A typhoon MPI multi-factor collaborative calculation evaluation method, medium and program product

Abstract

The present invention discloses a typhoon MPI multi-factor collaborative calculation evaluation method, medium and program product, which belongs to the field of typhoon intensity evaluation numerical calculation technology, including: collecting sea-air dynamic thermal field and typhoon core structure data, generating a multi-source data fusion feature matrix; extracting multi-factor key features related to typhoon enhancement; constructing a typhoon MPI physical calculation model, calculating MPI and its occurrence time; introducing an attention mechanism based on a convolutional neural network, constructing a machine learning model for typhoon MPI multi-factor collaborative evaluation, and training through optimizing the loss function by physical constraint terms; finally, combining the results of the physical calculation model and the machine learning model, using a weighted fusion strategy to dynamically correct the results, and performing uncertainty analysis and error evaluation. The present invention realizes the collaborative application of multi-source data fusion and physical mechanism and machine learning, and significantly improves the accuracy of calculation evaluation in the rapid enhancement stage of typhoons.

Description

Technical Field

[0001] The present invention belongs to the technical field of typhoon intensity calculation and evaluation and numerical calculation, and specifically provides a typhoon MPI multi-factor collaborative calculation and evaluation method, medium and program product, which are used to realize the calculation and evaluation of typhoon MPI and provide a scientific basis for disaster prevention and mitigation and meteorological risk assessment. Background Art

[0002] Typhoons are extremely destructive natural disasters. Accurately assessing their intensity, especially their maximum potential intensity (MPI) and the time of their occurrence, is of great significance for formulating effective disaster prevention and mitigation measures. MPI refers to the theoretical maximum intensity that a typhoon can reach under given environmental conditions, which reflects the upper limit of the potential destructive power of a typhoon. However, due to the high nonlinearity and complexity of typhoon structure and physical processes, existing computational assessment methods still face many challenges in practical applications, especially in the assessment of MPI and its occurrence time, which have significant limitations.

[0003] Traditional MPI calculation and evaluation methods mainly rely on thermodynamic and kinetic model frameworks, such as the Carnot heat engine model and the wind-induced surface heat exchange mechanism. The Carnot heat engine model estimates the MPI by calculating the temperature difference between the sea surface temperature and the outflow layer temperature, but it ignores the influence of the core processes inside the typhoon, such as eyewall replacement. The wind-induced surface heat exchange mechanism emphasizes the positive feedback mechanism between the near-surface wind speed and the sea surface enthalpy flux, but it is still insufficient in dealing with turbulent losses and environmental constraints under high wind speed conditions. For example, MPI is often overestimated under strong typhoon conditions. The Energy Growth Rate (EGR) theory has been introduced into typhoon intensity assessment. Its core idea is to calculate and evaluate the intensity change by calculating the balance between energy input and output. A positive EGR value indicates that the typhoon is strengthening, and a zero EGR indicates that the typhoon has reached its maximum intensity. However, the EGR calculation relies on accurate sea-air flux data, which is difficult to obtain. In addition, the EGR theory mainly considers the energy exchange between the typhoon and the environment, ignoring the impact of changes in the internal structure of the typhoon, such as eyewall replacement, convective bursts (CBs) and warm core structure evolution (WC), which limits its evaluation accuracy.

[0004] Statistical models and empirical formulas are still widely used in current typhoon intensity assessment, but these methods are usually based on linear fitting of historical data, which makes it difficult to capture the complex synergistic effects of multiple factors. For example, many statistical models regard the moment when the energy growth rate is zero as the time when the typhoon reaches MPI, but actual observations show that the maximum intensity usually lags behind this moment. The mechanism of this lag effect has not yet been reasonably explained in existing methods. In addition, statistical models have poor adaptability to assessments in new environments, and there is great uncertainty in the assessment of rapid intensification stages. In recent years, the rise of machine learning technology has provided a new solution for typhoon intensity assessment. Compared with traditional methods, machine learning can better capture the nonlinear characteristics of typhoon intensity changes and has stronger generalization capabilities. By combining machine learning algorithms with multi-source data fusion technology, a more sophisticated and accurate typhoon MPI assessment model can be constructed.

[0005] In summary, the existing typhoon MPI calculation and evaluation methods have problems such as insufficient calculation and evaluation accuracy, low reliability, and failure to fully consider the synergy of multiple factors, which makes it difficult to meet the actual needs of disaster prevention and mitigation. Therefore, how to comprehensively consider multiple influencing factors and combine machine learning algorithms and data fusion technology to develop a more accurate and reliable typhoon MPI calculation and evaluation method is a technical problem that needs to be solved urgently. Summary of the invention

[0006] 1. Purpose of the invention

[0007] In view of the above-mentioned defects and shortcomings in the prior art and to solve at least one of the above-mentioned and other technical problems, the present invention aims to provide a typhoon MPI multi-factor collaborative calculation evaluation method, medium and program product. Based on the multi-source data of the sea-air dynamic and thermal field and the typhoon inner core structure in the typhoon generation area and its surrounding areas, a collaborative evaluation model combining multi-source data fusion and machine learning is constructed to achieve accurate evaluation of typhoon MPI and its occurrence time, providing a scientific basis for meteorological disaster warning and risk assessment.

[0008] (II) Technical solution

[0009] In order to achieve the purpose of the invention and solve the technical problems, the present invention adopts the following technical solutions:

[0010] The first invention object of the present invention is to provide a typhoon MPI multi-factor collaborative calculation and evaluation method for calculating and evaluating the maximum possible intensity of a typhoon and its occurrence time, and improving the evaluation accuracy and reliability during the rapid intensification of a typhoon. The method comprises the following steps:

[0011] SS1. Collection and fusion of typhoon multi-source data

[0012] Collect multi-source data including sea-air dynamic thermal field and typhoon core structure data in and around the typhoon generating area, perform quality control and standardization on the collected multi-source data, integrate the pre-processed multi-source data through data fusion technology, and generate a multi-source data fusion feature matrix;

[0013] SS2. Extraction of multi-factor features related to typhoon MPI

[0014] Based on the multi-source data pre-processed in step SS1, extract the multi-factor key features closely related to typhoon intensification and its MPI, including at least: calculating the typhoon energy growth rate EGR and its characteristics of change over time, and determining the moment when it changes from positive to zero t mpi And the corresponding intensity V mpi ; Count the time-varying characteristics of the number of strong convective grid points in the typhoon eyewall region and determine the time when the convective bursts (CBs) reach their peak t CBs ; Calculate the difference between the temperature in the typhoon eyewall area and the ambient temperature, quantify the intensity of the warm heart and its characteristics of change over time, and determine the time when the intensity of the warm heart reaches its peak t WC ;

[0015] SS3. Establishing a typhoon MPI physical calculation model

[0016] Based on the multi-factor key features extracted in step SS2, and according to the coupling relationship of typhoon energy growth-convective outbreak-warm core structure-maximum intensity, a typhoon MPI physical calculation model is established:

[0017]

[0018] in, MPI phy , t phy are the typhoon MPI and its appearance time obtained based on the physical calculation model, is the average intensification rate of the typhoon during its rapid intensification stage, Δt for t mpi Relative to t phy time delay;

[0019] SS4. Constructing a machine learning model for typhoon MPI multi-factor collaborative evaluation

[0020] Construct and train a machine learning model for the multi-factor collaborative evaluation of typhoon MPI. The model input layer is used to receive the multi-source data fusion feature matrix generated in step SS1. The hidden layer uses multiple convolutional layers and pooling layers to extract multi-scale spatiotemporal evolution features including at least EGR, convective bursts, and warm-heart structures, and uses a fully connected layer for feature fusion. At the same time, an attention mechanism is introduced to at least capture the key moments when EGR turns from positive to negative, when the convective burst reaches a peak, and when the warm-heart intensity is the largest. The output layer is used to evaluate the typhoon MPI and its occurrence time, and the physical calculation results of the typhoon MPI in step SS3 are used as training data labels, and a loss function containing physical constraints is used for model training and parameter optimization.

[0021] SS5. Output the calculation and evaluation results of typhoon MPI and its appearance time

[0022] Based on the machine learning model of typhoon MPI multi-factor collaborative evaluation constructed in step SS4, typhoon-related real-time multi-source observation data are input to obtain the initial evaluation results of typhoon MPI and its occurrence time. Combined with the physical calculation results in step SS3, the initial evaluation results are corrected and optimized through a weighted fusion strategy. At the same time, uncertainty analysis and error evaluation are performed on the corrected and optimized results. Finally, a complete result including calculated evaluation values, confidence intervals and physical consistency evaluation reports is output.

[0023] The second inventive object of the present invention is to provide a computer program product, including computer instructions, which are used to execute the above-mentioned typhoon MPI multi-factor collaborative calculation and evaluation method of the present invention.

[0024] The third inventive 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 typhoon MPI multi-factor collaborative calculation and evaluation method is implemented.

[0025] (III) Technical Effect

[0026] Compared with the prior art, the typhoon MPI multi-factor collaborative calculation evaluation method, medium and program product of the present invention have the following beneficial and significant technical effects:

[0027] (1) This invention solves the error accumulation problem caused by relying on a single theoretical model in traditional typhoon calculation and evaluation methods by introducing a collaborative computing framework of machine learning algorithms and physical computing models, combining multi-factor characteristics such as typhoon energy growth rate EGR, convective burst and warm core intensity. This invention achieves accurate characterization of the rapid intensification stage of typhoons, especially accurate evaluation of the time lag effect of MPI, by dynamically adjusting the weight of energy growth rate and analyzing the spatiotemporal evolution of key characteristics, thus improving the defect of ignoring the dynamic characteristics of convective burst and warm core in traditional calculation and evaluation methods.

[0028] (2) The present invention establishes a physical calculation model by constructing the coupling relationship of typhoon energy growth-convective burst-warm core structure-maximum intensity, so that the evaluation results can meet the physical consistency requirements. At the same time, the present invention introduces the attention mechanism and multi-task learning framework of the machine learning model to capture the key moments when EGR turns from positive to negative, the convective burst reaches its peak, and the warm core intensity is the largest, thereby realizing dynamic modeling and accurate evaluation of the rapid enhancement process of typhoon intensity. This collaborative evaluation model overcomes the problem of insufficient representation ability of complex nonlinear relationships in traditional statistical methods.

[0029] (3) The present invention combines the weighted fusion strategy of physical computing model and machine learning model to dynamically correct the typhoon MPI and its occurrence time obtained by evaluation, ensuring the accuracy and robustness of the evaluation results. At the same time, the present invention introduces the Bayesian method to analyze the uncertainty of the correction results, quantify the confidence interval of the evaluation results, and provide a scientific basis for typhoon disaster warning and emergency management decision-making.

[0030] (4) The present invention constructs a collaborative computing and evaluation framework based on the typhoon core characteristics and the external environmental field. The present invention is not only applicable to the intensity calculation and evaluation of typhoons in the rapid intensification stage, but can also be extended to the intensity evolution research of other complex meteorological systems, providing technical support for the numerical forecasting and scientific research of tropical cyclones, and has important theoretical significance and practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Shown is a schematic diagram of the implementation process of the typhoon MPI multi-factor collaborative calculation and evaluation method of the present invention. DETAILED DESCRIPTION

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Example 1

[0034] Figure 1 FIG. 1 is a schematic diagram of the implementation process of the typhoon MPI multi-factor collaborative calculation and evaluation method of the present invention. According to the embodiment disclosed in the present invention, Figure 1 As shown, the typhoon MPI multi-factor collaborative calculation and evaluation method of the present invention comprehensively considers the sea-air dynamic thermal field and typhoon core structure data in the typhoon generation area and its surroundings, and constructs a collaborative evaluation model combining multi-source data fusion and machine learning to achieve accurate evaluation of the typhoon's maximum possible intensity MPI and its occurrence time. Specifically, the method mainly includes the following steps:

[0035] SS1. Collection and fusion of typhoon multi-source data

[0036] First, through high-precision observation and historical data integration, multi-source data that affect typhoon MPI are collected, including sea-air dynamic thermal field data and typhoon core structure data. After that, the collected multi-source data are quality controlled (such as outlier removal and spatiotemporal alignment processing) and standardized (such as minimum-maximum standardization and Z-score standardization), and pre-processed multi-source data are integrated through data fusion technologies such as principal component analysis (PCA), spatiotemporal variation assimilation and deep nested interpolation to generate a multi-source data fusion feature matrix, providing high-quality input for subsequent feature extraction.

[0037] SS2. Extraction of multi-factor features related to typhoon MPI

[0038] Based on the multi-source data pre-processed in step SS1, extract the multi-factor key features closely related to typhoon intensification and its MPI, including at least: calculating the typhoon energy growth rate EGR and its characteristics of change over time, and determining the moment when it changes from positive to zero t mpi And the corresponding intensity V mpi ; Count the time-varying characteristics of the number of strong convective grid points in the typhoon eyewall region and determine the time when the convective bursts (CBs) reach their peak t CBs ; Calculate the difference between the temperature in the typhoon eyewall area and the ambient temperature, quantify the intensity of the warm core and its characteristics of change over time, and determine the time when the intensity of the warm core (WC) reaches its peak t WC These multi-factor characteristics reflect the key nonlinear dynamic and thermodynamic characteristics of the typhoon's rapid intensification stage.

[0039] SS3. Establishing a typhoon MPI physical calculation model

[0040] Based on the multi-factor key features extracted in step SS2, and according to the coupling relationship of typhoon energy growth-convective outbreak-warm core structure-maximum intensity, a typhoon MPI physical calculation model is established:

[0041]

[0042] in, MPI phy , t phy are the typhoon MPI and its appearance time obtained based on the physical calculation model, is the average intensification rate of the typhoon during its rapid intensification stage, Δt for t mpi Relative to t phy This model combines the characteristics of multiple factors to describe the rapid intensification process of typhoons, providing physical constraints and benchmarks for subsequent machine learning models.

[0043] SS4. Constructing a machine learning model for typhoon MPI multi-factor collaborative evaluation

[0044] Taking the multi-source data fusion feature matrix generated in step SS1 as input, a machine learning model for multi-factor collaborative evaluation of typhoon MPI is constructed and trained. The hidden layer uses multiple convolutional layers and pooling layers to extract multi-scale spatiotemporal evolution features including at least EGR, convective bursts and warm-heart structures, and uses a fully connected layer for feature fusion. At the same time, an attention mechanism is introduced to capture at least the key moment features when EGR turns from positive to negative, when the convective burst reaches a peak, and when the warm-heart intensity is the largest. The output layer is used to evaluate the typhoon MPI and its occurrence time, and the physical calculation results of the typhoon MPI in step SS3 are used as training data labels, and a loss function containing physical constraints is used for model training and parameter optimization.

[0045] SS5. Output the calculation and evaluation results of typhoon MPI and its appearance time

[0046] Based on the machine learning model of typhoon MPI multi-factor collaborative evaluation constructed in step SS4, typhoon-related real-time multi-source observation data are input to obtain the initial evaluation results of typhoon MPI and its occurrence time. Combined with the physical calculation results in step SS3, the initial evaluation results are corrected and optimized through a weighted fusion strategy. At the same time, uncertainty analysis and error evaluation are performed on the corrected and optimized evaluation results. Finally, a complete result including calculated evaluation values, confidence intervals and physical consistency evaluation reports is output.

[0047] The embodiment of the present invention achieves accurate assessment of typhoon MPI and its occurrence time by introducing multi-source data fusion and machine learning technology. The coordinated application of physical computing models and machine learning models overcomes the limitations of a single model and improves the accuracy and reliability of the assessment results. At the same time, the result output including uncertainty analysis provides a scientific basis for typhoon disaster warning and disaster prevention and mitigation, reflecting the significant technical advantages and practical application value of the present invention in the field of typhoon intensity assessment.

[0048] Example 2: Collection and fusion of typhoon multi-source data

[0049] Based on the above-mentioned embodiment 1, as a further refinement and supplement, this embodiment 2 provides a more detailed description of step SS1. By clarifying the specific application of data types, preprocessing methods and fusion technologies, the integrity and consistency of the collected data are ensured, providing a high-quality data foundation for subsequent feature extraction and model building, and improving the evaluation accuracy and reliability.

[0050] In an embodiment of the present invention, the collected sea-atmosphere dynamic thermal field and typhoon inner core structure data should mainly include sea temperature environment parameters, atmospheric environment parameters, wind field data and convective system parameters, wherein: sea temperature environment parameters at least include sea surface temperature, ocean heat capacity, sea temperature vertical profile, ocean mixed layer depth and sea surface enthalpy flux; atmospheric environment parameters at least include atmospheric temperature field, humidity field, air pressure field and atmospheric stability parameters; wind field data at least include wind speed close to the surface, maximum wind speed radius, radial wind field, tangential wind field and wind speed axis symmetry parameters; convective system parameters at least include convective cloud top brightness temperature, convective cloud top height, convective vertical velocity, convective outbreak area and warm core structure parameters.

[0051] Among the multiple sea-atmosphere dynamic thermal fields and typhoon inner core structure data that must be collected as mentioned above, the sea surface temperature is used to characterize the thermal conditions of the underlying ocean of the target typhoon, the ocean heat capacity is used to characterize the heat storage of the upper ocean water and reflect the sustainability of the ocean's thermal support for the typhoon, the sea temperature vertical profile is used to describe the temperature distribution from the sea surface to the deeper water in the sea area where the target typhoon is located, the ocean mixed layer depth is used to characterize the thickness of the mixed layer in the underlying ocean of the typhoon, and the sea surface enthalpy flux includes sensible heat flux and latent heat flux and is used to evaluate the total amount of heat energy obtained by the target typhoon from the ocean.

[0052] The atmospheric temperature field includes the outflow layer temperature and the middle and lower troposphere temperature. The humidity field includes the relative humidity and specific humidity of the target area and is used to describe the water vapor distribution in the target typhoon area. The pressure field includes the central minimum pressure and sea level pressure of the target typhoon and is used to characterize the intensity and development state of the typhoon. The atmospheric stability parameters include the convective effective potential energy and convective inhibition energy and are used to evaluate the possibility of convective outbreaks in the target area. The wind speed near the surface is used to calculate the energy dissipation rate of the target area. The maximum wind speed radius is used to reflect the structural characteristics of the target typhoon core area. The radial wind field and tangential wind field are used to describe the dynamic characteristics of the typhoon core and its interaction with the surrounding atmosphere. The wind speed axis symmetry parameter is used to measure the symmetry of the target typhoon core.

[0053] The convective cloud top brightness temperature is used to characterize the top temperature of the convective cloud cluster, the convective cloud top height is used to quantify the vertical development of the convective cloud cluster, the convective vertical velocity is used to quantify the intensity of the upward movement within the convective system, and the convective outbreak area is used to characterize the horizontal range of the convective system. The warm heart structure parameters include the warm heart height, the warm heart temperature difference and the warm heart thickness, among which the warm heart height is used to characterize the vertical height range dominated by the warm heart effect, the warm heart temperature difference is used to characterize the difference between the central temperature of the eyewall area within the warm heart height and the ambient temperature, and the warm heart thickness is used to characterize the extension range of the warm heart temperature difference on the horizontal plane.

[0054] In an embodiment of the present invention, when preprocessing and fusing multi-source data, the quality control of the collected multi-source data at least includes outlier removal, data interpolation and spatiotemporal alignment processing, etc. The standardization processing includes dimensionless processing of different physical quantities, using the minimum-maximum standardization method to map the data to the [0,1] interval, and using the Z-score standardization method to eliminate the dimension effect. When performing data fusion, the data fusion technology can use principal component analysis to extract the main information of multi-source data, use spatiotemporal variational assimilation methods to integrate observation data and simulation data, introduce Kalman filtering methods to deal with the uncertainty of dynamic data, and use deep nested interpolation algorithms to perform high-precision integration of multi-source data, integrate the preprocessed multi-source data into a unified multi-source data fusion feature matrix, and provide complete and consistent input for subsequent feature extraction. At the same time, the data fusion results contain multi-dimensional information on the power, thermal and structure of the typhoon, providing accurate support for the collaborative evaluation model.

[0055] Example 3: Extraction of multi-factor features related to typhoon MPI

[0056] As a further refinement and supplement based on the above-mentioned embodiments 1 and 2, this embodiment 3 further provides supplementary explanation and detailed description of the specific implementation of step SS2 in the above-mentioned typhoon MPI multi-factor collaborative calculation and evaluation method of the present invention.

[0057] In a preferred embodiment of the present invention, the calculation method of the typhoon energy growth rate EGR is: based on the typhoon thermodynamics and kinetics theory, the typhoon energy generation rate PE and the typhoon energy dissipation rate DE are calculated, and then the typhoon energy growth rate EGR is calculated based on the difference between PE and DE. The relevant algorithm formula is:

[0058]

[0059] in, r is the radial distance from the typhoon center, r 0 is the radius of the outer circulation boundary of the typhoon, r m is the maximum wind speed radius of the typhoon; e is the Carnot engine efficiency and ，T s is the sea surface temperature, T 0 is the outflow layer temperature; r is the air density near the surface, is the near-surface wind speed; is the sea surface saturation enthalpy, is the near-surface atmospheric enthalpy; C k is the surface enthalpy exchange coefficient, C D is the drag coefficient.

[0060] and, C k The value of is affected by the wind speed at 10 meters V 10 Dynamically adjust according to the changes of V 10 When ≤10m / s, C k The value is 0.001. V 10 ≥70m / s, C k The value is 0.003, when 10m / s< V 10 <70m / s, C k The value of increases with the wind speed; C D The value of is a piecewise function. V 10 When <30m / s, C D The value of increases linearly with wind speed to 0.002. V 10 ≥30m / s, C DKeep a constant value of 0.002; when calculating EGR, r 0 is 300-500km, r m Based on wind field data, the EGR calculation results are updated in real time to reflect the dynamic characteristics of typhoon intensity changes.

[0061] Preferably, in step SS2, the method for counting the number of strong convection grid points in the typhoon eyewall area is: identifying, tracking and counting the number of grid points with an average vertical velocity greater than 1m / s at the 2-11km altitude layer in the typhoon eyewall area, establishing a time evolution sequence of the number of strong convection grid points, extracting the periodic characteristics of the sequence using a wavelet analysis method, and determining the time when the convective burst reaches its peak in combination with a maximum detection algorithm. t CBs At the same time, the spatial distribution characteristics of severe convective areas are calculated, and the contribution of convective outbreaks to the enhancement of typhoon intensity is dynamically monitored based on the time and spatial distribution characteristics of convective outbreaks.

[0062] In addition, the following conditions must be met simultaneously for the identification and judgment of convective bursts: the convective cloud top brightness temperature is less than -70°C, the convective duration is greater than 3 hours, and the convective area is greater than 100 km. 2 . And through the algorithm formula To quantify the intensity of convective bursts, CBI is the convective burst index. n is the total number of convective bursts, i is an integer and i =1,2,…, n , w i For the i The weight coefficient of the convective burst is ΔT i For the i The convective cloud top brightness temperature anomaly of the convective outburst is A i For the i The area of ​​the convective region of a convective outbreak, Δt i For the i The convection duration of a convective burst.

[0063] Preferably, in step SS2, the method for calculating the difference between the temperature inside the typhoon eyewall and the ambient temperature is as follows: first, determine the spatial range of the typhoon eyewall area, and calculate the average temperature profiles of each layer at an altitude of 1-18 km in the area; then, take the area 500 km away from the typhoon center as the environmental field, and calculate the average profile of the ambient temperature; finally, calculate the difference between the two temperature profiles, obtain the vertical distribution, horizontal distribution and time evolution characteristics of the warm heart intensity, and determine the time when the warm heart intensity reaches its peak through time series analysis. tWC , and analyze the delayed response relationship between the warm-heart enhancement and the rapid intensification of the typhoon. Further, in step SS2, the calculation of the warm-heart intensity also includes calculating the warm-heart index through an algorithm formula To quantify the warm intensity characteristics of the typhoon eyewall, z is the height, z 1=1km, z 2 = 18 km, ΔT ( z ) is the difference between the temperature of the typhoon eyewall area and the surrounding environment in the range of 1-18 km, and is based on the warm heart evolution index Monitor the dynamic changes of warm heart intensity over time, where t For time.

[0064] This Example 3 describes in detail the method for extracting multi-factor characteristics related to typhoon MPI, including EGR calculation, convective burst quantification and warm core intensity analysis. These characteristics reflect the key physical mechanisms and dynamic thermodynamic characteristics in the rapid intensification process of typhoons, and provide comprehensive and dynamic feature data support for subsequent physical calculation models and machine learning models.

[0065] Example 4: Establishing a typhoon MPI physical calculation model

[0066] Based on the above embodiments 1 to 3, as a further refinement and supplement, this embodiment 4 further provides additional explanations on the specific implementation of step SS3 to enhance the description capability and evaluation accuracy of the physical calculation model for the rapid intensification stage of a typhoon.

[0067] In some preferred embodiments, in step SS3, the average intensification rate of the typhoon rapid intensification stage is calculated based on the time series data of the typhoon intensity change:

[0068] First, through the time series data of wind speed changes during the typhoon life cycle, the time period with a wind speed change rate greater than 5.5m / s / 6h is identified as the rapid intensification stage, and the start and end time of the rapid intensification stage is determined to provide time period division for the subsequent calculation of the average intensification rate;

[0069] Then, in the rapid enhancement stage, the piecewise linear fitting method was used to extract the trend characteristics of the intensity change and calculate the average enhancement rate in each time period. V 0;

[0070] Finally, combined with the sea surface temperature and vertical wind shear environmental conditions, the average enhancement rate is corrected to obtain the final average enhancement rate ,in α , β is the correction coefficient based on historical data fitting, SST is the sea surface temperature, VWS is the vertical wind shear intensity.

[0071] Example 5: Constructing a machine learning model for typhoon MPI multi-factor collaborative evaluation

[0072] Based on the above-mentioned embodiments 1 to 4, as a further refinement and supplement, this embodiment 5 provides supplementary explanation and detailed description for the specific implementation of step SS4, clarifying the structural design of the machine learning model, input feature selection, training process and physical constraint fusion method.

[0073] In some preferred embodiments, in step SS4 of the present invention, the machine learning model may adopt a convolutional neural network model, and the input feature is the multi-source data fusion feature matrix generated in step SS1, and the shape is N × M × C , where N is the number of time steps, M is the number of spatial grids, and C is the number of feature channels. The model input layer uses a 3×3 convolution kernel to extract local features to ensure the effective capture of spatiotemporal information. It contains 4 convolution blocks in the middle, each of which consists of two convolution layers and a maximum pooling layer. The number of channels of the convolution layers is 32, 64, 128, and 256, respectively. The step size of the pooling layer is 2, which is used to extract features layer by layer and reduce the computational complexity. The fully connected layer contains 1024 neurons, and the ReLU activation function is used to improve the nonlinear fitting ability. Dropout regularization (dropout rate is 0.5) is introduced to prevent overfitting. The attention mechanism uses a self-attention structure to calculate the correlation weights of features at different time steps, focusing on the key moments when EGR turns from positive to negative, the convective outbreak reaches its peak, and the warm heart intensity reaches its maximum value, and enhances the model's response to important time points by weighted features.

[0074] Further preferably, the input features of the machine learning model also include the vertical wind shear distribution VWS, atmospheric humidity gradient and / or convective effective potential energy CAPE and their spatiotemporal evolution characteristics reflecting the external environmental field of the typhoon, and the spatiotemporal correlation information of the environmental field characteristics is extracted through multi-layer convolution of the convolutional neural network. At the same time, the time series model is combined to capture the dynamic changes of the internal structural characteristics of the typhoon, so as to enhance the physical consistency and generalization ability of the machine learning model for typhoon MPI assessment.

[0075] A multi-task learning mechanism based on physical constraints is introduced in the training process of the machine learning model. The main task is used to evaluate the MPI of the typhoon and its occurrence time, and the auxiliary task is used to evaluate the evolution trend of the energy growth rate EGR, the peak intensity of convective outbreaks and the warm heart intensity. The model's learning ability of the synergistic relationship of multiple factors is enhanced by sharing the weights of the hidden layer, and the physical rationality and dynamic stability of the model evaluation results are ensured by the loss function containing physical constraints.

[0076] As a preferred option, the algorithm formula of the loss function is: ,in: L is the loss function, w m , w p , w r are weight coefficients respectively and the optimal value is determined through cross validation; L mse is the mean square error constraint term, which is used to measure the difference between the evaluation value and the true value; L phy It is a physical constraint term used to ensure that the evaluation results satisfy the energy conservation and typhoon dynamics equations; L reg is a regularization constraint term used to prevent the model from overfitting.

[0077] This embodiment comprehensively improves the evaluation accuracy and physical consistency of typhoon MPI and its occurrence time by designing a typhoon MPI multi-factor collaborative evaluation model based on convolutional neural networks, combining attention mechanisms, multi-task learning, and physical constraints. The model's innovations in input feature selection, structural design, and training optimization provide solid technical support for the correction and practical application of subsequent evaluation results.

[0078] Example 6: Outputting the calculation and evaluation results of typhoon MPI and its appearance time

[0079] Based on the above embodiments 1 to 5, as a further refinement and supplement, this embodiment 6 further provides supplementary explanation and detailed description of the specific implementation of step SS5.

[0080] In some preferred embodiments, in step SS5 of the present invention, an ensemble learning method is used to weightedly fuse the evaluation results of the machine learning model with the physical calculation results. The algorithm formula for weighted fusion is: , ,in t f is the calculated evaluation value of the MPI appearance time after fusion, t phy , t ml MPI epoch values ​​obtained for physical computing and machine learning models, respectively; MPI f Calculate the maximum possible intensity after fusion, MPI phy 、MPI ml The maximum possible intensity values ​​obtained for the physical calculation and machine learning models, respectively; w ml and w phyare weight coefficients and w ml + w phy =1, and their values ​​are determined by minimizing the error in the training data.

[0081] Preferably, when performing error assessment on the corrected and optimized typhoon MPI assessment results, the deviation and uncertainty of the assessment results are assessed through comparative verification with historical typhoon MPI observation data, deviation analysis based on spatial distribution and / or time series consistency detection methods, and a post-processing module integrating machine learning and physical calculation is constructed to ensure that the assessment results can meet the requirements for typhoon intensity assessment accuracy in actual application scenarios.

[0082] This embodiment significantly improves the accuracy and reliability of typhoon MPI and its occurrence time assessment by weighted fusion of the assessment results of the physical calculation model and the machine learning model, and combines dynamic correction and error assessment methods. At the same time, based on the construction of uncertainty analysis and post-processing modules, it ensures that the assessment results can meet the needs of actual application scenarios.

[0083] Through the above embodiments, the purpose of the present invention is fully and effectively achieved. Those skilled in the art can understand that the present invention includes but is not limited to the contents described in the drawings and the above specific embodiments. Although the present invention has been described with respect to the most practical and preferred embodiments currently considered, it should be understood that the present invention is not limited to the disclosed embodiments, and any modification that does not deviate from the functional and structural principles of the present invention will be included in the scope of the claims.

Claims

1. A typhoon MPI multi-factor collaborative calculation and evaluation method, characterized in that: The method comprises the following steps: SS1. Collect multi-source data including sea-air dynamic thermal field and typhoon core structure data in and around the typhoon generating area, perform quality control and standardization on the collected multi-source data, integrate the pre-processed multi-source data through data fusion technology, and generate a multi-source data fusion feature matrix; SS2. Calculate the typhoon energy growth rate EGR and its characteristics over time, and determine the moment when it changes from a positive value to zero t mpi And the corresponding intensity V mpi ; Count the time-varying characteristics of the number of strong convection grid points in the typhoon eyewall area and determine the time when the convection outbreak reaches its peak t CBs ; Calculate the difference between the temperature in the typhoon eyewall area and the ambient temperature, quantify the intensity of the warm heart and its characteristics of change over time, and determine the time when the intensity of the warm heart reaches its peak t WC ; SS3. Based on the extracted multi-factor key features and the coupling relationship between typhoon energy growth-convective outbreak-warm core structure-maximum intensity, a typhoon MPI physical calculation model is established: in, MPI phy , t phy are the typhoon MPI and its appearance time obtained based on the physical calculation model, is the average intensification rate of the typhoon during its rapid intensification stage, Δt for t mpi Relative to t phy time delay; SS4. Construct and train a machine learning model for the multi-factor collaborative evaluation of typhoon MPI. The model input layer is used to receive the multi-source data fusion feature matrix. The hidden layer uses multiple convolutional layers and pooling layers to extract multi-scale spatiotemporal evolution features including at least EGR, convective bursts and warm-heart structures, and uses a fully connected layer for feature fusion. At the same time, an attention mechanism is introduced to capture at least the key moments when EGR turns from positive to negative, when the convective burst reaches its peak, and when the warm-heart intensity is the largest. The output layer is used to evaluate the typhoon MPI and its occurrence time, and the physical calculation results of the typhoon MPI in step SS3 are used as training data labels. A loss function containing physical constraints is used for model training and parameter optimization. SS5. Based on the machine learning model of typhoon MPI multi-factor collaborative evaluation, typhoon-related real-time multi-source observation data are input to obtain the initial evaluation results of typhoon MPI and its occurrence time, and the initial evaluation results are corrected and optimized through a weighted fusion strategy in combination with the physical calculation results. At the same time, uncertainty analysis and error evaluation are performed on the corrected and optimized results, and the final output includes a complete result of calculated evaluation value, confidence interval and physical consistency evaluation.

2. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1 is characterized in that: In step SS1, the sea-air dynamic thermal field and typhoon inner core structure data of the typhoon generation area and its surroundings include sea temperature environment parameters, atmospheric environment parameters, wind field data and convection system parameters, wherein: the sea temperature environment parameters at least include sea surface temperature, ocean heat capacity, sea temperature vertical profile, ocean mixed layer depth and sea surface enthalpy flux; the atmospheric environment parameters at least include atmospheric temperature field, humidity field, air pressure field and atmospheric stability parameters; the wind field data at least include wind speed close to the surface, maximum wind speed radius, radial wind field, tangential wind field and wind speed axis symmetry parameters; the convection system parameters at least include convective cloud top brightness temperature, convective cloud top height, convective vertical velocity, convective outbreak area and warm core structure parameters.

3. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1 is characterized in that: In step SS1, the quality control of the collected multi-source data at least includes outlier removal, data interpolation and spatiotemporal alignment processing. The standardization processing includes dimensionless processing of different physical quantities, mapping the data to the [0,1] interval using the minimum-maximum standardization method, and using the Z-score standardization method to eliminate the dimension effect; the data fusion technology uses principal component analysis, spatiotemporal variational assimilation, Kalman filtering and / or deep nested interpolation algorithm to integrate the pre-processed multi-source data into a unified multi-source data fusion feature matrix.

4. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1 is characterized in that: In step SS2, the method for calculating the typhoon energy growth rate EGR is: firstly, the typhoon energy generation rate PE and the typhoon energy dissipation rate DE are calculated based on thermodynamics and kinetic theory, and then the typhoon energy growth rate EGR is calculated based on the difference between the two: in, r is the radial distance from the typhoon center, r 0 is the radius of the outer circulation boundary of the typhoon. r m is the maximum wind speed radius of the typhoon; ε is the Carnot engine efficiency and , T s is the sea surface temperature, T 0 is the outflow layer temperature; ρ is the air density near the surface, is the near-surface wind speed; is the sea surface saturation enthalpy, is the near-surface atmospheric enthalpy; C k is the surface enthalpy exchange coefficient, C D is the drag coefficient.

5. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 4 is characterized in that: C k The value of is affected by the wind speed at 10 meters V 10 Dynamically adjust according to the changes of V 10 When ≤10m / s, C k The value is 0.

001. V 10 ≥70m / s, C k The value is 0.003, when 10m / s< V 10 <70m / s, C k The value of increases with the wind speed; C D The value of is a piecewise function. V 10 When <30m / s, C D The value of increases linearly with wind speed to 0.

002. V 10 ≥30m / s, C D Keep a constant value of 0.002; when calculating EGR, r 0 is 300-500km, r m Based on wind field data, the EGR calculation results are updated in real time to reflect the dynamic characteristics of typhoon intensity changes.

6. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1 is characterized in that: In step SS2, the method for counting the number of strong convection grid points in the typhoon eyewall area is as follows: identify, track and count the number of grid points with an average vertical velocity greater than 1m / s in the 2-11km altitude layer in the typhoon eyewall area, establish a time evolution sequence of the number of strong convection grid points, use wavelet analysis to extract the periodic characteristics of the sequence, and combine the maximum detection algorithm to determine the time when the convection burst reaches the peak. t CBs At the same time, the spatial distribution characteristics of severe convective areas are calculated, and the contribution of convective outbreaks to the enhancement of typhoon intensity is dynamically monitored based on the time and spatial distribution characteristics of convective outbreaks.

7. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 6 is characterized in that: The identification of convective bursts requires that the convective cloud top brightness temperature is less than -70°C, the convective duration is greater than 3 hours, and the convective area is greater than 100 km. 2 , and through the algorithm formula To quantify the intensity of convective bursts, CBI is the convective burst index. n is the total number of convective bursts, i is an integer and i =1,2,…, n , w i For the i The weight coefficient of the convective burst is ΔT i For the i The convective cloud top brightness temperature anomaly of the convective outburst is A i For the i The area of ​​the convective region of a convective outbreak, Δt i For the i The convection duration of a convective burst.

8. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1 is characterized in that: In step SS2, the method for calculating the difference between the temperature inside the typhoon eyewall and the ambient temperature is as follows: determine the spatial range of the typhoon eyewall area, calculate the average temperature profiles of each layer at an altitude of 1-18km; take the area 500km away from the typhoon center as the environmental field, and calculate the average profile of the ambient temperature; calculate the difference between the two temperature profiles, obtain the vertical distribution, horizontal distribution and time evolution characteristics of the warm heart intensity, and determine the time when the warm heart intensity reaches its peak through time series analysis t WC , and analyze the delayed response relationship between the warm-heart enhancement and the rapid intensification of the typhoon.

9. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 8 is characterized in that: The calculation of the warmth intensity also includes the calculation of the warmth index To quantify the warmth intensity characteristics, z is the height, z 1=1km, z 2 = 18 km, ΔT ( z ) is the difference between the temperature of the typhoon eyewall area and the surrounding environment in the range of 1-18 km, and is based on the warm heart evolution index Monitor the dynamic changes of warm heart intensity over time, where t For time.

10. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1, characterized in that: The average intensification rate of the typhoon during the rapid intensification phase in step SS3 Calculation is performed based on the time series data of typhoon intensity changes: through the time series data of wind speed changes during the typhoon life cycle, the time period with a wind speed change rate greater than 5.5m / s / 6h is identified as the rapid intensification stage; the trend characteristics of intensity changes are extracted using the piecewise linear fitting method, and the average intensification rate in each time period is calculated. V 0; Combined with the sea surface temperature and vertical wind shear environmental conditions, the average enhancement rate is corrected to obtain the final average enhancement rate ,in α , β is the correction coefficient based on historical data fitting, SST is the sea surface temperature, VWS is the vertical wind shear intensity.

11. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1, characterized in that: In step SS4, the machine learning model adopts a convolutional neural network model. The model input layer uses a 3×3 convolution kernel to extract local features. It contains 4 convolution blocks in the middle. Each convolution block consists of two convolution layers and a maximum pooling layer. The number of channels of the convolution layer is 32, 64, 128, and 256 respectively, and the step size of the pooling layer is 2; the fully connected layer contains 1024 neurons, using ReLU activation function and Dropout regularization; the attention mechanism uses a self-attention structure to calculate the correlation weights of features at different time steps.

12. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 11, characterized in that: The input features of the machine learning model also include the vertical wind shear distribution, atmospheric humidity gradient and / or convective effective potential energy and their spatiotemporal evolution characteristics that reflect the external environmental field of the typhoon, and the spatiotemporal correlation information of the environmental field characteristics is extracted through multi-layer convolution of the convolutional neural network. At the same time, the time series model is combined to capture the dynamic changes of the internal structural characteristics of the typhoon, so as to enhance the physical consistency and generalization ability of the machine learning model for typhoon MPI assessment.

13. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 12 is characterized in that: A multi-task learning mechanism based on physical constraints is introduced in the training process of the machine learning model, in which the main task is used to evaluate the typhoon MPI and its occurrence time, and the auxiliary task is used to evaluate the evolution trend of the energy growth rate EGR, the peak intensity of convective outbreaks and the warm heart intensity. The model's learning ability for the synergistic relationship of multiple factors is enhanced by sharing the weights of the hidden layer, and the physical rationality and dynamic stability of the evaluation results are ensured by a loss function containing physical constraints.

14. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 13, characterized in that: The algorithm formula of the loss function containing physical constraints is: ,in: L is the loss function, w m , w p , w r are weight coefficients respectively and the optimal value is determined through cross validation; L mse is the mean square error constraint term, which is used to measure the difference between the evaluation value and the true value; L phy It is a physical constraint term used to ensure that the evaluation results satisfy the energy conservation and typhoon dynamics equations; L reg is a regularization constraint term used to prevent the model from overfitting.

15. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1, characterized in that: In step SS5, an ensemble learning method is used to weightedly fuse the evaluation results of the machine learning model with the physical calculation results. The algorithm formula for weighted fusion is: , ,in t f is the calculated evaluation value of the MPI appearance time after fusion, t phy , t ml MPI epoch values ​​obtained for physical computing and machine learning models, respectively; MPI f MPI is the calculated evaluation value of the maximum possible intensity after fusion. phy 、MPI ml The maximum possible intensity values ​​obtained for the physical calculation and machine learning models, respectively; w ml and w phy are weight coefficients and w ml + w phy =1, and their values ​​are determined by minimizing the error in the training data.

16. The typhoon MPI multi-factor collaborative calculation and evaluation method according to claim 1, characterized in that: In step SS5, when the error of the typhoon MPI result after correction and optimization is evaluated, the deviation and uncertainty of the evaluation result are evaluated by comparison and verification with historical typhoon MPI observation data, deviation analysis based on spatial distribution and / or time series consistency detection method, and a post-processing module integrating machine learning and physical calculation is constructed to ensure that the evaluation result can meet the requirements for typhoon intensity assessment accuracy in actual application scenarios.

17. A computer program product comprising computer instructions, characterized in that The computer instructions are used to execute the typhoon MPI multi-factor collaborative calculation and evaluation method described in any one of claims 1 to 16.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the typhoon MPI multi-factor collaborative calculation and evaluation method described in any one of claims 1 to 16 is implemented.

Images