Method for constructing prediction model of monthly generation frequency of low vortex over the qinghai-tibet plateau, and method and device for predicting low vortex over the qinghai-tibet plateau

Abstract

The application provides a Qinghai-Tibet Plateau low vortex monthly generation frequency prediction model construction method, a Qinghai-Tibet Plateau low vortex prediction method and device, and relates to the technical field of weather prediction. The model construction method comprises: dividing the whole year into multiple time periods, and dividing the generation area of the Qinghai-Tibet Plateau low vortex into multiple sub-regions to form a refined space-time grid. For each space-time unit, a candidate feature set containing nonlinear terms and interaction terms is constructed from basic meteorological factors. Then, starting from an empty model containing only an intercept term, a bidirectional stepwise regression algorithm is used to automatically select the most statistically significant and stable feature combination through iterative loops to form the optimal model structure for the time period and the sub-region. For each specific month, the corresponding model structure of the time period is called, and only the historical data of the month is used to re-calculate the regression to estimate the exclusive model parameters. The application realizes reliable prediction of the generation frequency of the Qinghai-Tibet Plateau low vortex on a monthly scale.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting technology, and in particular to a method for constructing a monthly frequency prediction model for low-pressure vortices in the Qinghai-Tibet Plateau, as well as a method and apparatus for predicting low-pressure vortices in the Qinghai-Tibet Plateau. Background Technology

[0002] Accurate prediction of the Tibetan Plateau low-pressure system is of crucial scientific value and practical significance for extreme precipitation forecasting, disaster risk assessment, and water resource allocation. Currently, the prediction of the Tibetan Plateau low-pressure system relies entirely on numerical weather prediction models.

[0003] However, the reliable forecast lead time of numerical weather prediction models is usually less than 15 days. For mesoscale systems like the Tibetan Plateau Low, once this lead time is exceeded, the forecast results will rapidly become distorted in key elements such as location, intensity, and formation time. Therefore, existing methods based on direct identification from numerical models cannot meet the operational requirements for monthly-scale climate prediction of the Tibetan Plateau Low, and there is an urgent need to explore and establish new long-term forecasting technologies. Summary of the Invention

[0004] This invention provides a method for constructing a monthly frequency prediction model for low-pressure vortices on the Tibetan Plateau, as well as a method and apparatus for predicting low-pressure vortices on the Tibetan Plateau. This addresses the shortcomings of existing technologies that cannot perform monthly-scale climate prediction of low-pressure vortices on the Tibetan Plateau, thereby enabling reliable prediction of the frequency of low-pressure vortices on the Tibetan Plateau on a monthly scale.

[0005] In a first aspect, the present invention provides a method for constructing a monthly generation frequency prediction model for low-level eddies in the Qinghai-Tibet Plateau, comprising: The 12 months are divided into multiple time periods, and the region where the Tibetan Plateau vortex forms is divided into multiple sub-regions; For each time period and each sub-region, the following steps are performed to construct the model structure corresponding to each time period and each sub-region: In the current time period and the current sub-region, the historical monthly generation frequency of the Tibetan Plateau low-pressure system is obtained as the dependent variable, and the historical monthly average values ​​of multiple candidate meteorological factors are obtained as basic independent variables; based on the basic independent variables, extended features are constructed, and the basic independent variables and the extended features together constitute a candidate feature set; a regression model containing only the intercept term is used as the starting point for iteration, and a bidirectional stepwise regression algorithm is used to iteratively filter the candidate feature set until the model structure converges, thus obtaining the model structure corresponding to the current time period and the current sub-region; For each month and each sub-region, the following steps are performed to obtain the monthly generation frequency prediction model corresponding to each month and each sub-region: the model structure corresponding to the current month's time period in the current sub-region is taken as the target model structure corresponding to the current month and the current sub-region; regression calculation is performed based on the independent and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; the target model structure and the model parameters corresponding to the target model structure are combined to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region.

[0006] In some embodiments, the stepwise bidirectional regression algorithm is used to iteratively filter the candidate feature set, including: Perform the following steps in each loop iteration: From the unselected candidate features in the candidate feature set, select and add a candidate feature that meets the feature addition criteria to the current model. The feature addition criteria include: in the univariate regression analysis of the candidate feature and the dependent variable, the p-value of the F-test is less than or equal to the first statistical significance threshold; after the candidate feature is added to the current model, the variance inflation factor of all features in the model is less than or equal to the multicollinearity threshold; and the adjusted R-squared value of the model... 2 The improvement is greater than the threshold for improving the explanatory power of the first model; From all the selected candidate features in the current model, remove one candidate feature that meets the feature removal criteria. These criteria include: the p-value of the candidate feature's F-test in the current model is greater than the second statistical significance threshold, and after removing the candidate feature from the current model, the adjusted R-squared value of the model decreases. 2 The improvement is greater than the threshold for improving the explanatory power of the second model.

[0007] In some embodiments, constructing extended features based on the basic independent variables includes: The extended feature is obtained by constructing nonlinear terms and pairwise interaction terms on the basic independent variables; wherein the nonlinear terms include quadratic terms and cubic terms.

[0008] In some embodiments, obtaining the historical monthly average values ​​of multiple candidate meteorological factors includes: Within the current time period and sub-region, obtain the historical monthly average values ​​of multiple meteorological factors; Based on the correlation coefficient between the historical monthly generation frequency and the historical monthly average value of each meteorological factor, and the multicollinearity among the historical monthly average values ​​of the multiple meteorological factors, the historical monthly average values ​​of the multiple meteorological factors are screened to obtain the historical monthly average values ​​of the candidate meteorological factors.

[0009] In some embodiments, the method further includes: The historical monthly average values ​​of each meteorological factor were preprocessed by positive transformation, standardization, and logarithmic transformation in sequence.

[0010] Secondly, the present invention also provides a method for predicting low-pressure vortices in the Qinghai-Tibet Plateau, comprising: Obtain the input feature data corresponding to the target month and the target sub-region. The input feature data is determined based on the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and the target sub-region. The input feature data is input into the monthly generation frequency prediction model, and the monthly generation frequency prediction value corresponding to the target month and the target sub-region is output. The monthly generation frequency prediction model is obtained based on the method for constructing the monthly generation frequency prediction model of the Tibetan Plateau low vortex as described in any of the first aspects.

[0011] Thirdly, the present invention also provides a device for constructing a prediction model for the monthly generation frequency of low-level vortices in the Qinghai-Tibet Plateau, comprising: The segmentation module is used to divide the 12 months into multiple time periods and to divide the region where the Tibetan Plateau low vortex is generated into multiple sub-regions; The construction module is used to perform the following steps for each time period and each sub-region to construct the model structure corresponding to each time period and each sub-region: In the current time period and the current sub-region, the historical monthly generation frequency of the Tibetan Plateau low vortex is obtained as the dependent variable, and the historical monthly average value of multiple candidate meteorological factors is obtained as the basic independent variable; based on the basic independent variable, extended features are constructed, and the basic independent variable and the extended features together constitute a candidate feature set; a regression model containing only the intercept term is used as the starting point for iteration, and a bidirectional stepwise regression algorithm is used to iteratively filter the candidate feature set until the model structure converges, thus obtaining the model structure corresponding to the current time period and the current sub-region; The first acquisition module is used to perform the following steps for each month and each sub-region to acquire the monthly generation frequency prediction model corresponding to each month and each sub-region: taking the model structure corresponding to the current month's time period in the current sub-region as the target model structure corresponding to the current month and the current sub-region; performing regression calculations based on the independent and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; and combining the target model structure and the model parameters corresponding to the target model structure to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region.

[0012] Fourthly, the present invention also provides a device for predicting low-level vortices on the Qinghai-Tibet Plateau, comprising: The second acquisition module is used to acquire input feature data corresponding to the target month and the target sub-region. The input feature data is determined based on the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and the target sub-region. The prediction module is used to input the input feature data into the monthly generation frequency prediction model and output the monthly generation frequency prediction value corresponding to the target month and the target sub-region. The monthly generation frequency prediction model is obtained based on the method for constructing the monthly generation frequency prediction model of the Tibetan Plateau low vortex as described in any of the first aspects.

[0013] Fifthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for constructing a monthly generation frequency prediction model of a low-pressure zone in the Tibetan Plateau as described above, or the method for predicting a low-pressure zone in the Tibetan Plateau as described above.

[0014] Sixthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a monthly generation frequency prediction model of low-pressure zones in the Tibetan Plateau as described above, or the method for predicting low-pressure zones in the Tibetan Plateau as described above.

[0015] In a seventh aspect, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for constructing a monthly generation frequency prediction model of a low-pressure zone in the Tibetan Plateau as described above, or the method for predicting a low-pressure zone in the Tibetan Plateau as described above.

[0016] The present invention provides a method for constructing a monthly frequency prediction model for low-pressure vortices in the Tibetan Plateau, as well as a method and apparatus for predicting low-pressure vortices in the Tibetan Plateau. By constructing model structures by region and time period, spatial heterogeneity and seasonal non-stationarity are explicitly addressed. Based on the model structures, the model parameters corresponding to each month are determined to obtain a variable-coefficient monthly frequency prediction model for low-pressure vortices in the Tibetan Plateau with a unified physical framework, thereby achieving reliable prediction of the frequency of low-pressure vortex formation in the Tibetan Plateau over 1-3 months. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a flowchart illustrating the method for constructing a monthly generation frequency prediction model for low-level vortices in the Qinghai-Tibet Plateau provided by the present invention.

[0019] Figure 2 This is a flowchart illustrating the method for predicting low-pressure vortexes in the Qinghai-Tibet Plateau provided by this invention.

[0020] Figure 3 This invention provides scatter plots of observed and fitted values ​​and residual QQ plots corresponding to the frequent and infrequent periods in a region.

[0021] Figure 4 This invention provides scatter plots of observed and fitted values ​​and residual QQ plots corresponding to the two-frequency and low-frequency periods in a region, respectively.

[0022] Figure 5 This is a schematic diagram of the device for constructing a monthly generation frequency prediction model for low-level vortices in the Qinghai-Tibet Plateau provided by the present invention.

[0023] Figure 6 This is a schematic diagram of the structure of the low-vortex prediction device for the Qinghai-Tibet Plateau provided by the present invention.

[0024] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] Figure 1 This is a flowchart illustrating the method for constructing a monthly generation frequency prediction model for low-level vortices in the Qinghai-Tibet Plateau provided by the present invention. Figure 1 As shown, the method includes the following steps: Step 110: Divide the 12 months into multiple time periods and divide the region where the Tibetan Plateau vortex forms into multiple sub-regions.

[0027] Specifically, the formation mechanism of the Tibetan Plateau low-pressure system may fundamentally differ across seasons. Therefore, from a temporal perspective, based on the monthly frequency threshold of the low-pressure system, the 12 months are divided into multiple periods. For example, using the multi-year average monthly formation frequency of 10 as a threshold, the year is divided into two periods: a high-frequency period and a low-frequency period. The high-frequency period covers April to October, when the Tibetan Plateau low-pressure system is most active, while the low-frequency period corresponds to November to March, when low-frequency activity is relatively rare.

[0028] The formation mechanism of the Tibetan Plateau low-pressure system may also vary geographically. Therefore, it is spatially divided into multiple sub-regions. For example, focusing on the main formation area of ​​the Tibetan Plateau low-pressure system, namely the range of 30°N to 36°N and 80°E to 100°E, and further dividing it along the 90°E line, we can form a western sub-region and an eastern sub-region.

[0029] By dividing the data into time and space, a solid foundation is laid for establishing dedicated forecasting models adapted to different seasons and regions.

[0030] Step 120: For each time period and each sub-region, perform the following steps to construct the model structure corresponding to each time period and each sub-region: In the current time period and the current sub-region, obtain the historical monthly generation frequency of the Tibetan Plateau low-pressure vortex as the dependent variable, and obtain the historical monthly average values ​​of multiple candidate meteorological factors as basic independent variables; construct extended features based on the basic independent variables, and combine the basic independent variables and extended features to form a candidate feature set; take a regression model containing only the intercept term as the starting point for iteration, and use a bidirectional stepwise regression algorithm to iteratively filter the candidate feature set until the model structure converges, thus obtaining the model structure corresponding to the current time period and the current sub-region.

[0031] Specifically, for a specific spatiotemporal combination within a given time period and a specific sub-region, a dedicated and optimal model structure is constructed for that specific spatiotemporal combination. Specifically, as follows: For the current time period and the current sub-region, the historical monthly formation frequency of the Tibetan Plateau low vortex is obtained as the dependent variable that needs to be predicted. At the same time, the historical monthly average values ​​of a series of candidate meteorological factors (such as vorticity, relative humidity, etc.) are collected as the basic independent variables (i.e., the initial prediction variables).

[0032] In order to capture the complex physical relationships in the real world, the construction of the model structure is not limited to these basic independent variables. Instead, the basic independent variables are used to construct features to obtain extended features, and the basic independent variables and extended features are used together as part of the candidate feature set.

[0033] It should be noted that after obtaining the candidate feature set, missing value processing can be performed on the candidate feature set to delete samples with missing values ​​for any variable, resulting in a clean candidate feature set without missing values.

[0034] To ensure objectivity and fairness, the screening process begins with a regression model that only includes the intercept term (e.g., an Ordinary Least Squares (OLS) model). A bidirectional stepwise regression algorithm of "forward selection and backward elimination" is used to iteratively screen the candidate feature set. This cyclical process of adding and removing features continues until the model structure reaches a convergent state, meaning that there are no more suitable features to add and no redundant features to remove.

[0035] After the iteration process terminates, a complete multivariate nonlinear regression equation is constructed. This equation can be clearly decomposed into two components: model structure and model parameters. The model structure consists of the intercept term, the final selected feature combinations, and their specific mathematical forms; the model parameters are the specific regression coefficients and intercept terms corresponding to the feature combinations in the model structure. The model structure corresponding to the current time period and the current sub-region is obtained.

[0036] Step 130: Perform the following steps for each month and each sub-region to obtain the monthly generation frequency prediction model for each month and each sub-region: Take the model structure corresponding to the current month's time period in the current sub-region as the target model structure for the current month and the current sub-region; Perform regression calculations based on the independent and dependent variables defined by the target model structure for the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; Combine the target model structure with the model parameters corresponding to the target model structure to obtain the monthly generation frequency prediction model for the current month and the current sub-region.

[0037] Specifically, based on the model structure corresponding to each time period and each sub-region, the model parameters are re-estimated for each month, forming a monthly model and a monthly parameter table. Specifically, as follows: For the current month and the current sub-region, determine the model structure corresponding to the time period of the current month within the current sub-region, and use this model structure as the target model structure for the current month and the current sub-region. Obtain observational data for the current month and the current sub-region, including the dependent variable (monthly formation frequency of the Tibetan Plateau low-pressure system) and the independent variables defined by the target model structure. Based on this observational data, perform a new regression calculation to obtain a set of model parameters (i.e., the specific values ​​of each regression coefficient and intercept term) that best fits the current month and the current sub-region. Finally, combine the target model structure with this set of model parameters to obtain the monthly formation frequency prediction model for the current month and the current sub-region.

[0038] By defining the model structure using a large sample of data spanning multiple months, the risk of selecting spurious variables due to insufficient sample size or random fluctuations when using only single-month data is effectively avoided, ensuring that the model captures long-term and robust physical laws. Recalibrating the model parameters using dedicated data for each month allows the model to sensitively capture subtle differences in meteorological conditions between months. This monthly adaptive calibration ensures that the final forecasting model has both a solid theoretical foundation and closely reflects the actual conditions of each month. This not only produces a series of high-precision monthly forecasting tools but also a table showing the changes in model parameters for each month, providing profound insights for scientific research and clearly revealing the seasonal evolution of the intensity of key influencing factors.

[0039] The method for constructing a monthly frequency prediction model for low-pressure vortices on the Tibetan Plateau provided by this invention constructs model structures by region and time period, explicitly handling spatial heterogeneity and seasonal non-stationarity; then, based on the model structure, the model parameters corresponding to each month are determined, resulting in a variable-coefficient monthly frequency prediction model for low-pressure vortices on the Tibetan Plateau with a unified physical framework, thereby achieving reliable prediction of the frequency of low-pressure vortices on the Tibetan Plateau over 1-3 months.

[0040] In some embodiments, a bidirectional stepwise regression algorithm is used to iteratively filter the candidate feature set, including: Perform the following steps in each loop iteration: From the unselected candidate features in the candidate feature set, select and add a candidate feature that meets the feature addition criteria to the current model. These criteria include: in the univariate regression analysis of the candidate feature and the dependent variable, the p-value of the F-test is less than or equal to the first statistical significance threshold; after the candidate feature is added to the current model, the variance inflation factor of all features in the model is less than or equal to the multicollinearity threshold; and the adjusted R-squared value of the model is... 2 The improvement is greater than the threshold for improving the explanatory power of the first model; From all the selected candidate features in the current model, remove one candidate feature that meets the feature removal criteria. These criteria include: the candidate feature's p-value in the F-test of the current model is greater than the second statistical significance threshold, and after removing the candidate feature from the current model, the adjusted R-squared value of the model decreases. 2 The improvement is greater than the threshold for improving the explanatory power of the second model.

[0041] Specifically, in the forward selection phase, the algorithm iterates through all candidate features not yet included in the model and evaluates their value for inclusion. For a candidate feature to be included in the current model, it must meet certain inclusion criteria. Specifically: First, it must possess preliminary statistical relevance; that is, when the candidate feature is used alone to regress the dependent variable, its F-test p-value must be less than or equal to the first statistical significance threshold (e.g., 0.25). This ensures that only features with potential explanatory power are eligible for the next round of evaluation. Second, its inclusion must not disrupt the structural stability of the model. Specifically, after the candidate feature is included in the current model, the variance inflation factor (VIF) of all features (i.e., independent variables) in the model must be less than or equal to the multicollinearity threshold (e.g., 20.0). This aims to avoid distortion in model parameter estimation caused by high correlation between predictor variables. Finally, the inclusion of the feature must substantially contribute to the overall performance of the model, quantified by increasing the adjusted coefficient of determination (R-squared, R0). 2 This produces a gain greater than the first model's explanatory power enhancement threshold (e.g., 0.001). Adjusted R 2 The proportion of the dependent variable that the characterization model can explain.

[0042] Following this is the backward elimination phase, which aims to examine and remove features already existing in the model that may become redundant or insignificant due to the addition of new features. For a selected candidate feature to be removed, it must meet certain removal criteria. Specifically: First, the significance of the candidate feature has significantly decreased in the current multivariate model environment, as evidenced by its p-value in the F-test of the current model being greater than the second statistical significance threshold (e.g., 0.30). Second, removing the candidate feature will benefit the model's simplicity and generalization ability; this is judged by the adjusted R-squared value of the model after removing the feature. 2 On the contrary, it can achieve growth exceeding the second model explanatory power improvement threshold (e.g., 0.001). The benefit of reducing model complexity by removing a noisy variable with a small contribution outweighs the cost of its small information loss, thus improving the adjusted R. 2 .

[0043] The entire screening process will continue in this iterative process of combining advancement and retreat until the algorithm can neither add any new features nor remove any old features in a complete loop. At this point, the model structure reaches a convergent state, and the screening process terminates.

[0044] In some embodiments, constructing extended features based on basic independent variables includes: Extended features are obtained by constructing nonlinear terms and pairwise interaction terms for the basic independent variables; among them, nonlinear terms include quadratic and cubic terms.

[0045] Specifically, to capture the complex relationships that are prevalent in natural phenomena, feature engineering methods are used to extend the basic independent variables to enhance the model's expressive power. This process mainly involves the construction of two types of extended features. The first type is nonlinear terms, that is, calculating the square of each basic independent variable Xi (Xi²). 2 ) and the third power (Xi 3 The inclusion of these higher-order terms aims to enable the model to fit curvilinear relationships, thereby accurately characterizing physical processes whose effects are not constant but may accelerate, saturate, or even reverse as the independent variables increase. The second type consists of pairwise interaction terms, calculated by multiplying any two distinct fundamental independent variables Xi and Xj (Xi...). Xj) is used to generate this feature. The role of this type of feature is to quantify the synergistic or restrictive effects between variables, that is, the combined effect when two factors coexist may be significantly greater or less than the simple sum of their independent effects, thus revealing a deeper intrinsic driving mechanism.

[0046] In some embodiments, obtaining the historical monthly average of multiple candidate meteorological factors includes: Within the current time period and sub-region, obtain the historical monthly average values ​​of multiple meteorological factors; Based on the correlation coefficient between the historical monthly generation frequency and the historical monthly average of each meteorological factor, as well as the multicollinearity among the historical monthly averages of multiple meteorological factors, the historical monthly averages of multiple meteorological factors are screened to obtain the historical monthly averages of candidate meteorological factors.

[0047] Specifically, to ensure the quality of the model input, the original meteorological factor data is first rigorously screened to construct a set of candidate meteorological factors that is both representative and free of redundancy.

[0048] Within a specified time period and sub-region, the historical monthly average values ​​of all potentially relevant meteorological factors are first obtained. These factors then undergo a dual filtering process based on correlation analysis and multicollinearity diagnosis. The first filtering aims to assess the correlation coefficient between each meteorological factor and the dependent variable (i.e., historical monthly generation frequency), serving as a preliminary measure of its predictive potential. The second filtering focuses on the internal correlations among meteorological factors, identifying and addressing multicollinearity issues by calculating the correlation matrix between meteorological factors.

[0049] When the absolute value of the correlation coefficient between any two meteorological factors exceeds the threshold of 0.90, one of the meteorological factors is deemed redundant and removed. In deciding which meteorological factor to retain, the system follows an optimization rule, prioritizing those with higher correlation to the dependent variable or lower initial variance inflation factors. This preprocessing step aims to reduce collinearity among variables at the source and sets strict constraints for subsequent stepwise regression analysis. Specifically, during model building, the VIF value of all selected meteorological factors will be continuously monitored and ensured to not exceed 20.0, thereby guaranteeing the stability and reliability of the final model.

[0050] In some embodiments, the method for constructing a monthly generation frequency prediction model for low-level eddies in the Tibetan Plateau provided by the present invention further includes: The historical monthly average values ​​of each meteorological factor were preprocessed by positive transformation, standardization, and logarithmic transformation respectively.

[0051] Specifically, to enhance the stability and convergence of the subsequent model, a rigorous and orderly data preprocessing procedure was implemented on the historical monthly average values ​​of all original meteorological factors before model construction.

[0052] The process begins with positiveification and standardization. For meteorological factors with negative or zero values, methods such as translation are used to ensure their numerical domain is strictly positive, a necessary prerequisite for subsequent logarithmic transformation. Next, all factors are standardized to eliminate scale differences caused by different physical units, ensuring comparability of factors under the same standard. Based on this, the final step is to perform a logarithmic transformation on the processed data. This transformation effectively compresses the dynamic range of the data, reduces the excessive influence of extreme values ​​on the model, and helps to make the data distribution more symmetrical and normal, thus better satisfying the statistical assumptions of regression analysis. This series of preprocessing steps ensures that all independent variables input into the model are within an optimized and well-suited numerical space.

[0053] Figure 2 This is a flowchart illustrating the method for predicting low-pressure vortices in the Qinghai-Tibet Plateau provided by this invention, as shown below. Figure 2 As shown, the method includes: Step 210: Obtain the input feature data corresponding to the target month and the target sub-region. The input feature data is determined based on the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and the target sub-region.

[0054] Step 220: Input the input feature data into the monthly generation frequency prediction model corresponding to the target month and the target sub-region, and output the monthly generation frequency prediction value corresponding to the target month and the target sub-region.

[0055] The monthly generation frequency prediction model is based on the above-mentioned method for constructing the monthly generation frequency prediction model of the low-pressure zone in the Qinghai-Tibet Plateau.

[0056] Specifically, in the model inference phase, the prediction request is first clearly defined, i.e., the target month and target sub-region are determined. Then, the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and target sub-region are determined, and the predicted values ​​of these specific independent variables in the target month and target sub-region are obtained to obtain the input feature data.

[0057] The prepared input feature data is imported into the monthly generation frequency prediction model corresponding to the target month and the target sub-region. This model encapsulates the optimized independent variable structure, the coefficients of each independent variable structure, and the specific values ​​of the intercept term for this spatiotemporal background. The model applies these preset parameters to the input feature data, performs regression calculations, and outputs the predicted frequency of the Tibetan Plateau low-pressure vortex in the target sub-region during the target month.

[0058] For example, to predict the frequency of low-pressure vortices over the Tibetan Plateau in the eastern sub-region in June 2026, the system first loads a dedicated prediction model for the eastern region for June, identifying the required combination of input features, such as the quadratic term of relative humidity and the interaction term between vorticity and geopotential height. Next, the system collects relevant meteorological forecast data for June 2026, constructs this feature vector, and inputs it into the dedicated prediction model. The model can then instantly calculate and output the predicted frequency of low-pressure vortices over the Tibetan Plateau for that month.

[0059] To comprehensively evaluate the effectiveness and practicality of the model, this invention constructs a monthly low-vortex formation frequency prediction model for the Tibetan Plateau through a refined modeling strategy based on different regions and time periods, combined with adaptive calibration of monthly coefficients. Performance validation results show that the model performs excellently on several key indicators and demonstrates outstanding reconstruction and interpretation capabilities for historical data. Specific effects are described below: Firstly, regarding the effectiveness of regional and time-segmented modeling, this invention designates the western region (30-36°N, 80-90°E) as Region 1 and the eastern region (30-36°N, 90-100°E) as Region 2. Based on whether the monthly average number of occurrences reaches 10, the year is divided into a high-incidence period (April-October) and a low-incidence period (November-March). The nonlinear model, constructed based on seven core meteorological variables—vorticity, vertical velocity, divergence, atmospheric apparent heat source vertical gradient parameters, specific humidity, relative humidity, and geopotential height—is evaluated using the Pearson correlation coefficient r and the coefficient of determination R. 2 The F-statistic and its corresponding p-value are comprehensively evaluated. Compared with a unified annual model for the entire region without spatiotemporal division, its coefficient of determination R0... 2 Compared to 0.173, the explanatory power of the model in this invention is significantly improved.

[0060] As shown in Table 1, the hierarchical model proposed in this invention significantly improves the explanatory power of the model in all four scenarios (frequent occurrence period in Region 1, low occurrence period in Region 1, frequent occurrence period in Region 2, and low occurrence period in Region 2). The coefficient of determination R for the model in each of the four scenarios is also shown. 2 The corresponding Pearson correlation coefficients (r) between the predicted and observed values ​​were 0.546, 0.710, 0.635, and 0.664, respectively, with the corresponding values ​​also ranging from 0.74 to 0.86. The p-values ​​for the F-tests of all models were significantly lower than the 0.05 significance level, indicating that their regression relationships were statistically significant.

[0061] Table 1

[0062] Secondly, regarding the effectiveness of monthly adaptive calibration of coefficients, this method, while maintaining the set of physical variables determined by the time-period model, re-estimates the specific values ​​of the model coefficients and intercept term for each month. The calibration results are shown in Table 2, demonstrating that the model accuracy remains highly stable at a finer monthly scale. For Region 1, the average coefficient of determination R0 for each month of its frequent occurrence period is [value missing]. 2 The average R value during the low incidence period was 0.54. 2 The value is 0.45; for region two, the average R value during the frequent and infrequent periods is... 2 The correlation coefficients reached 0.41 and 0.40 respectively. The average correlation coefficients (r) for each time period also remained at or above 0.63. This fully demonstrates the success of the monthly calibration strategy, which effectively improves the model's ability to capture subtle features of each month without sacrificing the overall stability of the model.

[0063] Table 2

[0064] Figure 3 This invention provides scatter plots of observed and fitted values ​​and residual QQ plots corresponding to frequent and infrequent occurrence periods in a given region. Figure 4 This invention provides scatter plots of observed and fitted values, and residual QQ plots corresponding to the high-frequency and low-frequency periods of regional dual-occurrence. The scatter plot of observed and fitted values ​​is used to visually evaluate the predictive accuracy of the model. In the scatter plot, each black dot represents a sample, with the horizontal axis representing the actual observed value of that sample and the vertical axis representing the fitted value of the model. The dashed line is a reference line, representing a perfect prediction where the fitted value equals the actual observed value. The residual QQ plot is used to verify whether the model's residuals conform to a normal distribution. In the residual QQ plot, each black dot represents the distribution of each actual residual value in the model, and the dashed line represents the ideal normal distribution.

[0065] from Figure 3 and Figure 4Scatter plot analysis of the observed and fitted values ​​shows that they are closely clustered near the diagonal in the coordinate system, exhibiting a high degree of linear correlation. The model's coefficient of determination R0... 2 The value can reach as high as 0.710, which indicates that the model has an excellent fitting effect on the frequency of low-pressure vortex formation on the Tibetan Plateau. Figure 3 and Figure 4 The residual QQ plot shows that the residual scatter points of the model basically follow the normal distribution reference line, and no systematic bias is observed, which proves that the statistical properties of the model residuals are good.

[0066] In summary, the excellent fitting accuracy and robust statistical properties demonstrate that the regional and time-specific models constructed in this invention are stable and reliable. This model successfully captures the spatial heterogeneity and seasonal variation patterns of the low-pressure system formation on the Tibetan Plateau, providing a solid scientific foundation for accurate monthly-scale predictions.

[0067] Table 3

[0068] It should be noted that in Table 3, r represents the historical monthly average of relative humidity at 500 hPa, w represents the historical monthly average of vertical velocity at 500 hPa, Q represents the historical average of the atmospheric apparent heat source vertical gradient from 450 hPa to 550 hPa, d represents the historical monthly average of divergence at 500 hPa, vo represents the historical monthly average of vorticity at 500 hPa, z represents the historical monthly average of geopotential height at 500 hPa, and q represents the historical monthly average of specific humidity at 500 hPa.

[0069] As shown in Table 3, vorticity and divergence are key variables in both regional and time-period models, reflecting their widespread influence on the formation of low-pressure systems on the Tibetan Plateau. Further analysis reveals significant regional differences in the meteorological factors affecting the frequency of low-pressure system formation on the Tibetan Plateau: during the frequent occurrence period in Region 1, relative humidity, vertical velocity, and the vertical gradient of the atmospheric apparent heat source are the dominant factors; while during the frequent occurrence period in Region 2, vorticity, divergence, and relative humidity contribute more significantly. This regional difference in dynamic and thermal influences not only highly matches the actual meteorological characteristics of the Tibetan Plateau but also strongly verifies the scientific validity and rationality of the regional and time-period modeling strategy adopted in this invention.

[0070] Therefore, this invention, by modeling by region and time period, constructing nonlinear features to expand features, and filtering features through bidirectional stepwise regression, can accurately capture the physical mechanism of low-pressure vortex formation in different regions and seasons of the Tibetan Plateau, thereby significantly improving the accuracy and reliability of monthly-scale prediction.

[0071] Figure 5 This is a schematic diagram of the structure of the device for constructing a monthly generation frequency prediction model for low-level eddies in the Qinghai-Tibet Plateau provided by the present invention, as shown below. Figure 5 As shown, the device includes: The division module 510 is used to divide the 12 months into multiple time periods and divide the region where the Tibetan Plateau vortex is generated into multiple sub-regions. The construction module 520 is used to perform the following steps for each time period and each sub-region to construct the model structure corresponding to each time period and each sub-region: In the current time period and the current sub-region, the historical monthly generation frequency of the Tibetan Plateau low vortex is obtained as the dependent variable, and the historical monthly average value of multiple candidate meteorological factors is obtained as the basic independent variable; based on the basic independent variable, extended features are constructed, and the basic independent variable and the extended features together constitute a candidate feature set; a regression model containing only the intercept term is used as the starting point for iteration, and a bidirectional stepwise regression algorithm is used to iteratively filter the candidate feature set until the model structure converges, thus obtaining the model structure corresponding to the current time period and the current sub-region; The first acquisition module 530 is used to perform the following steps for each month and each sub-region to acquire the monthly generation frequency prediction model corresponding to each month and each sub-region: taking the model structure corresponding to the current month's time period in the current sub-region as the target model structure corresponding to the current month and the current sub-region; performing regression calculations based on the independent and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; and combining the target model structure and the model parameters corresponding to the target model structure to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region.

[0072] In some embodiments, when the construction module 520 performs iterative filtering of the candidate feature set using a bidirectional stepwise regression algorithm, it is configured to: Perform the following steps in each loop iteration: From the unselected candidate features in the candidate feature set, select and add a candidate feature that meets the feature addition conditions to the current model; the feature addition conditions include: the candidate feature itself meets the first statistical significance threshold, and after the candidate feature is added to the model, the current model meets the multicollinearity threshold and the first model explanatory power improvement threshold; From all the selected candidate features in the current model, remove one candidate feature that meets the feature removal criteria. The feature removal criteria include: the candidate feature no longer meets the second statistical significance threshold in the current model, and its removal would cause the current model to meet the second model explanatory power improvement threshold.

[0073] In some embodiments, when constructing extended features based on the basic independent variables, the construction module 520 is used to: The extended feature is obtained by constructing nonlinear terms and pairwise interaction terms on the basic independent variables; wherein the nonlinear terms include quadratic terms and cubic terms.

[0074] In some embodiments, when obtaining the historical monthly average values ​​of multiple candidate meteorological factors, the construction module 520 is used to: Within the current time period and sub-region, obtain the historical monthly average values ​​of multiple meteorological factors; Based on the correlation coefficient between the historical monthly generation frequency and the historical monthly average value of each meteorological factor, and the multicollinearity among the historical monthly average values ​​of the multiple meteorological factors, the historical monthly average values ​​of the multiple meteorological factors are screened to obtain the historical monthly average values ​​of the candidate meteorological factors.

[0075] In some embodiments, the apparatus further includes: The preprocessing module is used to perform positive transformation, standardization, and logarithmic transformation on the historical monthly average values ​​of each meteorological factor in sequence.

[0076] It should be noted that the above-mentioned Qinghai-Tibet Plateau low-vortex monthly generation frequency prediction model construction device provided by the present invention can realize all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0077] Figure 6 This is a schematic diagram of the structure of the Tibetan Plateau low-vortex prediction device provided by the present invention, as shown below. Figure 6 As shown, the device includes: The second acquisition module 610 is used to acquire input feature data corresponding to the target month and the target sub-region. The input feature data is determined based on the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and the target sub-region. Prediction module 620 is used to input the input feature data into the monthly generation frequency prediction model and output the monthly generation frequency prediction value corresponding to the target month and the target sub-region; The monthly generation frequency prediction model is obtained based on the method for constructing the monthly generation frequency prediction model of the Tibetan Plateau low vortex described above.

[0078] It should be noted that the Tibetan Plateau low-vortex prediction device provided by the present invention can realize all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0079] Figure 7This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740. The processor 710, communication interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a method for constructing a prediction model for the monthly formation frequency of the Tibetan Plateau low-pressure system. This method includes: dividing the 12 months into multiple time periods and dividing the formation area of ​​the Tibetan Plateau low-pressure system into multiple sub-regions; performing the following steps for each time period and each sub-region to construct a model structure corresponding to each time period and each sub-region: within the current time period and current sub-region, obtaining the historical monthly formation frequency of the Tibetan Plateau low-pressure system as the dependent variable, and obtaining the historical monthly average values ​​of multiple candidate meteorological factors as basic independent variables; constructing extended features based on the basic independent variables, and combining the basic independent variables and the extended features to form a candidate feature set; using a regression model containing only the intercept term as the starting point for iteration, and employing a bidirectional... The stepwise regression algorithm iteratively filters the candidate feature set until the model structure converges, obtaining the model structure corresponding to the current time period and the current sub-region. For each month and each sub-region, the following steps are performed to obtain the monthly generation frequency prediction model for each month and each sub-region: the model structure corresponding to the current month's time period in the current sub-region is used as the target model structure for the current month and the current sub-region; regression calculations are performed based on the independent and dependent variables defined by the target model structure for the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; the target model structure and the model parameters corresponding to the target model structure are combined to obtain the monthly generation frequency prediction model for the current month and the current sub-region.

[0080] Alternatively, the processor 710 may call logic instructions in the memory 730 to execute a method for predicting low-pressure vortices in the Tibetan Plateau. This method includes: acquiring input feature data corresponding to a target month and a target sub-region, wherein the input feature data is determined based on independent variables defined by a monthly generation frequency prediction model corresponding to the target month and the target sub-region; inputting the input feature data into the monthly generation frequency prediction model; and outputting the monthly generation frequency prediction value corresponding to the target month and the target sub-region; wherein the monthly generation frequency prediction model is obtained based on the method for constructing a monthly generation frequency prediction model for low-pressure vortices in the Tibetan Plateau described above.

[0081] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for constructing a monthly frequency prediction model for the Tibetan Plateau low-pressure system provided by the above methods. This method includes: dividing the 12 months into multiple time periods and dividing the region of formation of the Tibetan Plateau low-pressure system into multiple sub-regions; performing the following steps for each time period and each sub-region to construct a model structure corresponding to each time period and each sub-region: within the current time period and the current sub-region, obtaining the historical monthly frequency of formation of the Tibetan Plateau low-pressure system as a dependent variable, and obtaining the historical monthly average values ​​of multiple candidate meteorological factors as basic independent variables; constructing extended features based on the basic independent variables, and combining the basic independent variables and the extended features to form a model structure. Candidate feature set; using a regression model containing only the intercept term as the starting point for iteration, and employing a bidirectional stepwise regression algorithm to iteratively filter the candidate feature set until the model structure converges, obtaining the model structure corresponding to the current time period and the current sub-region; for each month and each sub-region, perform the following steps to obtain the monthly generation frequency prediction model corresponding to each month and each sub-region: take the model structure corresponding to the current month's time period in the current sub-region as the target model structure corresponding to the current month and the current sub-region; perform regression calculations based on the independent and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; combine the target model structure and the model parameters corresponding to the target model structure to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region.

[0083] Alternatively, when the computer program is executed by the processor, the computer can execute a method for predicting low-pressure vortices in the Tibetan Plateau, the method comprising: acquiring input feature data corresponding to a target month and a target sub-region, the input feature data being determined based on independent variables defined in a monthly generation frequency prediction model corresponding to the target month and the target sub-region; inputting the input feature data into the monthly generation frequency prediction model, and outputting the monthly generation frequency prediction value corresponding to the target month and the target sub-region; wherein the monthly generation frequency prediction model is obtained based on the method for constructing a monthly generation frequency prediction model for low-pressure vortices in the Tibetan Plateau described in any of the above embodiments.

[0084] Furthermore, this invention also provides a non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for constructing a monthly frequency prediction model for low-pressure systems on the Tibetan Plateau provided by the aforementioned methods. This method includes: dividing the 12 months into multiple time periods and dividing the region where low-pressure systems on the Tibetan Plateau originate into multiple sub-regions; performing the following steps for each time period and each sub-region to construct a model structure corresponding to each time period and each sub-region: within the current time period and current sub-region, obtaining the historical monthly frequency of low-pressure systems on the Tibetan Plateau as the dependent variable, and obtaining the historical monthly average values ​​of multiple candidate meteorological factors as basic independent variables; constructing extended features based on the basic independent variables, and combining the basic independent variables and the extended features to form a candidate feature set; using a model containing only an intercept term... The regression model is used as the starting point for iteration, and a bidirectional stepwise regression algorithm is used to iteratively filter the candidate feature set until the model structure converges, thus obtaining the model structure corresponding to the current time period and the current sub-region. For each month and each sub-region, the following steps are performed to obtain the monthly generation frequency prediction model corresponding to each month and each sub-region: the model structure corresponding to the current month's time period in the current sub-region is used as the target model structure corresponding to the current month and the current sub-region; regression calculation is performed based on the independent variables and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; the target model structure and the model parameters corresponding to the target model structure are combined to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region.

[0085] Alternatively, when the computer program is executed by a processor, it implements a method for predicting low-pressure vortices in the Tibetan Plateau. This method includes: acquiring input feature data corresponding to a target month and a target sub-region, wherein the input feature data is determined based on independent variables defined by a monthly generation frequency prediction model corresponding to the target month and the target sub-region; inputting the input feature data into the monthly generation frequency prediction model; and outputting the monthly generation frequency prediction value corresponding to the target month and the target sub-region; wherein the monthly generation frequency prediction model is obtained based on the method for constructing a monthly generation frequency prediction model for low-pressure vortices in the Tibetan Plateau described in any of the above embodiments.

[0086] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0087] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a monthly generation frequency prediction model for low-level vortices in the Qinghai-Tibet Plateau, characterized in that, include: The 12 months are divided into multiple time periods, and the region where the Tibetan Plateau vortex forms is divided into multiple sub-regions; For each time period and each sub-region, the following steps are performed to construct the model structure corresponding to each time period and each sub-region: In the current time period and the current sub-region, the historical monthly generation frequency of the Tibetan Plateau low vortex is obtained as the dependent variable, and the historical monthly average value of multiple candidate meteorological factors is obtained as the basic independent variable; based on the basic independent variable, extended features are constructed, and the basic independent variable and the extended features together constitute a candidate feature set; Using a regression model containing only the intercept term as the starting point of the iteration, and employing a bidirectional stepwise regression algorithm to iteratively filter the candidate feature set until the model structure converges, the model structure corresponding to the current time period and the current sub-region is obtained. For each month and each sub-region, perform the following steps to obtain the monthly generation frequency prediction model for each month and each sub-region: take the model structure corresponding to the current month's time period in the current sub-region as the target model structure for the current month and the current sub-region; Regression calculations are performed based on the independent and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; the target model structure and the model parameters corresponding to the target model structure are combined to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region. The construction of extended features based on the basic independent variables includes: The extended feature is obtained by constructing nonlinear terms and pairwise interaction terms on the basic independent variables; wherein the nonlinear terms include quadratic terms and cubic terms.

2. The method according to claim 1, characterized in that, The stepwise bidirectional regression algorithm is used to iteratively filter the candidate feature set, including: Perform the following steps in each loop iteration: From the unselected candidate features in the candidate feature set, select and add a candidate feature that meets the feature addition criteria to the current model. The feature addition criteria include: in the univariate regression analysis of the candidate feature and the dependent variable, the p-value of the F-test is less than or equal to the first statistical significance threshold; after the candidate feature is added to the current model, the variance inflation factor of all features in the model is less than or equal to the multicollinearity threshold; and the adjusted R-squared value of the model... 2 The improvement is greater than the threshold for improving the explanatory power of the first model; From all the selected candidate features in the current model, remove one candidate feature that meets the feature removal criteria. These criteria include: the p-value of the candidate feature's F-test in the current model is greater than the second statistical significance threshold, and after removing the candidate feature from the current model, the adjusted R-squared value of the model decreases. 2 The improvement is greater than the threshold for improving the explanatory power of the second model.

3. The method according to claim 1, characterized in that, The process of obtaining the historical monthly average values ​​of multiple candidate meteorological factors includes: Within the current time period and sub-region, obtain the historical monthly average values ​​of multiple meteorological factors; Based on the correlation coefficient between the historical monthly generation frequency and the historical monthly average value of each meteorological factor, and the multicollinearity among the historical monthly average values ​​of the multiple meteorological factors, the historical monthly average values ​​of the multiple meteorological factors are screened to obtain the historical monthly average values ​​of the candidate meteorological factors.

4. The method according to claim 3, characterized in that, The method further includes: The historical monthly average values ​​of each meteorological factor were preprocessed by positive transformation, standardization, and logarithmic transformation in sequence.

5. A method for predicting low-level vortices in the Qinghai-Tibet Plateau, characterized in that, include: Obtain the input feature data corresponding to the target month and the target sub-region. The input feature data is determined based on the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and the target sub-region. The input feature data is input into the monthly generation frequency prediction model, and the monthly generation frequency prediction value corresponding to the target month and the target sub-region is output. The monthly generation frequency prediction model is obtained based on the method for constructing the monthly generation frequency prediction model of the low-pressure zone in the Tibetan Plateau as described in any one of claims 1-4.

6. A device for constructing a predictive model for the monthly generation frequency of low-level vortices in the Qinghai-Tibet Plateau, characterized in that, include: The segmentation module is used to divide the 12 months into multiple time periods and to divide the region where the Tibetan Plateau low vortex is generated into multiple sub-regions; The construction module is used to perform the following steps for each time period and each sub-region to construct the model structure corresponding to each time period and each sub-region: In the current time period and the current sub-region, the historical monthly generation frequency of the Tibetan Plateau low vortex is obtained as the dependent variable, and the historical monthly average value of multiple candidate meteorological factors is obtained as the basic independent variable; based on the basic independent variable, extended features are constructed, and the basic independent variable and the extended features together constitute a candidate feature set; Using a regression model containing only the intercept term as the starting point of the iteration, and employing a bidirectional stepwise regression algorithm to iteratively filter the candidate feature set until the model structure converges, the model structure corresponding to the current time period and the current sub-region is obtained. The first acquisition module is used to perform the following steps for each month and each sub-region to obtain the monthly generation frequency prediction model corresponding to each month and each sub-region: take the model structure corresponding to the current month's time period in the current sub-region as the target model structure corresponding to the current month and the current sub-region. Regression calculations are performed based on the independent and dependent variables defined by the target model structure corresponding to the current month and the current sub-region to obtain the model parameters corresponding to the target model structure; the target model structure and the model parameters corresponding to the target model structure are combined to obtain the monthly generation frequency prediction model corresponding to the current month and the current sub-region. The construction of extended features based on the basic independent variables includes: The extended feature is obtained by constructing nonlinear terms and pairwise interaction terms on the basic independent variables; wherein the nonlinear terms include quadratic terms and cubic terms.

7. A device for predicting low-level vortices on the Qinghai-Tibet Plateau, characterized in that, include: The second acquisition module is used to acquire input feature data corresponding to the target month and the target sub-region. The input feature data is determined based on the independent variables defined by the monthly generation frequency prediction model corresponding to the target month and the target sub-region. The prediction module is used to input the input feature data into the monthly generation frequency prediction model and output the monthly generation frequency prediction value corresponding to the target month and the target sub-region. The monthly generation frequency prediction model is obtained based on the method for constructing the monthly generation frequency prediction model of the low-pressure zone in the Tibetan Plateau as described in any one of claims 1-4.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for constructing a monthly generation frequency prediction model for low-pressure vortices in the Tibetan Plateau as described in any one of claims 1 to 4, or the method for predicting low-pressure vortices in the Tibetan Plateau as described in claim 5.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for constructing the monthly generation frequency prediction model of the low-pressure zone in the Tibetan Plateau as described in any one of claims 1 to 4, or the method for predicting the low-pressure zone in the Tibetan Plateau as described in claim 5.

Images