formula (3), and decomposes various mobility tensors of different travel modes into basic life pattern tensors described by three basic aspects: intensity change between regions, time period, and day.

[0117] (3) in, represents the liquidity tensor, The intensity matrix representing the basic life patterns, coefficient matrix representing the temporal pattern; The time-space embedding generation module 214 is used to form a time-space embedding according to the area and time information of residents searching for typhoon keywords captured by the typhoon data capture module 210.

[0118] The causal network processing module 215 is used to input the time-space embedding generated by the time-space embedding generation module 214 into the causal network structure to capture unobserved confounding factors.

[0119] The confounding factors are defined by formula (2), as follows: (2) in, t is the time step, n is the number of selected geographic regions. These confounders represent unobserved time-varying confounders and are obtained by adding proxy variables obtained through analysis and learning.

[0120] Specifically, the causal network processing module 215 can form a comprehensive time-space data structure by embedding time-space data and combining the region (spatial information) and time information of residents' search for typhoon keywords. The time-space embedding is used as the input of the causal network structure to control these unobserved confounding factors in the analysis, thereby reducing their impact on the causal effect estimation. This structure can more comprehensively describe and analyze the changes in human mobility patterns under the influence of typhoons.

[0121] Please also see Fig. 9 The time-varying confounding factor learning unit 22 includes: an extended observable data capture module 220, a causal network construction module 221, a causal effect estimation module 222, a recurrent neural network processing module 223, a graph convolutional network processing module 224, and a time-varying confounding factor prediction module 225.

[0122] The extended observable data capturing module 220 is used to capture the extended observable data.

[0123] The extended observable data include extreme weather data , Human Mobility Index , mobile streaming network wait.

[0124] Specifically, the extended observable data acquisition module 220 collects terminal weather data related to typhoons from the Japan Meteorological Agency. , including the rainfall and wind speed of the stage, the time interval data set is set to 10 minutes; based on the travel mode selection of the marked GPS data, the mobility changes are analyzed to obtain the human mobility index ; Construct a mobile flow network between regions, simulate the potential relationship between unobserved confounding factors in the current time step, and obtain the mobile flow network through renormalization .

[0125] The renormalized mobile flow network is shown in formula (4): (4) In formula (4), the mobile flow network variable is a n × n The matrix ofn Indicates the number of selected regions; R Represents the set of real numbers.

[0126] The causal network construction module 221 is used to construct a causal network structure according to the extended observable data captured by the extended observable data capture module 220.

[0127] The causal network building module 221 is constructed by confounding factors. Implementing a causal network structure for assessing extreme weather How does it affect different human mobility indices for each time period t .

[0128] Furthermore, the causal network construction module 221 constructs a causal network structure based on the extended observable data and the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur.

[0129] Specifically, the causal network construction module 221 uses and Respectively represent at a certain time step t The potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur construct the causal network structure, as shown in formulas (5) and (6) respectively: (5) (6) In formulas (5) and (6), and Respectively represent the time step t potential outcomes of the human mobility index for typhoon occurrence and typhoon non-occurrence; Indicates that at time step t within, no. n potential results of the human mobility index for each region during a typhoon; Indicates that at time step t within, no. n Potential results of the human mobility index for each region when a typhoon does not occur.

[0130] The causal effect estimation module 222 is used to calculate the average treatment effect through the causal network structure constructed by the causal network construction module 221 to perform causal effect estimation.

[0131] Specifically, the causal effect estimation module 222 calculates the time period by formula (7): t The average difference between all regions in the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur is used to obtain the average treatment effect for causal effect estimation, as shown below: (7) in, t is the time step, Indicates that at time step t , for the region i , potential results of the human mobility index during typhoons; Indicates that at time step t , for the region i , potential results of the human mobility index when the typhoon did not occur; Indicates that at time step t ,area i unobserved confounders; Indicates that at time step t ,area i extreme weather data; E(⋅) represents the expected value, which is used to estimate the potential results under given confounding factors and weather data conditions; Represents the time step t Average Treatment Effect (ATE)

[0132] The recurrent neural network processing module 223 is used to process historical covariate data using a recurrent neural network to capture dynamic features in time series data.

[0133] Specifically, the recurrent neural network processing module 223 uses a recurrent neural network (RNN) to extract historical information from the changing variables (i.e., confounding factors, treatments, and results) through formula (8) to capture the dynamic characteristics in the time series data. The historical information is represented by the hidden vector It is expressed as shown in formula (8): (8) in, Indicates that at time step t -1, which captures the time steps up to t -1 historical information; RNN stands for recurrent neural network, a neural network that can process sequence data and capture time dependencies; Indicates that at time step t A hidden vector of -2, which serves as the initial state of the RNN; Indicates that at time step t -1 unobserved confounder; Indicates that at time step t -1 covariate; Indicates that at time step t -1 extreme weather data; Indicates that at time stept The human mobility result of -1; ⊕ represents the concatenation operation, which connects different vectors to form a longer vector.

[0134] The graph convolutional network processing module 224 is used to use a graph convolutional network to process the mobile flow network to capture spatial dependencies.

[0135] Regions with large population flows usually have similar confounding factors. The graph convolutional network processing module 224 models the relationship information between all regions by combining the graph convolutional network (GCN) with formula (9), and uses two layers of GCNs to obtain the confounding factors of the current time period from the current flow network. , which includes the covariate and hidden vector .

[0136] The time-varying confounding factor prediction module 225 is used to predict the time-varying confounding factor by combining the captured results of the recurrent neural network processing module 223 and the graph convolution network processing module 224 through formula (9). prediction.

[0137] (9) in, Indicates that at time step t unobserved confounders; Indicates that at time step t The mobile flow network matrix; ReLU represents the rectified linear unit activation function, which is used to introduce nonlinearity and accelerate the learning of the model; and Respectively represent the two-layer model parameters of GCN; Indicates that at time step t -1 covariate; Indicates that at time step t -1 hidden vector; ⊕ represents a concatenation operation, which connects different vectors to form a longer vector.

[0138] Please also see Fig.10 The causal effect evaluation unit 23 includes: a conditional density calculation module 230, a human mobility index estimation module 231, an objective function construction module 232, and an average dose-response curve drawing module 233.

[0139] The conditional density calculation module 230 is used to calculate the conditional density by using probability density estimation based on the confounding factors of the current time step learned by the time-varying confounding factor prediction module 225 and the observable data for analysis.

[0140] The probability density estimation may be a technique such as kernel density estimation or maximum likelihood estimation. The observable data for analysis include typhoon impact data and human mobility data.

[0141] Furthermore, the conditional density calculation module 230 applies an activation function to estimate the conditional density under given confounding factors at multiple grid points through formula (10) to obtain a probability distribution.

[0142] Specifically, the continuous normalization process Divide evenly into J grid, and a simple network using linear interpolation To estimate the conditional density on the grid points. In this embodiment, the conditional density calculation module 230 applies the softmax activation function to the given confounding factor by formula (10). = z Conditional density under conditions exist J +1 grid points to obtain the probability distribution, as shown in formula (10): (10) in, , The dimension is The weight matrix of Confounding factors The softmax function converts the input vector into a probability distribution so that the sum of all output values ​​is 1. The result is J +1-dimensional real vector.

[0143] The conditional density calculation module 230 also adjusts the conditional density by ensuring that the estimated conditional density is continuous for any given value at multiple grid points from 0 to 1 and satisfies the valid density conditions that all probability values ​​are non-negative and the sum of the entire probability distribution is 1.

[0144] Specifically, the conditional density is estimated in the softmax layer exist J +1 On the grid point interval [0,1], for any given , the estimated is always continuous, so it needs to be rescaled to a valid density condition, that is, and , and calculate the conditional density by linear interpolation exist J The specific value at +1 grid point is as shown in formula (11): (11) in, Indicates the corresponding weather data processing level , in j The conditional density estimate at grid points; where j ∈{0, 1, ..., J}, J The total number of grid cells minus one.

[0145] The conditional density calculation module 230 also accurately estimates the conditional density of any specific weather treatment level by performing linear interpolation processing on predefined grid points based on formula (12) to obtain a function that continuously and accurately reflects the probability distribution at the corresponding level.

[0146] Specifically, for any weather processing level The conditional density The conditional density calculation module 230 calculates the conditional density by J +1 grid points are linearly interpolated to obtain the value, as shown in formula (12): (12) in, and Respectively means less than or equal to The largest integer sum greater than or equal to The smallest integer of ; and Respectively, at the grid points and The conditional density estimate at ; Indicates that given the confounding factors Under the conditions of weather treatment level The conditional density estimate of ; and Represents the grid points and The conditional density estimate at ; Represents the total number of grid points minus one.

[0147] The human mobility index estimation module 231 is used to calculate the expected value of the human mobility index using a machine learning model based on the confounding factors of the current time step learned by the time-varying confounding factor prediction module 225 and the human mobility index.

[0148] Specifically, given extreme weather data, and confounding factors Under the condition of Expected value The goal of the resulting estimator, expressed by equation (13), is to predict the expected value of human mobility under specific weather conditions and confounding factors.

[0149] (13) In order to avoid the problem of discontinuous average dose-response curve caused by discretizing extreme weather data, the present invention does not discretize the weather processing of the result estimator, but directly converts the continuous extreme weather data into and confounding factors As input, we can directly train a simple network such as multi-layer perceptrons (MLPs) to As input, to simplify the operation and output the result. However, due to the way this method is handled in the network model Similar, resulting in a probability loss related To solve the problem of loss caused by direct input To solve the problem of information, a variable coefficient prediction head is introduced, the parameters of which depend on the processing data. w , as shown in formula (14): (14) in, represents a neural network with parameters Depends on processing data Equation (14) allows the model to adjust its predictions of human mobility according to different weather conditions.

[0150] Neural Networks right z Apply the ReLU activation function, as shown in formula (15): (15) in, represents the degree of the truncated basis, is dependent on w The network parameters are as follows. Formula (15) is z and The dot product result is passed through the ReLU activation function to generate a nonlinear prediction.

[0151] In order to flexibly and continuously model the processing effect w , using the spline odd function w It is modeled as a linear combination of basis functions, as shown in formula (16): (16) in, represents the spline basis, Formula (16) allows the model to depend on the treatment effect in a flexible and continuous way.w , By changing the parameters to effectively influence the human mobility index.

[0152] The objective function construction module 232 is used to construct an objective function based on the conditional density calculated by the conditional density calculation module 230 and the causal effect estimation result obtained by the causal effect estimation module 222.

[0153] Specifically, the objective function construction module 232 uses the conditional density and results , and perform the resulting estimator and the conditional density estimator The prediction loss is calculated, and the parameter α is used to balance these losses to obtain the objective function that comprehensively considers the prediction accuracy and probability density fitting degree .

[0154] In the above, we get the conditional density and results Then, use these values ​​to construct the objective function , as shown in formula (17): (17) in, The resulting estimator is calculated The prediction loss of and, that is, the actual observation value and expected value The difference between. Based on the negative log-likelihood, the conditional density estimator is calculated The prediction loss, that is, the conditional density The logarithm of . α is a regularization hyperparameter used to balance the weight of prediction error and conditional density estimation loss; by manually setting the hyperparameter α , you can adjust α The value of controls the emphasis of the model on prediction accuracy (the first term) and conditional density estimate accuracy (the second term).

[0155] The average dose-response curve drawing module 233 is used to draw the average dose-response curve using the confounding factors of the current time step learned by the time-varying confounding factor prediction module 225 and the extreme weather data captured by the extended observable data capture module 220, so as to obtain the causal effect of the continuous weather treatment level on the human mobility index.

[0156] Specifically, the average dose-response curve drawing module 233 is based on the observation samples, including the results of whether the typhoon occurs. , the learned unobserved confounders and extreme weather data , as shown in formula (18); through these data, a curve between continuous weather data and liquidity index is drawn, as shown in formula (19): (18) in, t is the time step, Indicates that at time step t , No. i The result of whether a typhoon occurs in the observation sample; Indicates that at time step t , No. i unobserved confounders learned from the observed sample; Indicates that at time step t , No. i Extreme weather data in observation samples; n is the total number of observed samples.

[0157] (19)

[0158] in, E (⋅) indicates expected value; represents the human mobility index; and may represent other covariates or network structure that are related to the level of weather treatment together, for predicting human mobility; Indicates a given weather treatment level.

[0159] The average dose-response curve drawing module 233 draws a curve between continuous weather data and mobility index, avoiding dependence on observation samples and different assumptions of potential causal relationships.

[0160] The advantages and beneficial effects of a typhoon effect assessment method have been described above and will not be repeated here. Since a typhoon effect assessment device is applied to a typhoon effect assessment method, it satisfies the typhoon effect assessment method and has the same advantages and beneficial effects.

[0161] One of the embodiments of the present invention further provides a readable storage medium, in which a computer program is stored. The computer program includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes the steps of the typhoon effect assessment method described in any of the above embodiments.

[0162] One of the embodiments of the invention also provides an electronic device, including: a processor and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, when the processor executes the computer instructions, the electronic device executes the steps of the typhoon effect assessment method described in any of the above embodiments.

[0163] See also Fig.11 , is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention.

[0164] The electronic device 3 includes a processor 31, a memory 32, an input device 33, and an output device 34. The processor 31, the memory 32, the input device 33, and the output device 34 are coupled via a connector, and the connector includes various interfaces, transmission lines, or buses, etc., which are not limited in the embodiments of the present invention. It should be understood that in various embodiments of the present invention, coupling refers to mutual connection in a specific manner, including direct connection or indirect connection through other devices, for example, through various interfaces, transmission lines, buses, etc.

[0165] The processor 31 described in the embodiment of the present application can be implemented by hardware, firmware, software or a combination thereof, and can use a circuit, a single or multiple application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD), a programmable logic device (Programmable Logic Device, PLD), a field programmable gate array (Field Programmable Gate Array, FPGA), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, at least one of a microprocessor, and also includes other physical, biological or chemical structures that can achieve similar or equivalent functions to the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of the present application, or any combination of the steps mentioned therein. In the electronic device 3 of this embodiment, the processor 31 can be a single processor or a combination of two or more processors in terms of physical entity, such as a combination of a CPU and a GPU, a combination of a CPU and an FPGA, a combination of a CPU and an ASIC, a combination of a CPU and a coprocessor, and a combination of a CPU and a DSP, etc. In the embodiment, these combinations are collectively referred to as processors like a single processor.

[0166] The computer program 35 involved in the embodiment can be stored in a computer device readable storage medium, which includes but is not limited to a disk, a tape, a magnetic card, a floppy disk, a flash memory, an optical disk, an optical card, a read-only memory (ROM), a random access memory (RAM), an erasable programmable ROM (EPROM) and an electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical or chemical structures that can achieve similar or equivalent functions to the storage media listed above, such as DNA, RNA, protein and other units with information storage capabilities. In a specific embodiment, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiment can be stored in a single medium in a centralized manner, or can be stored in multiple media in a distributed manner.

[0167] The memory 32 containing a computer device readable storage medium may be a non-volatile memory or a random access memory. These computer device readable storage media may be built into the device, or may be connected to the device involved in the embodiment as an external device or part of an external device. In some embodiments, the memory with a computer device readable storage medium is deployed locally; in other embodiments, a scheme of deploying the memory away from the processor may also be adopted, such as a network attached memory accessed via an RF circuit or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), etc., or a suitable combination thereof, as long as the computer device can access the memory. In addition, the computer program involved in the embodiment may be stored in plaintext / ciphertext form, or may be designed as training data, which may be integrated and reorganized by model training and implicitly stored in the parameter state of a deep neural network or other machine learning model.

[0168] like Fig.11 In the illustrated embodiment, the computer program 35 is stored in the memory 32 of the electronic device 3, but the reader should be aware that other solutions that enable the processor to execute the computer program are feasible.

[0169] The input device 33 is used to input data and / or signals, and the output device 34 is used to output data and / or signals. The output device 34 and the input device 33 may be independent devices or an integrated device.

[0170] It is understandable that in the embodiment of the present invention, the memory 32 is not only used to store related instructions, and the embodiment of the present invention does not limit the specific data stored in the memory.

[0171] Understandably, Fig.11Only a simplified design of an electronic device is shown. In practical applications, the electronic device may also include other necessary components, including but not limited to any number of input / output devices, processors, memories, etc., and all ship flow prediction methods based on spatiotemporal multi-graph convolutional networks that can implement the embodiments of the present invention are within the protection scope of the present invention.

[0172] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical or other forms.

[0173] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0174] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0175] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the technical solution of the present invention can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a management server, or a network device, etc.) or a processor to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (English: read-only memory, abbreviated: ROM), random access memory (English: Random Access Memory, abbreviated: RAM), disk or optical disk and other media that can store program codes.

[0176] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of this application, this does not limit the scope of patent protection of this application. All technical solutions generated by replacing or modifying equivalent structures or equivalent processes based on the essential concept of this application using the contents recorded in the specification and drawings of this application, as well as directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are included in the scope of patent protection of this application.

Claims

1. A typhoon effect assessment method, characterized in that: The method comprises: Analyze the target typhoon data and incorporate covariates of residents in different geographical areas to capture unobserved time-varying confounding factors; Capturing and learning representations of time-varying confounders through machine learning based on extended observable data as well as regional covariate data and unobserved confounders at the previous time step; Conditional density is calculated based on the confounding factors and observable data used for analysis at the current time step, the expected value of the human mobility index is estimated, an objective function incorporating the estimates is constructed, and average dose-response curves are plotted to assess the causal effects of successive weather treatment levels on human mobility.

2. The typhoon effect assessment method according to claim 1, characterized in that: The target typhoon data are analyzed and combined with covariates of residents in different geographical areas to capture unobserved time-varying confounding factors, including: Analyze the target typhoon data to capture basic observable data, select and quantify variables representing unobserved confounding factors based on the basic observable data to define them as proxy variables, and convert the observable data into feature vectors; wherein the basic observable data include typhoon weather data, resident alertness data, and resident basic flow pattern data; the proxy variables include resident alertness data volume and resident basic flow pattern data; the proxy variables are as described in formula (1): （1） In formula (1), t represents the time step, n Indicates the number of selected geographic regions; Conduct time series analysis on the basic mobility pattern data of residents to capture their time dependence; The mobility tensor of different travel modes is decomposed into a basic life mode tensor by formula (3); wherein the basic life mode tensor includes a regional intensity change mode tensor, a time period mode tensor and a day mode tensor; （3） In formula (3), represents the liquidity tensor, The intensity matrix representing the basic life patterns, coefficient matrix representing the temporal pattern; A time-space embedding is formed based on the region and time information of the residents' search for typhoon keywords, and the time-space embedding is input into the causal network structure to capture unobserved confounding factors; wherein the confounding factors are as described in formula (2): （2）。 3. The typhoon effect assessment method according to claim 2, characterized in that: Based on the extended observable data and the regional covariate data and unobserved confounders of the previous time step, machine learning is used to capture and learn the representation of time-varying confounders, including: Capturing the extended observable data; wherein the extended observable data includes extreme weather data , Human Mobility Index , mobile streaming network ; The mobile streaming network As described in formula (4): （4） In formula (4), R represents the set of real numbers; According to the extended observable data, causal effects are estimated by constructing a causal network structure and calculating the average treatment effect; Using a recurrent neural network to process historical covariate data through formula (8) to capture dynamic features in time series data, and using a graph convolutional network to process a mobile flow network to capture spatial dependencies, and combining the dynamic features in the time series data captured by the recurrent neural network and the spatial dependencies captured by the graph convolutional network through formula (9) to predict time-varying confounding factors; （8） In formula (8), Indicates that at time step t -1 hidden vector, capturing the time steps up to t -1 historical information; Indicates that at time step t The hidden vector of -2 is the initial state of the recurrent neural network; Indicates that at time step t -1 unobserved confounder; Indicates that at time step t -1 covariate; Indicates that at time step t -1 extreme weather data; Indicates that at time step t The potential result of the human mobility index of -1; ⊕ represents the concatenation of different vectors; （9） In formula (9), Indicates that at time step t unobserved confounders; Indicates that at time step t The mobile flow network matrix; ReLU Represents the rectified linear unit activation function, which is used to introduce nonlinearity and accelerate the learning of the model; and They respectively represent the two-layer model parameters of the graph convolutional network.

4. The typhoon effect assessment method according to claim 3, characterized in that: Based on the extended observable data, causal effect estimation is performed by constructing a causal network structure and calculating the average treatment effect, specifically including: According to the extended observable data, a causal network structure is constructed based on the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur; wherein the causal network structure is as described in formulas (5) and (6): （5） （6） In formulas (5) and (6), and Respectively represent the time step t potential outcomes of the human mobility index for typhoon occurrence and typhoon non-occurrence; Indicates that at time step t within, no. n potential results of the human mobility index for each region during a typhoon; Indicates that at time step t within, no. n potential results of the human mobility index for each region when a typhoon does not occur; The time step is calculated by formula (7): t The potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur are averaged across all regions to obtain the average treatment effect for causal effect estimation; （7） In formula (7), Indicates that at time step t , for the region i , potential results of the human mobility index during typhoons; Indicates that at time step t , for the region i , potential results of the human mobility index when the typhoon did not occur; Indicates that at time step t ,area i unobserved confounders; Indicates that at time step t ,area i extreme weather data; E(⋅) represents the expected value, which is used to estimate the potential results under given confounding factors and weather data conditions; Represents the time step t The average treatment effect.

5. The typhoon effect assessment method according to claim 1, characterized in that: The conditional density is calculated based on the confounding factors and observable data of the current time step, the expected value of the human mobility index is estimated, the objective function containing the estimate is constructed, and the average dose-response curve is plotted to evaluate the causal effect of the continuous weather treatment level on human mobility, including: Based on the learned confounding factors of the current time step and observable data for analysis, conditional density is calculated using probability density estimation; wherein the observable data for analysis includes typhoon impact data and human mobility data; Based on the learned confounding factors of the current time step and the human mobility index, the expected value of the human mobility index is calculated using the machine learning model; Constructing an objective function based on the conditional density and the result estimation; wherein the objective function is as described in formula (17); （17） In formula (17), represents the objective function, n represents the total number of observed samples; Indicates i Actual human mobility index for a sample of observations; Indicates i The predicted value of the human mobility index for a sample of observations; Indicates i Extreme weather data for observation samples; Indicates i Unobserved confounders of the observed sample; Indicates that given the unobserved confounder Under the conditions of extreme weather data The probability density of α is a regularization hyperparameter used to balance the weight of prediction error and conditional density estimation loss; The average dose-response curve is plotted by using the confounding factors and extreme weather data of the current time step obtained based on learning, and the causal effect of the continuous weather treatment level on the human mobility index is obtained; wherein the confounding factors and extreme weather data of the current time step are as described in formula (18); the average dose-response curve is as described in formula (19); （18） In formula (18), Indicates that at time step t , No. i The result of whether a typhoon occurs in the observation sample; Indicates that at time step t , No. i unobserved confounders learned from the observed sample; Indicates that at time step t , No. i Extreme weather data in observation samples; （19） In formula (19), E (⋅) indicates expected value; Indicates that at time step t Human Mobility Index; and Indicates that at time step t Other covariates or network structure data; Indicates that at time step t The level of weather treatment; A specific value representing a given weather treatment level.

6. The typhoon effect assessment method according to claim 5, characterized in that: Based on the confounding factors of the current time step obtained through learning and the observable data for analysis, the conditional density is calculated using probability density estimation, including: Apply the activation function to estimate the conditional density under given confounding factors at multiple grid points through formula (10) to obtain the probability distribution; （10） In formula (10), Confounding factors The eigenvector of The result is J +1-dimensional real vector; Indicates that given the confounding factors Under the conditions of J + Conditional density at 1 grid point; , The dimension is The weight matrix of Confounding factors Dimensions; The conditional density is adjusted by ensuring that the estimated conditional density is continuous for any given value at a number of grid points from 0 to 1 and satisfies the valid density conditions that all probability values ​​are non-negative and that the sum of the entire probability distribution is 1; The specific value of the conditional density at each grid point is calculated based on formula (11) by linear interpolation to obtain the detailed probability distribution of different weather treatment levels under given confounding factors; （11） In formula (11), Indicates the corresponding weather data processing level , in j The conditional density estimate at grid points; where j ∈{0, 1, ..., J }, J The total number of grids minus one; By performing linear interpolation based on formula (12) at predefined grid points, the conditional density of any specific weather treatment level is accurately estimated to obtain a function that continuously and accurately reflects the probability distribution of the corresponding level; （12） In formula (12), Indicates that given the confounding factors Under the conditions of weather treatment level The conditional density estimate of ; and Represents the grid points and The conditional density estimate at ; and , which means less than or equal to The largest integer sum greater than or equal to The smallest integer of ; Represents the total number of grid points minus one.

7. The typhoon effect assessment method according to claim 6, characterized in that: Based on the learned confounding factors of the current time step and the human mobility index, the expected value of the human mobility index is calculated using a machine learning model, including: The human mobility index is defined under given extreme weather data and confounding factors by formula (13) to obtain the expected value of the human mobility index; （13） In formula (13), represents the expected value of the human mobility index; The continuous extreme weather data and confounding factors are used as input to train the multilayer perceptron network, and the extreme weather data and confounding factors are processed by integrating the variable coefficient prediction head through formula (14) to obtain the expected value of the human mobility index; where, （14） In formula (14), Represents the parameters of the variable coefficient prediction head; Indicates that it has parameters Neural network function; represents the confounding factors as input vector The eigenvector of By applying the activation function through formula (15) and the spline odd function modeling through formula (16) in the multilayer perceptron network, nonlinear changes and continuity modeling of the input data are performed to obtain a predicted value of the human mobility index with accuracy and flexibility; （15） In formula (15), represents the degree of the truncated basis, Indicates i The parameter vector of neurons depends on extreme weather data ; Represents the parameter vector The transpose of the input vector Perform dot product operation; （16） In formula (16), represents the linear combination of spline basis functions, represents the spline basis, Indicates i Neurons and l The coefficients between the spline basis functions; Indicates l spline basis functions.

8. A typhoon effect assessment device, characterized in that: The device comprises: The typhoon effect analysis unit is used to analyze the target typhoon data and incorporate covariates of residents in different geographical areas to capture unobserved time-varying confounding factors; A time-varying confounding factor learning unit, used to capture and learn the representation of time-varying confounding factors through machine learning based on the extended observable data and the regional covariate data and unobserved confounding factors of the previous time step generated by the typhoon effect analysis unit; A causal effect evaluation unit is used to calculate the conditional density based on the confounders of the current time step generated by the time-varying confounder learning unit and the observable data for analysis, estimate the expected value of the human mobility index, construct an objective function containing the estimation, and draw an average dose-response curve to evaluate the causal effect of continuous weather treatment levels on human mobility.

9. The typhoon effect assessment device according to claim 8, characterized in that: The typhoon effect analysis unit comprises: A typhoon data capture module, used to analyze target typhoon data to capture basic observable data; wherein the basic observable data includes typhoon weather data, resident alertness data, and resident basic flow pattern data; The proxy variable definition module is used to select and quantify variables representing unobserved confounding factors based on the basic observable data captured by the typhoon data capture module and define them as proxy variables, and convert the observable data into feature vectors; wherein the proxy variables include the amount of residents' alertness data and the basic flow pattern data of residents; the proxy variables are as described in formula (1): （1） In formula (1), t represents the time step, n Indicates the number of selected geographic regions; A time series analysis module, used for performing time series analysis on the basic flow pattern data of residents captured by the typhoon data capture module to capture its time dependency; A non-negative matrix decomposition module is used to decompose the mobility tensors of different travel modes into basic life pattern tensors through formula (3); wherein the basic life pattern tensor includes a regional intensity change pattern tensor, a time period pattern tensor and a day pattern tensor; （3） In formula (3), represents the liquidity tensor, The intensity matrix representing the basic life patterns, coefficient matrix representing the temporal pattern; A time-space embedding generation module, used to form a time-space embedding according to the area and time information of residents searching for typhoon keywords captured by the typhoon data capture module; A causal network processing module is used to input the time-space embedding generated by the time-space embedding generation module into the causal network structure to capture unobserved confounding factors; wherein the confounding factors are as described in formula (2): （2）。 10. The typhoon effect assessment device according to claim 9, characterized in that: The time-varying confounding factor learning unit includes: An extended observable data capture module is used to capture the extended observable data; wherein the extended observable data includes extreme weather data , Human Mobility Index , mobile streaming network ; The mobile streaming network As described in formula (4): （4） In formula (4), R represents the set of real numbers; A causal network construction module, used to construct a causal network structure according to the extended observable data captured by the extended observable data capture module; A causal effect estimation module, used to calculate the average treatment effect through the causal network structure constructed by the causal network construction module, and perform causal effect estimation; A recurrent neural network processing module, used for processing historical covariate data using a recurrent neural network through formula (8) to capture dynamic features in time series data; （8） In formula (8), Indicates that at time step t -1 hidden vector, capturing the time steps up to t -1 historical information; Indicates that at time step t The hidden vector of -2 is the initial state of the recurrent neural network; Indicates that at time step t -1 unobserved confounder; Indicates that at time step t -1 covariate; Indicates that at time step t -1 extreme weather data; Indicates that at time step t The potential result of the human mobility index of -1; ⊕ represents the concatenation of different vectors; A graph convolutional network processing module, which is used to process the mobile flow network using a graph convolutional network to capture spatial dependencies; A time-varying confounding factor prediction module, used to predict time-varying confounding factors by combining the capture results of the recurrent neural network processing module and the graph convolution network processing module through formula (9); （9） In formula (9), Indicates that at time step t unobserved confounders; Indicates that at time step t The mobile flow network matrix; ReLU Represents the rectified linear unit activation function, which is used to introduce nonlinearity and accelerate the learning of the model; and They respectively represent the two-layer model parameters of the graph convolutional network.

11. The typhoon effect assessment device according to claim 10, characterized in that: The causal network construction module is used to construct a causal network structure based on the extended observable data and the potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur; wherein the causal network structure is as described in formulas (5) and (6): （5） （6） In formulas (5) and (6), and Respectively represent the time step t potential outcomes of the human mobility index for typhoon occurrence and typhoon non-occurrence; Indicates that at time step t within, no. n potential results of the human mobility index for each region during a typhoon; Indicates that at time step t within, no. n potential results of the human mobility index for each region when a typhoon does not occur; The causal effect estimation module is used to calculate the time step t The potential results of the human mobility index when a typhoon occurs and when a typhoon does not occur are averaged across all regions to obtain the average treatment effect for causal effect estimation; （7） In formula (7), Indicates that at time step t , for the region i , potential results of the human mobility index during typhoons; Indicates that at time step t , for the region i , potential results of the human mobility index when the typhoon did not occur; Indicates that at time step t ,area i unobserved confounders; Indicates that at time step t ,area i extreme weather data; E(⋅) represents the expected value, which is used to estimate the potential results under given confounding factors and weather data conditions; Represents the time step t The average treatment effect.

12. The typhoon effect assessment device according to claim 11, characterized in that: The causal effect assessment unit comprises: A conditional density calculation module, used to calculate the conditional density using probability density estimation based on the confounding factors of the current time step learned by the time-varying confounding factor prediction module and observable data for analysis; wherein the observable data for analysis include typhoon impact data and human mobility data; A human mobility index estimation module, configured to calculate an expected value of the human mobility index using a machine learning model based on the confounding factors of the current time step learned by the time-varying confounding factor prediction module and the human mobility index; An objective function construction module, used to construct an objective function based on the conditional density calculated by the conditional density calculation module and the causal effect estimation result obtained by the causal effect estimation module; wherein the objective function is as described in formula (17); （17） In formula (17), represents the objective function, n represents the total number of observed samples; Indicates i Actual human mobility index for a sample of observations; Indicates i The predicted value of the human mobility index for a sample of observations; Indicates i Extreme weather data for observation samples; Indicates i Unobserved confounders of the observed sample; Indicates that given the unobserved confounder Under the conditions of extreme weather data The probability density of α is a regularization hyperparameter used to balance the weight of prediction error and conditional density estimation loss; The average dose-response curve drawing module is used to draw the average dose-response curve using the confounding factors of the current time step learned by the time-varying confounding factor prediction module and the extreme weather data captured by the extended observable data capture module, so as to obtain the causal effect of the continuous weather treatment level on the human mobility index; wherein the confounding factors of the current time step and the extreme weather data are as described in formula (18); and the average dose-response curve is as described in formula (19); （18） In formula (18), Indicates that at time step t , No. i The result of whether a typhoon occurs in the observation sample; Indicates that at time step t , No. i unobserved confounders learned from the observed sample; Indicates that at time step t , No. i Extreme weather data in observation samples; （19） In formula (19), E (⋅) indicates expected value; Indicates that at time step t Human Mobility Index; and Indicates that at time step t Other covariates or network structure data; Indicates that at time step t The level of weather treatment; A specific value representing a given weather treatment level.

13. The typhoon effect assessment device according to claim 12, characterized in that: The conditional density calculation module is used for: Apply the activation function to estimate the conditional density under given confounding factors at multiple grid points through formula (10) to obtain the probability distribution; （10） In formula (10), Confounding factors The eigenvector of The result is J +1-dimensional real vector; Indicates that given the confounding factors Under the conditions of J + Conditional density at 1 grid point; , The dimension is The weight matrix of Confounding factors Dimensions; The conditional density is adjusted by ensuring that the estimated conditional density is continuous for any given value at a number of grid points from 0 to 1 and satisfies the valid density conditions that all probability values ​​are non-negative and that the sum of the entire probability distribution is 1; The specific value of the conditional density at each grid point is calculated based on formula (11) by linear interpolation to obtain the detailed probability distribution of different weather treatment levels under given confounding factors; （11） In formula (11), Indicates the corresponding weather data processing level , in j The conditional density estimate at grid points; where j ∈{0, 1, ..., J }, J The total number of grids minus one; By performing linear interpolation based on formula (12) at predefined grid points, the conditional density of any specific weather treatment level is accurately estimated to obtain a function that continuously and accurately reflects the probability distribution of the corresponding level; （12） In formula (12), Indicates that given the confounding factors Under the conditions of weather treatment level The conditional density estimate of ; and Represents the grid points and The conditional density estimate at ; and , which means less than or equal to The largest integer sum greater than or equal to The smallest integer of ; Represents the total number of grid points minus one.

14. The typhoon effect assessment device according to claim 13, characterized in that: The human mobility index estimation module is used to: The human mobility index is defined under given extreme weather data and confounding factors by formula (13) to obtain the expected value of the human mobility index; （13） In formula (13), represents the expected value of the human mobility index; The continuous extreme weather data and confounding factors are used as input to train the multilayer perceptron network, and the extreme weather data and confounding factors are processed by integrating the variable coefficient prediction head through formula (14) to obtain the expected value of the human mobility index; where, （14） In formula (14), Represents the parameters of the variable coefficient prediction head; Indicates that it has parameters Neural network function; represents the confounding factors as input vector The eigenvector of By applying the activation function through formula (15) and the spline odd function modeling through formula (16) in the multilayer perceptron network, nonlinear changes and continuity modeling of the input data are performed to obtain a predicted value of the human mobility index with accuracy and flexibility; （15） In formula (15), represents the degree of the truncated basis, Indicates i The parameter vector of neurons depends on extreme weather data ; Represents the parameter vector The transpose of the input vector Perform dot product operation; （16） In formula (16), represents the linear combination of spline basis functions, represents the spline basis, Indicates i Neurons and l The coefficients between the spline basis functions; Indicates l spline basis functions.

15. An electronic device, comprising: A processor and a memory, characterized in that the memory is used to store computer program codes, the computer program codes include computer instructions, and when the processor executes the computer instructions, the electronic device executes the steps of the typhoon effect assessment method as described in any one of claims 1 to 7.

16. A readable storage medium having a computer program stored therein, characterized in that: The computer program includes program instructions, and when the program instructions are executed by a processor of an electronic device, the processor executes the steps of the typhoon effect assessment method as described in any one of claims 1 to 7.

Images