A method, medium and program product for calculating and evaluating the change in water vapor mass of tropical cyclones based on machine learning and integrating the exact differential equation of water vapor pressure

Abstract

The present invention discloses a method, medium and program product for calculating and evaluating the water vapor mass change of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation, belonging to the technical fields of meteorological science and numerical calculation. The method includes steps such as constructing an exact water vapor pressure differential equation considering the nonlinear effect of specific humidity, establishing a calculation model for the water vapor mass change of the air column per unit area, obtaining meteorological data in the tropical cyclone region, dynamically tracking the tropical cyclone and determining the calculation area, constructing a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of the tropical cyclone and calculating and evaluating the water vapor mass change of different height layers layer by layer, and outputting the calculation and evaluation results of the water vapor mass change. By introducing the nonlinear effect of specific humidity, the present invention can more accurately reflect the non-uniformity of the water vapor distribution in the atmosphere, thereby improving the accuracy of the calculation of the water vapor mass change, and providing key technical support for the prediction of tropical cyclone paths, the simulation of intensity and the optimization of numerical models.

Description

Technical Field

[0001] The present invention belongs to the technical fields of meteorological science and numerical calculation, and relates to the assessment and analysis of the change in water vapor mass of tropical cyclones and numerical calculation techniques. Specifically, it is a method, medium, and program product for calculating and evaluating the change in water vapor mass of tropical cyclones based on machine learning and integrating an exact water vapor pressure differential equation, which can be used for quantitative analysis of the spatio-temporal evolution characteristics of tropical cyclone water vapor content and simulation and prediction of intensity changes. Background Art

[0002] A tropical cyclone (TC) is an extreme phenomenon in the global weather system, and its development and evolution are of great significance to the research of atmospheric dynamics and thermodynamics. The generation, evolution, and dissipation of tropical cyclones are closely related to the distribution and change of atmospheric humidity. The latent heat release during the phase change of water vapor is the main energy source for the development of tropical cyclone intensity, and the horizontal and vertical transports of the humidity field directly affect the intensity change and internal water vapor redistribution of tropical cyclones. Therefore, accurately calculating and evaluating the change in tropical cyclone water vapor mass is of great significance for revealing its intensity change mechanism, improving the simulation accuracy of numerical models, and enhancing the tropical cyclone intensity prediction ability.

[0003] Atmospheric humidity is an important physical quantity characterizing the water vapor content in the atmosphere, and its commonly used measurement indicators include water vapor pressure, water vapor density, mixing ratio, specific humidity, relative humidity, and dew point temperature, etc. Among them, water vapor pressure is the weight of water vapor in an air column per unit area, directly reflecting the water content in the atmosphere. Specific humidity represents the mass of water vapor in unit mass of moist air and is a key parameter for describing the characteristics of moist air. Research shows that the rapid intensification of tropical cyclones is often accompanied by a significant increase in low-level humidity and a significant decrease in high-level humidity, and this humidity distribution characteristic directly affects the latent heat release efficiency in the cyclone core area and the intensity development process. Under the background of global warming, the atmospheric water holding capacity continues to increase, the atmospheric humidity shows an upward trend, and the TC intensity also shows an increasing trend, and this correlation has been widely recognized at the climate scale. However, at the synoptic scale, the influence mechanism of atmospheric humidity on TC intensity is quite complex, and the humidity and its changes at different heights, positions, and degrees will cause different responses of TC structure and intensity. Current research shows that the increase in humidity in the TC core area is conducive to its rapid intensification, but the response of TC intensity to atmospheric humidity is not linear, and there is a complex non-linear feedback mechanism.

[0004] The calculation of water vapor mass change is the core link in humidity field analysis. The existing water vapor pressure differential equation usually adopts in the form of, where p v is the water vapor pressure, p is the ambient atmospheric pressure, qv is the specific humidity of the atmosphere, t t is time. This equation completely ignores the nonlinear effect of the specific humidity of the atmosphere and cannot accurately reflect the non-uniformity characteristics of the water vapor distribution in the atmosphere. Specifically, the above equation only considers the zero-order term of the specific humidity and ignores the contributions of the first-order and higher-order nonlinear terms. This simplified treatment leads to a large deviation between the calculated results of the equation and the actual observations. Especially in the critical stage of TC development, this deviation may affect the accurate assessment of the TC intensity change. For example, the humidification rate of water vapor is usually overestimated in the lower layer, while the drying rate of water vapor may be underestimated in the middle and upper layers.

[0005] In summary, the existing differential equation of water vapor pressure has problems such as insufficient calculation accuracy, ignoring the non-uniformity of water vapor distribution, and not considering the nonlinear effect of specific humidity when calculating the change of TC water vapor mass. Therefore, it is an urgent technical problem to establish a calculation and evaluation method for the change of TC water vapor mass that considers the nonlinear effect of specific humidity and can provide higher calculation accuracy, so as to more accurately calculate and evaluate the change of water vapor mass in the tropical cyclone area, and thus more deeply understand the impact of water vapor on the change of tropical cyclone intensity. Summary of the Invention

[0006] (I) Object of the Invention

[0007] In view of the defects and deficiencies that most of the existing calculation methods for the change of tropical cyclone water vapor mass are based on simplified differential equations of water vapor pressure, ignoring the nonlinear effect of specific humidity, resulting in insufficient calculation accuracy and difficulty in accurately evaluating the impact of water vapor on the change of tropical cyclone intensity, etc., to solve at least one of the above and other technical problems in the prior art, the present invention aims to provide a calculation and evaluation method, medium and program product for the change of tropical cyclone water vapor mass based on machine learning and integrating an accurate differential equation of water vapor pressure. By introducing an accurate differential equation of water vapor pressure considering the nonlinear effect of specific humidity and combining dynamic tracking and analysis techniques, high-precision calculation and evaluation of the change of water vapor mass at different vertical levels during the development of tropical cyclones are realized, significantly improving the accuracy of humidity change calculation and the adaptability to complex weather systems, so as to provide key technical support for the intensity prediction, path simulation of tropical cyclones and the development of related numerical models.

[0008] (II) Technical Solution

[0009] To achieve the object of the invention and solve its technical problems, the present invention adopts the following technical solutions:

[0010] The first object of the present invention is to provide a method for calculating and evaluating the change in water vapor mass of tropical cyclones based on machine learning and integrating an accurate differential equation of water vapor pressure, which is used to calculate and evaluate the change in water vapor mass at different vertical levels during the development of tropical cyclones, and improve the calculation accuracy of the non-uniformity and change law of humidity distribution. When the method is implemented, it at least includes the following steps:

[0011] SS1. Establish an accurate differential equation of water vapor pressure considering the non-linear effect of specific humidity

[0012] Combining the atmospheric physical characteristics and the relationship between specific humidity and water vapor pressure, introducing the non-linear term of specific humidity, and establishing an accurate differential equation of water vapor pressure containing the non-linear effect of specific humidity:

[0013]

[0014] Among them, p is the ambient atmospheric pressure, t is time, p v is the water vapor pressure, q v is the atmospheric specific humidity, γ v is the non-linear coefficient of specific humidity, dp / dt is the rate of change of air pressure per unit time, dp v / dt is the rate of change of water vapor pressure per unit time, dq v / dt is the rate of change of specific humidity per unit time;

[0015] SS2. Construct a calculation model for the change in water vapor mass of an air column per unit area

[0016] Substitute the accurate differential equation of water vapor pressure established in step SS1 into the calculation formula of the water vapor mass of an air column per unit area to construct a calculation model for the change in water vapor mass of an air column per unit area , and expand the non-linear term of specific humidity into an infinite series form in the constructed calculation model , to fully consider the adjustment effect of the non-linear effect of specific humidity on water vapor distribution, and improve the calculation accuracy of water vapor mass change to the non-linear order, where m v is the water vapor mass of an air column per unit area, g is the acceleration of gravity, p b and p t are the bottom air pressure and the top air pressure of an air column per unit area respectively, kis the order of the terms in the infinite series expansion;

[0017] SS3. Dynamically track tropical cyclones and determine the calculation area

[0018] Obtain the meteorological data of the tropical cyclone area to be evaluated for calculation. Determine the intensity and position of the tropical cyclone center at each time based on the lowest mean sea level pressure value and its position in the meteorological data. By dynamically tracking the tropical cyclone path, establish a dynamic coordinate system based on the tropical cyclone center position. Take the tropical cyclone center position at each time as the origin and the center of the dynamic calculation area of the dynamic coordinate system, and map the relevant meteorological physical quantities into the dynamic coordinate system. By adjusting the range and position of the calculation area in real time to make it precisely match the moving path of the tropical cyclone center, realize the dynamic tracking of the tropical cyclone, and ensure the accuracy of the spatio-temporal calculation and evaluation of the water vapor mass change;

[0019] SS4. Construct a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones

[0020] Based on the calculation model of the water vapor mass change of the air column per unit area constructed in step SS2, use a machine learning algorithm to construct a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones. The multi-layer convolutional neural network model includes an input layer, a hidden layer, and an output layer. The input layer of the model is used to receive the meteorological data of each height layer in the tropical cyclone area under the dynamic coordinate system and the water vapor pressure change rate calculated according to the exact water vapor pressure differential equation. The hidden layer uses multiple convolutional layers and pooling layers to extract multi-dimensional features of the water vapor distribution of tropical cyclones and uses a fully connected layer for feature fusion. At the same time, an attention mechanism module is introduced to capture the correlation of water vapor mass change between different height layers. The output layer uses the water vapor mass change calculated according to the results of the calculation model in step SS2 as the label of the training data, and calculates and evaluates the water vapor mass change in the air column per unit area of each height layer respectively;

[0021] SS5. Output the calculation and evaluation results of the water vapor mass change

[0022] Use the multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones constructed in step SS4, combined with the dynamically determined calculation area and the meteorological data of each height level in step SS3, to generate the spatio-temporal distribution data of the water vapor mass change of tropical cyclones at different height levels.

[0023] The second object of the present invention is to provide a computer program product, including computer instructions, to execute the above-mentioned method for calculating and evaluating the water vapor mass change of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation.

[0024] The third object of the present invention is to provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned method for calculating and evaluating the change of water vapor mass of tropical cyclones based on machine learning and integrating the exact differential equation of water vapor pressure is realized.

[0025] (III) Technical effects

[0026] Compared with the prior art, the method for calculating and evaluating the change of water vapor mass of tropical cyclones, the medium and the program product of the present invention based on machine learning and integrating the exact differential equation of water vapor pressure have the following beneficial and remarkable technical effects:

[0027] (1) By constructing an exact differential equation of water vapor pressure and introducing the nonlinear effect of specific humidity, the present invention significantly improves the calculation accuracy of the change of water vapor mass. The traditional differential equation of water vapor pressure fails to fully consider the nonlinear characteristics of specific humidity. Based on the relationship between specific humidity and water vapor pressure, the present invention constructs an exact differential equation of water vapor pressure containing the nonlinear term of specific humidity, and uses the infinite series expansion method to accurately characterize the contribution of specific humidity to the non-uniformity of water vapor distribution. Compared with the traditional method, the present invention can more accurately reflect the true change characteristics of humidification in the lower layer and drying in the middle and upper layers. Especially during the development of tropical cyclones, the introduction of the nonlinear term reduces the calculation error, and the average residual error of the change of water vapor mass is significantly reduced.

[0028] (2) By establishing a calculation model for the change of water vapor mass of the air column per unit area and combining the exact differential equation of water vapor pressure, the present invention realizes the hierarchical calculation and evaluation of the change of water vapor mass at different height levels of tropical cyclones. By combining the corrected calculation of the exact equation and the physical level division, the present invention can calculate the change of water vapor mass of the air column per unit area at the lower layer, middle layer and upper layer respectively, and clarify the hierarchical characteristics of water vapor distribution. Especially, the significant humidification rate in the lower layer and the drying rate in the middle and upper layers are accurately quantified, providing a scientific basis for studying the influence of humidity distribution on the intensity change of tropical cyclones.

[0029] (3) By dynamically tracking the path of tropical cyclones and establishing a dynamic coordinate system, the present invention realizes the real-time matching of the calculation area with the position of the center of tropical cyclones. Traditional methods mostly use fixed-area calculations, and for weather systems such as tropical cyclones with significant mobility and non-uniformity, they cannot accurately reflect the spatio-temporal dynamic characteristics of their humidity fields. By using the dynamic tracking method, the present invention takes the position of the center of the tropical cyclone as the origin of the dynamic coordinate system, and adjusts the range and position of the calculation area in real time, so that the diagnostic results of the change of water vapor mass can more accurately reflect the actual development process of tropical cyclones. At the same time, the dynamic tracking method effectively solves the problem of distortion of the diagnostic results of the humidity field and significantly improves the spatio-temporal adaptability of the calculation and evaluation.

[0030] (4) The present invention further quantifies the regulatory effect of the specific humidity nonlinear effect on the change of water vapor mass, providing a new perspective for studying the influence of humidity on the intensity change of tropical cyclones. By introducing the specific humidity nonlinear term, the present invention discovers that the nonlinear effect of specific humidity can slow down the humidification rate in the lower layer and the drying rate in the middle and upper layers, promoting the concentration of water vapor to higher layers. This not only reveals the complex nonlinear law of humidity distribution but also provides technical support for the numerical simulation and model optimization of the humidity field.

[0031] (5) By constructing a multi-layer convolutional neural network model based on machine learning and combining it with the exact water vapor pressure differential equation and dynamic tracking technology, the present invention effectively captures the complex nonlinear relationships and spatio-temporal variation characteristics in meteorological data. Compared with traditional physical model-based calculation methods, the present invention shows higher accuracy and generalization ability in predicting the change of water vapor mass, and can more accurately predict the change of water vapor mass of tropical cyclones at different development stages and different intensities, providing more reliable technical support for the research on the mechanism of tropical cyclone intensity change, path simulation, and numerical model optimization. Brief Description of the Drawings

[0032] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. Hereinafter, the embodiments of the present invention will be described in detail with reference to the drawings, where:

[0033] Figure 1 The figure shows the implementation flowchart of the calculation and evaluation method for the change of water vapor mass of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation provided by the embodiment of the present invention;

[0034] Figure 2 The figure shows the schematic diagram of the change of (a) the central moving path and (b) the central mean sea level pressure and its 12-hour pressure change of tropical cyclone "Doksuri" from 00:00 on July 21, 2023, to 00:00 on July 29, 2023. Among them: In figure (a), the blue, red, and black dotted lines are the development, intensification, and mature stages of "Doksuri" respectively; in figure (b), the blue and red dotted lines are the boundaries between the development and intensification stages and between the intensification and mature stages of "Doksuri" respectively, and the red dotted line is the central mean sea level pressure, unit: hPa, and the gray column is the 12-hour pressure change, unit: hPa;

[0035] Figure 3 The figure shows the individual changes of the water vapor mass in the lower, middle, and upper layers of the tropical cyclone "Doksuri" per unit area of the air column calculated based on (a) the exact water vapor pressure differential equation and (b) the first-order nonlinear term of specific humidity (unit: 10 -4 kg / (m 2·s)), and the schematic diagram of the 12-hour change of the central mean sea level pressure of "Doksuri" (gray column, unit: hPa). In the figure: the black dotted line is the low layer, the red dotted line is the middle layer, and the blue dotted line is the high layer;

[0036] Figure 4 The figure shows the schematic diagram of the percentage of the residual difference of the exact water vapor pressure differential equation (solid line) and the commonly used water vapor pressure differential equation (dashed line) in the low layer (black dotted line), middle layer (red dotted line), and high layer (blue dotted line) of "Doksuri", respectively, and the 12-hour change of the central mean sea level pressure of "Doksuri" (gray column, unit: hPa). Specific implementation manner

[0037] To make the purpose, technical solution, and advantages of the implementation of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are part of the embodiments of the present invention, rather than all of the embodiments, and the described embodiments are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0038] The present invention aims to provide a method, medium, and program product for calculating and evaluating the water vapor mass change of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation. By introducing the exact water vapor pressure differential equation considering the nonlinear effect of specific humidity and combining the dynamic tracking and analysis technology, it realizes the high-precision calculation and evaluation of the water vapor mass change in different vertical layers during the development process of tropical cyclones, significantly improves the accuracy of humidity change calculation and the adaptability to complex weather systems, and thus provides key technical support for the intensity prediction, path simulation of tropical cyclones, and the development of related numerical models.

[0039] Embodiment 1

[0040] As a specific example, as Figure 1 shown, the method for calculating and evaluating the water vapor mass change of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation of the present invention includes the following steps:

[0041] SS1. Establish an exact water vapor pressure differential equation considering the nonlinear effect of specific humidity

[0042] Combining the atmospheric physical characteristics and the relationship between specific humidity and water vapor pressure, introducing the nonlinear term of specific humidity, and establishing an exact water vapor pressure differential equation containing the nonlinear effect of specific humidity:

[0043]

[0044] Among them,p is the ambient atmospheric pressure, t is the time, p v is the water vapor pressure, q v is the specific humidity of the atmosphere, γ v is the non - linear coefficient of specific humidity, dp / dt is the rate of change of air pressure per unit time, dq v / dt is the rate of change of specific humidity per unit time, dp v / dt is the rate of change of water vapor pressure per unit time.

[0045] Preferably, the derivation process of the accurate water vapor pressure differential equation includes: First, based on the ideal gas state equation and Dalton's law of partial pressures, construct the functional relationship between the partial pressure of water vapor and specific humidity; Second, consider the non - linear effect caused by the mass difference between water vapor molecules and dry air molecules in the atmosphere, and introduce the quadratic term of specific humidity; Finally, through the total differential operation of the relationship between water vapor pressure and specific humidity, obtain the accurate water vapor pressure differential equation including the non - linear effect of specific humidity. In addition, in the accurate water vapor pressure differential equation established by the present invention, the non - linear coefficient γ v has the calculation formula , M d is the molar mass of dry air, M v is the molar mass of water vapor. By introducing the non - linear coefficient γ v , to accurately reflect the non - linear relationship between water vapor pressure and specific humidity. And, the molar mass M d of dry air is preferably taken as 28.96 g / mol, and the molar mass M v of water vapor is preferably taken as 18.02 g / mol. Therefore, the value of the non - linear coefficient γ v is preferably taken as 0.61.

[0046] More specifically, the embodiments of the present invention construct the accurate water vapor pressure differential equation in the following manner. First, the water vapor pressure is the weight of water vapor in the air column per unit area, which directly reflects the water content of the atmosphere. When the water vapor pressure in the air reaches the maximum value, it is called saturated humid air, and the water vapor pressure at this time is the saturated water vapor pressure. The ratio of water vapor pressure to saturated water vapor pressure reflects the relative humidity of the atmosphere, that is, the water vapor saturation ratio U v :

[0047]

[0048] Among them, U v i.e., the relative humidity, also known as the water vapor saturation ratio, is the water vapor supersaturation ratio and is a second-order small quantity relative to 1, p v and p vs are the water vapor pressure and the saturated water vapor pressure respectively.

[0049] Secondly, let p be the ambient atmospheric pressure, and the calculation formulas for the proportion of water vapor pressure χ v and the proportion of saturated water vapor pressure χ vs are respectively , . It can be seen from the above definitions that the saturated water vapor pressure ratio can also be regarded as the molar ratio, and its value is calculated by a high-precision empirical formula (such as the Goff-Gratch formula or the Tetens formula). Under general atmospheric conditions, the value of the proportion of water vapor pressure χ v usually does not exceed 0.06. Further combining the relationship between water vapor pressure and specific humidity q v and combining the above equations, we can obtain:

[0050]

[0051] Among them, q v is the atmospheric specific humidity, γ v is the non-linear coefficient of specific humidity. By performing a total differential operation on the above equation, the accurate water vapor pressure differential equation established by the present invention can be obtained:

[0052]

[0053] Compared with the commonly used water vapor pressure differential equation in the prior art, the accurate water vapor pressure differential equation established by the present invention considers the non-linear effect of the atmospheric specific humidity q v by introducing the non-linear term of specific humidity, and improves the accuracy of the water vapor pressure differential equation.

[0054] SS2. Construct a calculation model for the change in the water vapor mass of an air column per unit area

[0055] Substitute the accurate water vapor pressure differential equation constructed in step SS1 into the calculation formula for the water vapor mass of an air column per unit area , a calculation model for the change in water vapor mass of the air column per unit area is constructed , and the specific humidity non - linear term in this calculation model is expanded in the form of an infinite series , so as to improve the calculation accuracy to the non - linear order and fully consider the adjustment effect of the non - linear effect of specific humidity on water vapor distribution, where m v is the water vapor mass of the air column per unit area, g is the acceleration due to gravity, p b and p t are the bottom pressure and the top pressure of the air column per unit area respectively, k is the order of the infinite series expansion term.

[0056] Preferably, when calculating the specific humidity non - linear term in the form of an infinite series, according to the typical magnitude of specific humidity q v under actual atmospheric conditions and the value range of the non - linear coefficient γ v of specific humidity, only the first several series terms can be selected as the approximate value of the calculation, so as to maintain the calculation accuracy while maintaining the high - precision description ability of the model for the non - linear effect of specific humidity.

[0057] SS3. Dynamically track tropical cyclones and determine the calculation area

[0058] Obtain the meteorological data of the tropical cyclone area to be evaluated and calculated. Determine the intensity and position of the tropical cyclone center at each time based on the lowest mean sea - level pressure value and its position in the meteorological data at each time. By dynamically tracking the tropical cyclone path, establish a dynamic coordinate system based on the tropical cyclone center position, take the tropical cyclone center position at each time as the origin of the moving coordinate and the center of the dynamic calculation area, and map the relevant meteorological physical quantities into the dynamic coordinate system. By adjusting the range and position of the calculation area in real - time to make it accurately match the moving path of the tropical cyclone center, realize the dynamic tracking of the tropical cyclone, and ensure the spatio - temporal calculation and evaluation accuracy of the water vapor mass change.

[0059] Preferably, meteorological data of a tropical cyclone to be evaluated and calculated are obtained based on reanalysis data, satellite remote sensing observation data, and / or surface meteorological observation data. The meteorological data includes at least horizontal wind field data with a time resolution of not less than 6 hours, a spatial resolution of not less than 1°×1°, and a vertical range from 1000 hPa to 50 hPa, vertical velocity data, geopotential height data, specific humidity data, relative humidity data, and mean sea level pressure data. The time range of the meteorological data covers the complete life history process of the tropical cyclone from generation to dissipation. The meteorological data from different sources are preprocessed through quality control and data assimilation methods to ensure the spatio-temporal continuity and physical consistency of the data. Through the acquisition and integration of the above meteorological data, comprehensive support can be provided for the dynamic tracking of the tropical cyclone area, the real-time adjustment of the calculation area, and the assessment of water vapor mass change, while improving the applicability and accuracy of the calculation model at different levels.

[0060] Preferably, the specific steps for dynamically tracking a tropical cyclone may include: determining the intensity and position of the tropical cyclone center at each time step using the lowest mean sea level pressure value and its position at each time step, constructing the movement trajectory of the tropical cyclone through the time series of the tropical cyclone center position, establishing a dynamic coordinate system with the tropical cyclone center as the origin, and constructing a dynamic calculation area (e.g., 5°×5° in size) centered on the longitude and latitude coordinates of the tropical cyclone center, and adjusting the position of the calculation area in real time as the tropical cyclone center moves to ensure that the calculation area always covers the core area of the tropical cyclone. In addition, dynamically tracking the tropical cyclone further includes mapping relevant meteorological physical quantities (including horizontal wind field, vertical velocity, specific humidity, relative humidity, etc.) at each time step to the dynamic coordinate system using interpolation methods, and updating the spatial distribution characteristics of the physical quantities in real time in combination with the changes in the dynamic tracking path to ensure the dynamic consistency between the assessment of water vapor mass change and the actual physical field.

[0061] SS4. Construct a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones

[0062] Based on the calculation model of the change in water vapor mass of the air column per unit area constructed in step SS2, a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones is constructed using a machine learning algorithm. The multi-layer convolutional neural network model includes an input layer, a hidden layer, and an output layer. The input layer of the model is used to receive meteorological data of each height layer in the tropical cyclone area in a dynamic coordinate system and the rate of change of water vapor pressure calculated according to the exact water vapor pressure differential equation. The hidden layer uses multiple convolutional layers and pooling layers to extract multi-dimensional features of the water vapor distribution of the tropical cyclone and uses a fully connected layer for feature fusion. At the same time, an attention mechanism module is introduced to capture the correlation of the change in water vapor mass between different height layers. The output layer uses the change in water vapor mass calculated according to the results of the calculation model in step SS2 as the label of the training data, and calculates and evaluates the change in water vapor mass in the air column per unit area of each height layer respectively.

[0063] Preferably, in the multi-layer convolutional neural network model of the present invention, its input layer divides the dynamic calculation area into three height levels in the vertical direction of the atmosphere for data reception. Among them, the lower layer takes the meteorological element field in the range of 1000 - 700 hPa, which is used to analyze the water vapor transportation and humidification process of the bottom atmosphere; the middle layer takes the meteorological element field in the range of 700 - 500 hPa, which is used to capture the water vapor fluctuation characteristics of the middle atmosphere; the upper layer takes the meteorological element field in the range of 500 - 50 hPa, which is used to evaluate the drying effect of the upper atmosphere. In addition, the output layer of the multi-layer convolutional neural network model uses the mean square error loss function to measure the difference between the predicted value of the multi-layer convolutional neural network model and the change in water vapor mass calculated according to the results of the calculation model in step SS2, and introduces the absolute value or square of the difference between the two as an additional loss term to optimize and adjust the loss function, so as to improve the prediction accuracy of the model and the consistency with the results of the calculation model.

[0064] Preferably, in step SS4, according to the development stages of the life cycle of the tropical cyclone, including the development stage, the intensification stage, and the mature stage, the change characteristics of water vapor mass at different height levels in different stages are calculated and dynamically evaluated layer by layer, so as to reveal the humidity change law of the tropical cyclone in different development stages and provide key data support and theoretical basis for the evolution of cyclone intensity and path prediction.

[0065] In addition, the hierarchical calculation for evaluating the change in water vapor mass at different altitude levels may include the analysis of the convergence effect on the change in water vapor mass in the lower layer. By calculating the horizontal convergence rate through the humidification rate of the change in water vapor mass in the lower layer, the contribution of the change in lower-layer humidity to the intensification of tropical cyclone intensity can be evaluated. Meanwhile, in combination with the changes in the middle and upper layers, the impact of the accumulation of lower-layer humidity on the overall humidity distribution can be analyzed. At the same time, the hierarchical calculation for evaluating the change in water vapor mass at different altitude levels may also include the analysis of the fluctuation characteristics of the change in water vapor mass in the middle layer. By calculating the fluctuation intensity of the change in water vapor mass in the air column per unit area in the middle layer, the impact of the middle-layer humidity fluctuation on the cyclone structure and convective activities can be evaluated, and the spatio-temporal coupling relationship between the middle-layer humidity fluctuation and the lower-layer humidification can be analyzed.

[0066] SS5. Output the calculation and evaluation results of the change in water vapor mass

[0067] Using the multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones constructed in step SS4, combined with the calculation area dynamically determined in step SS3 and the meteorological data at each altitude level, generate the spatio-temporal distribution data of the change in water vapor mass at different altitude levels of tropical cyclones.

[0068] Preferably, the calculation and evaluation results of the change in water vapor mass may further include the analysis of the contribution rate of the non-linear effect of specific humidity. The specific method is as follows: By comparing the calculation results of the change in water vapor mass based on the exact differential equation of water vapor pressure introducing the non-linear term of specific humidity with the calculation results of the change in water vapor mass based on the differential equation of water vapor pressure without introducing the non-linear term, calculate the contribution rate of the non-linear term of specific humidity to the lower-layer humidification rate, the middle-layer fluctuation intensity, and the upper-layer drying rate, and comprehensively evaluate the impact of the non-linear effect of specific humidity on the regulation of the overall humidity distribution.

[0069] Through the above steps, based on considering the non-linear effect of specific humidity, the present invention realizes the high-precision calculation and evaluation of the change in water vapor mass of tropical cyclones, improves the diagnostic accuracy of humidity distribution, and provides technical support for the intensity simulation, path prediction, and numerical model optimization of tropical cyclones.

[0070] Embodiment 2

[0071] On the basis of the above Embodiment 1, as a more specific and detailed example, in this Embodiment 2, Typhoon Doksuri (No. 5 in 2023) is taken as the research object. Combining the reanalysis data (FNL data) of NCEP, the change in water vapor mass at different vertical levels during the development of the typhoon is calculated and evaluated, fully verifying the effectiveness and applicability of the method for calculating and evaluating the change in water vapor mass of tropical cyclones based on machine learning and integrating the exact differential equation of water vapor pressure described in the present invention.

[0072] In this embodiment, the 6-hourly global analysis field data provided by NCEP is adopted. The data includes the following meteorological elements: horizontal wind field, vertical velocity, geopotential height, specific humidity, relative humidity, and mean sea level pressure. The horizontal resolution is 1°×1°, and the vertical direction is from 1000 hPa to 50 hPa. The time coverage is from 00:00 on July 21, 2023 to 00:00 on July 29, 2023, completely recording the whole process of the typhoon "Doksuri" from generation to dissipation. According to the minimum mean sea level pressure value and its position at each time step, the intensity and path changes of the typhoon "Doksuri" are determined, specifically as Figure 2 shown. Figure 2 In (a), the central movement path of "Doksuri" is shown. The typhoon was generated over the ocean east of the Philippines at 00:00 on July 21, 2023, and then continuously moved westward and strengthened, reaching the super typhoon intensity at 12:00 on July 24. At 21:00 on July 26, "Doksuri" weakened to a severe typhoon, and the path changed from westward to northward. It strengthened to a super typhoon again and made landfall along the coast of Fujian, China at 02:00 on July 28, and finally was decommissioned at 03:00 on July 29. Figure 2 In (b), the mean sea level pressure at the center of the typhoon "Doksuri" and its 12-hour pressure change are shown. Its intensity change is divided into three stages: the first 60 hours (from 00:00 on July 21 to 12:00 on July 23) is the development stage, the middle 60 hours (from 12:00 on July 23 to 00:00 on July 26) is the strengthening stage, and the last 72 hours (from 00:00 on July 26 to 00:00 on July 29) is the mature stage.

[0073] According to Figure 2 the typhoon path and intensity changes shown, using the position of the lowest mean sea level pressure center of "Doksuri", a dynamic coordinate system is established through a dynamic tracking method. The typhoon center position is used as the origin of the moving coordinate, and a 5°×5° area centered on the typhoon center is selected as the dynamic calculation area. The specific steps include: determining the longitude and latitude positions of the typhoon center at each time step, constructing a dynamic coordinate system with this position as the origin; mapping the physical quantities (such as specific humidity, relative humidity, horizontal wind field, etc.) in the meteorological data to the dynamic calculation area; and adjusting the position of the dynamic calculation area in real time as the typhoon center moves, so that it always covers the typhoon core area.

[0074] The dynamic calculation area is divided into three height layers in the vertical direction of the atmosphere: the lower layer (1000 - 700 hPa), the middle layer (700 - 500 hPa), and the upper layer (500 - 50 hPa). Using the water vapor mass change calculation model constructed in the present invention, the water vapor mass change of the air column per unit area in different height layers is calculated respectively, and its change trend is analyzed. The results are as Figure 3 shown. Figure 3Figure (a) shows the evolution trends of the water vapor mass changes in the air columns per unit area of the lower, middle, and upper layers: For the lower layer, during the development and intensification stages, the water vapor mass in the lower layer always remains positive, indicating continuous humidification, and the humidification rate continuously accelerates as the typhoon intensity increases; during the mature stage (especially after landing), the humidification rate significantly decreases, indicating that the humidification process is closely related to the typhoon intensity. For the middle layer, the change in water vapor mass in the middle layer shows a fluctuating characteristic, with a brief humidification during the development stage and mainly drying in the remaining periods. For the upper layer, the water vapor mass in the upper layer is always negative, indicating continuous drying, and the drying rate shows an anti-phase characteristic with the humidification rate in the lower layer. Figure 3 Figure (b) shows the regulating effect of the non-linear term of specific humidity on the water vapor mass changes in the lower, middle, and upper layers: For the lower layer, the non-linear term of specific humidity shows a trend of slowing down the humidification, and its effect increases as the typhoon intensity increases. For the middle and upper layers, the non-linear term of specific humidity shows a trend of slowing down the drying, especially during the intensification stage, and its effect is significantly enhanced.

[0075] Based on the above stratified calculations, a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones constructed according to step SS4 in Embodiment 1 is used to further evaluate the water vapor mass changes of Typhoon "Doksuri". For the input layer of the model, meteorological data at each height layer in the tropical cyclone area in the dynamic coordinate system (such as horizontal wind field, vertical velocity, specific humidity, relative humidity, etc.) and the change rate of water vapor pressure calculated by the exact water vapor pressure differential equation are input. The hidden layer uses 4 convolutional layers and 3 max-pooling layers to extract features. The size of the convolutional kernel is 3×3, and the number of channels is 32, 64, 128, and 256 in sequence. A spatial attention mechanism and a channel attention mechanism are introduced after the last convolutional layer to capture the correlation of water vapor mass changes between different height layers, and the features of the lower, middle, and upper layers are fused through a fully connected layer to generate comprehensive features. The output layer of the model outputs the water vapor mass changes in the air columns per unit area at each height layer, generating spatio-temporal distribution data of the humidification rate in the lower layer, the fluctuation intensity in the middle layer, and the drying rate in the upper layer. And when training the multi-layer convolutional neural network model, the Adam optimizer is used, the initial value of the learning rate is set to 0.001, and it decays to 0.1 of the original value every 50 epochs; the batch size is set to 32; the number of training rounds is 200 rounds; the loss function uses the mean square error function, and an L2 regularization term is introduced to prevent overfitting. The performance of the model is evaluated through 5-fold cross-validation, and the average validation set loss converges to below 0.015.

[0076] To verify the calculation accuracy of the method for calculating and evaluating the water vapor mass changes of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation described in the present invention, it is compared with the common water vapor pressure differential equation, and the results are as Figure 4 shown. Figure 4Shows the proportion of calculation residuals of the exact equation and the common equation in the lower, middle, and upper layers: For the lower layer, the average proportion of residuals of the exact equation is 1 / 7 of that of the common equation, indicating that introducing the specific humidity nonlinear term significantly improves the calculation accuracy of the humidification process in the lower layer. For the middle and upper layers, the average proportion of residuals of the exact equation is 1 / 2 of that of the common equation. In particular, the accuracy advantage is more significant during the humidification process, while the common equation slightly overestimates the water vapor content during the drying process.

[0077] Finally, based on the stratified calculation results, spatio-temporal distribution data of the water vapor mass change at different altitude layers during the evolution of "Doksuri" are generated, specifically including: distribution maps and change curves of the water vapor mass change at different altitude layers; analysis of the contribution rate of the specific humidity nonlinear effect to the humidification rate in the lower layer, the fluctuation intensity in the middle layer, and the drying rate in the upper layer; evaluation of the impact of the water vapor mass change on the typhoon intensity change, etc.

[0078] Through the above implementation process, the present invention verifies its applicability and advantages to complex weather systems in the case of "Doksuri", significantly improving the calculation accuracy of the water vapor mass change, and providing reliable support for tropical cyclone intensity simulation and path prediction.

[0079] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for calculating and evaluating the change in water vapor mass of tropical cyclones based on machine learning and integrating the exact differential equation of water vapor pressure, characterized in that, When the method is implemented, it at least includes the following steps: SS1. Combining the atmospheric physical characteristics and the relationship between specific humidity and water vapor pressure, introducing the non-linear term of specific humidity, and establishing an accurate differential equation of water vapor pressure containing the non-linear effect of specific humidity: wherein, p is the ambient atmospheric pressure, t is the time, p v is the water vapor pressure, q v is the specific humidity of the atmosphere, γ v is the non - linear coefficient of the specific humidity, dp / dt is the rate of change of the air pressure per unit time, dp v / dt is the rate of change of the water vapor pressure per unit time, dq v / dt is the rate of change of the specific humidity per unit time; SS2. Substitute the exact differential equation of water vapor pressure established in step SS1 into the calculation formula for the water vapor mass of the air column per unit area to construct a calculation model for the change in the water vapor mass of the air column per unit area , and the specific humidity non - linear term in this calculation model is expanded in the form of an infinite series , where m v is the water vapor mass of the air column per unit area, g is the acceleration due to gravity, p b and p t are the bottom pressure and top pressure of the air column per unit area respectively, k is the order of the infinite series expansion term; SS3. Obtain the meteorological data of the tropical cyclone area to be evaluated and calculated. Determine the intensity and position of the tropical cyclone center at each time based on the lowest mean sea level pressure value and its position in the meteorological data. By dynamically tracking the tropical cyclone path, establish a dynamic coordinate system based on the tropical cyclone center position. Take the tropical cyclone center position at each time as the origin of the moving coordinate and the center of the dynamic calculation area, and map the relevant meteorological physical quantities into the dynamic coordinate system. By adjusting the range and position of the calculation area in real time to make it accurately match the moving path of the tropical cyclone center, realize the dynamic tracking of the tropical cyclone, and ensure the accuracy of the spatio-temporal calculation and evaluation of the water vapor mass change; SS4. Based on the calculation model of the water vapor mass change of the air column per unit area constructed in step SS2, use a machine learning algorithm to construct a multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones. The multi-layer convolutional neural network model includes an input layer, a hidden layer, and an output layer. The input layer of the model is used to receive the meteorological data of each height layer in the tropical cyclone area under the dynamic coordinate system and the rate of change of water vapor pressure calculated according to the accurate differential equation of water vapor pressure. The hidden layer uses multiple convolutional layers and pooling layers to extract multi-dimensional features of the water vapor distribution of tropical cyclones and uses a fully connected layer for feature fusion. At the same time, an attention mechanism module is introduced to capture the correlation of the water vapor mass change between different height layers. The output layer uses the water vapor mass change calculated according to the result of the calculation model in step SS2 as the label of the training data, and calculates and evaluates the water vapor mass change in the air column per unit area of each height layer respectively; SS5. Using the multi-layer convolutional neural network model for calculating and evaluating the water vapor mass of tropical cyclones constructed in step SS4, combined with the dynamically determined calculation area and the meteorological data of each height level in step SS3, generate the spatio-temporal distribution data of the water vapor mass change of tropical cyclones at different height levels.

2. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact water vapor pressure differential equation according to claim 1, wherein In step SS1, the derivation process of the accurate differential equation of water vapor pressure includes: First, based on the ideal gas state equation and Dalton's law of partial pressures, construct the functional relationship between water vapor partial pressure and specific humidity; Second, consider the non-linear effect caused by the mass difference between water vapor molecules and dry air molecules in the atmosphere, and introduce the quadratic term of specific humidity; Finally, through the total differential operation of the relationship between water vapor pressure and specific humidity, obtain the accurate differential equation of water vapor pressure containing the non-linear effect of specific humidity.

3. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 1, characterized in that, In step SS1, the nonlinear coefficient of specific humidity in the exact water vapor pressure differential equation γ v , and its calculation formula is , where M d is the molar mass of dry air, M v is the molar mass of water vapor. By introducing the nonlinear coefficient γ v , the nonlinear relationship between water vapor pressure and specific humidity can be accurately reflected.

4. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact water vapor pressure differential equation according to claim 1, wherein, In step SS2, when calculating the specific humidity non - linear term in the form of an infinite series, according to the typical magnitude of specific humidity q v under actual atmospheric conditions and the value range of the non - linear coefficient of specific humidity γ v a number of the first - order series terms are selected as the approximate value for calculation, while ensuring the calculation accuracy and maintaining the high - precision description ability of the model for the non - linear effect of specific humidity.

5. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact water vapor pressure differential equation according to claim 1, wherein In step SS3, meteorological data of the tropical cyclone are obtained based on reanalysis data, satellite remote sensing observation data, and / or surface meteorological observation data. The meteorological data include wind field data, vertical velocity data, geopotential height data, specific humidity data, relative humidity data, and mean sea level pressure data with a time resolution of not less than 6 hours, a spatial resolution of not less than 1°×1°, and a vertical range from 1000 hPa to 50 hPa. Its time range covers the complete life history process of the tropical cyclone from generation to dissipation. The meteorological data from different sources are fused through quality control and data assimilation to ensure the spatio-temporal continuity and physical consistency of the data.

6. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 5, wherein In step SS3, the specific steps for dynamically tracking the tropical cyclone include: determining the intensity and position of the tropical cyclone center at each time step using the lowest mean sea level pressure value and its position at each time step, constructing the movement trajectory of the tropical cyclone through the time series of the tropical cyclone center position, establishing a dynamic coordinate system with the tropical cyclone center as the origin, and constructing a dynamic calculation area centered on the longitude and latitude coordinates of the tropical cyclone center. The position of the calculation area is adjusted in real time as the tropical cyclone center moves to ensure that the calculation area always covers the core area of the tropical cyclone.

7. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 6, characterized in that, In step SS3, dynamically tracking the tropical cyclone further includes mapping the relevant meteorological physical quantities at each time step to the dynamic coordinate system using an interpolation method, and combining the changes in the dynamic tracking path to update the spatial distribution characteristics of the physical quantities in real time to ensure the dynamic consistency between the evaluation of water vapor mass change and the actual physical field.

8. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 7, characterized in that, In step SS4, the input layer of the multi-layer convolutional neural network model divides the dynamic calculation area into three height levels in the atmospheric vertical direction for data reception. Among them, the lower layer takes the meteorological element field in the range of 1000 - 700 hPa to analyze the water vapor transport and humidification process in the lower atmosphere; the middle layer takes the meteorological element field in the range of 700 - 500 hPa to capture the water vapor fluctuation characteristics in the middle atmosphere; The upper layer takes the meteorological element field in the range of 500 - 50 hPa to evaluate the drying effect in the upper atmosphere.

9. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 8, characterized in that, In step SS4, the output layer of the multi-layer convolutional neural network model uses the mean square error loss function to measure the difference between the predicted value of the multi-layer convolutional neural network model and the water vapor mass change calculated according to the calculation model result in step SS2, and introduces the absolute value or square of the difference between the two as an additional loss term to optimize and adjust the loss function to improve the model prediction accuracy and consistency with the calculation model result.

10. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 9, characterized in that, In step SS4, according to the life cycle development stage of the tropical cyclone, the water vapor mass change characteristics at different height levels in the development stage, intensification stage, and mature stage of the tropical cyclone are calculated and dynamically evaluated in layers to reveal the humidity change law of the tropical cyclone at different development stages.

11. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 10, characterized in that, In step SS4, the hierarchical calculation and evaluation of the water vapor mass change in different altitude layers further includes the analysis of the convergence effect of the water vapor mass change in the lower layer. By calculating the horizontal convergence rate through the humidification rate of the water vapor mass change in the lower layer, the contribution of the lower layer humidity change to the intensification of the tropical cyclone intensity is evaluated, and the influence of the lower layer humidity accumulation on the overall humidity distribution is analyzed in combination with the changes in the middle and upper layers.

12. The method for calculating and evaluating the change of water vapor mass of tropical cyclones based on machine learning and integrating the exact water vapor pressure differential equation according to claim 11, wherein, In step SS4, the hierarchical calculation and evaluation of the water vapor mass change in different altitude layers further includes the analysis of the fluctuation characteristics of the water vapor mass change in the middle layer. By calculating the fluctuation intensity of the water vapor mass change in the middle layer of the air column per unit area, the influence of the middle layer humidity fluctuation on the cyclone structure and convective activities is evaluated, and the spatio-temporal coupling relationship between the middle layer humidity fluctuation and the lower layer humidification is analyzed.

13. The method for calculating and evaluating the change in water vapor mass of a tropical cyclone based on machine learning and integrating the exact differential equation of water vapor pressure according to claim 1, wherein In step SS5, the calculation and evaluation results of the water vapor mass change further include the contribution rate analysis of the non - linear effect of specific humidity. By comparing the exact differential equation of water vapor pressure based on the introduction of the non - linear term of specific humidity with the differential equation of water vapor pressure without the introduction of the non - linear term, calculate the contribution rates of the non - linear term of specific humidity to the low - layer humidification rate, the middle - layer fluctuation intensity, and the high - layer drying rate, and comprehensively evaluate the impact of the non - linear effect of specific humidity on the regulation of the overall humidity distribution.

14. A computer program product, comprising computer instructions, characterized in that, The computer instructions are used to execute the method for calculating and evaluating the water vapor mass change of a tropical cyclone based on machine learning and integrating the exact water vapor pressure differential equation according to any one of claims 1 to 13.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for calculating and evaluating the water vapor mass change of a tropical cyclone based on machine learning and integrating the exact water vapor pressure differential equation according to any one of claims 1 to 13.

Images