Method for estimating and predicting air-sea interface power exchange coefficient based on AI and multi-source fusion data, medium and program product

Abstract

The invention discloses a method for estimating and predicting a power exchange coefficient of an air-sea interface based on AI and multi-source fusion data, a medium and a program product, and aims to improve the accuracy of numerical simulation, forecasting, scientific research and the like of air-sea interaction. Comprehensively considering the wind speed at the height of 10 meters and other various meteorological and ocean variables, and reconstructing a calculation formula of an air-sea interface power exchange coefficient; collecting and preprocessing multi-source data, and constructing a multi-source fusion data set; constructing an AI model by adopting a deep learning algorithm, extracting spatial-temporal characteristics in the data, and estimating Cd '; by optimizing the model structure and parameters, the generalization ability and the prediction precision are improved; the prediction result is applied to numerical simulation of air-sea interaction, and the model is updated regularly based on new observation data. The method has the advantages of high precision, dynamic updating and wide applicability, and is suitable for the fields of weather forecast, ocean dynamics research and the like.

Description

Technical Field

[0001] The present invention belongs to the cross-technical field of marine meteorological forecasting and artificial intelligence (AI), and involves the calculation of key parameters in the numerical simulation and research of air-sea interaction. Specifically, it is a method based on artificial intelligence method and multi-source fusion data to estimate and predict the dynamic exchange coefficient of the air-sea interface. C d Methods, media and program products for improving C d The accuracy of the calculation is improved, thereby improving the accuracy of numerical simulations of sea-air interactions. Background Art

[0002] In the numerical simulation and research of air-sea interaction, the heat and momentum exchanged between the atmosphere and the ocean interface are very critical. For the ocean, the ocean surface or ocean mixed layer is directly forced by the heat and momentum (i.e. thermal drive and dynamic drive) of the bottom layer of the atmosphere, resulting in changes in sea surface temperature and surface currents, which then drive changes in the lower ocean as a whole. For the atmosphere, the heat and momentum transfer of the ocean to the atmosphere is also the key to the occurrence and development of weather systems. Therefore, accurately calculating the momentum and heat exchange at the air-sea interface is of great significance to improving the simulation accuracy of thermal and dynamic processes in the atmosphere and the ocean.

[0003] In the process of momentum and heat exchange at the air-sea interface, the dynamic exchange coefficient is a core parameter, generally called C d . C d The value is the core parameter used to describe the momentum exchange intensity between the atmosphere and the ocean interface, reflecting the interaction characteristics between the bottom of the atmosphere and the ocean surface. C d The main factor is the wind field at 10m height in the atmosphere U 10 ,because U 10 It is also a common variable in atmospheric observation and numerical simulation, and is easy to obtain through observation and analysis of data. Therefore, a large number of studies have been conducted on U 10 and C d However, despite extensive research, the results obtained by different studies vary widely, especially under strong wind conditions. C d The uncertainty and differences in the values ​​are particularly prominent. This difference mainly comes from the following issues: First, the limitation of parameterized solutions: most existing solutions only consider the 10m wind speed andC d The relationship between other important meteorological and oceanic factors C d In fact, other factors such as sea surface current velocity, sea surface temperature, ocean mixed layer thickness, precipitation, etc. may also affect the air-sea interaction process. C d These factors that are not fully considered lead to C d The accuracy of the estimation results is limited and cannot fully reflect the complex dynamic processes of the sea-air interface.

[0004] The second is the inconsistency of wind fields: In numerical simulations, the wind field data usually used are not real wind fields, but reanalysis wind fields, satellite wind fields or simulated wind fields from different data sources. Due to the different sources of wind field data, the temporal and spatial resolutions and accuracies vary, resulting in differences between different wind fields and actual atmospheric conditions. When these "non-real" wind field data are used to match the real wind field data obtained in the laboratory or observations, the wind field data obtained by the laboratory or observations will be different. C d When the value is too low, the result may be quite different from the actual atmospheric forcing on the ocean.

[0005] The third is the limitations of existing methods: C d Most estimation methods are based on laboratory experiments or empirical formulas, but due to limited experimental data, the results are not universal. In addition, most studies only consider static or approximate sea-air interaction scenarios, lacking dynamic multi-factor complex conditions. C d The exploration of the changing laws, while the actual sea-air interaction environment often involves the joint action of multiple influencing factors.

[0006] In addition, there are also problems with data sources and matching: although there are currently a variety of observation methods and data sources such as buoy observations, satellite remote sensing, and shipborne observations that can be used to estimate C d ,However, data from different sources may have differences in quality, temporal and spatial resolution, etc., but how to effectively integrate these multi-source heterogeneous data and assimilate them into numerical models remains a challenge.

[0007] In summary, it is necessary to develop a system that can integrate multi-source data, dynamically respond to multiple meteorological and oceanographic factors, and match the data and models used. C d Computational methods to improve the accuracy of numerical simulations of air-sea interactions, especially under extreme weather conditions, are technical issues that need to be addressed urgently. Summary of the invention

[0008] 1. Purpose of the invention For the existing sea-air interface dynamic exchange coefficient ( C d ) has defects and shortcomings such as inconsistent results, incomplete consideration of influencing factors, wind field data quality issues, parameterization scheme limitations and data fusion challenges in the calculation. In order to solve at least one of the above and other technical problems in the prior art, the present invention aims to provide a method, medium and program product for estimating and predicting the dynamic exchange coefficient of the sea-air interface based on AI and multi-source fusion data. By combining AI technology and multi-source heterogeneous data, considering 10m wind field ( U 10 ) and matching the data and models used to achieve the dynamic exchange coefficient ( C d ), thereby improving the accuracy of numerical simulations of air-sea interactions and providing more reliable technical support for marine meteorological forecasting and scientific research.

[0009] (II) Technical solution In order to achieve the purpose of the invention and solve the technical problems, the present invention adopts the following technical solutions: The first invention object of the present invention is to provide a method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data, so as to improve the accuracy of numerical simulation, prediction and scientific research of air-sea interaction. The method comprises at least the following steps when implemented: SS1. Reconstruct the calculation formula of the dynamic exchange coefficient of the air-sea interface: Based on the Lagrangian fluid motion equation and the geostrophic flow equation, a variety of meteorological and oceanic variable factors including at least 10m wind speed, sea surface current velocity, Earth's Coriolis force, sea surface geostrophic current, ocean mixed layer thickness, and atmosphere and ocean density are comprehensively considered to reconstruct a sea-air interface dynamic exchange coefficient that can integrate multiple meteorological and oceanic data and simultaneously consider multiple different variable factors C d ' : in, U 10 is the wind field at 10 meters above the atmosphere, u 10 and v 10 They are U 10 Components in the X and Y directions; u and v are the components of the sea surface current velocity in the X and Y directions respectively; fis the Earth's Coriolis force, which reflects the effect of the Earth's rotation on fluid motion; u g and v g are the components of the geostrophic current in the X and Y directions, respectively, describing the influence of the pressure gradient at the air-sea interface; K is the horizontal kinetic energy of a unit mass particle at the air-sea interface and ; SS2. Collection and preprocessing of multi-source data: Collect and organize data for estimating the dynamic exchange coefficient of the air-sea interface C d ' The multi-source meteorological and oceanographic data collected include at least ocean drifting buoy data, satellite altimeter remote sensing data and atmospheric reanalysis data, and then the quality control and preprocessing of the collected multi-source data are carried out. Through outlier detection, missing value processing, data consistency check, and spatiotemporal interpolation and format unification of data from different sources, a multi-source fusion data set containing multiple meteorological and oceanographic factors is constructed, and the multi-source fusion data set is divided into independent training set and validation set based on time series stratified sampling; SS3. Construction and training of AI deep learning models: Based on the reconstruction in step SS1 C d ' The AI ​​deep learning model suitable for multi-source fusion data is designed and constructed by using the calculation formula and deep learning algorithm. The input layer is used to receive the relevant meteorological and ocean factors in the multi-source fusion data, and the hidden layer is responsible for extracting the spatiotemporal characteristics and establishing the meteorological and ocean variable factors and the sea-air interface dynamic exchange coefficient through nonlinear mapping. C d ' The output layer is used to generate the dynamic exchange coefficient of the air-sea interface C d ' The model is trained based on the model training data set divided in step SS2, and the model parameters are updated using the Adam optimizer, the root mean square error (RMSE) loss function and / or the mini-batch gradient descent method, and a Dropout layer and L2 regularization are added to the model network to prevent overfitting; SS4. Evaluation and optimization of AI deep learning models: Based on the model validation data set divided in step SS2, the performance of the AI ​​deep learning model trained in step SS3 is evaluated. The evaluation indicators include root mean square error (RMSE), mean absolute error (MAE) and / or correlation coefficient (R 2), and use the K-fold cross-validation method to evaluate the generalization ability of the model. Then, based on the evaluation results, optimize the model performance by adjusting the network structure (such as the number of layers, the number of neurons, etc.), optimizing hyperparameters (such as learning rate, batch size, etc.), and trying different feature combinations. Iterate the optimization process until the model performance reaches the preset threshold or the performance is no longer significantly improved; SS5. Dynamic exchange coefficient of air-sea interface C d ' Estimates and forecasts: Based on the AI ​​deep learning model optimized in step SS4, input the current time t Meteorological and oceanographic variable data are used to generate the dynamic exchange coefficient of the air-sea interface C d ' The current estimated value of and the next moment are calculated by the following formula t + dt The kinetic energy of the ocean K t+dt and the components of the surface velocity in the X and Y directions u t+dt , v t+dt ,in dt is the time step, representing the time interval between each iteration: Calculating the next moment t + dt of u t+dt and v t+dt After that, it is used as the input of the air-sea interaction numerical simulation system and a round of air-sea coupling numerical simulation is completed to obtain t + dt The new meteorological and oceanographic variables at the moment are then passed as input data to the AI ​​deep learning model optimized in step SS4, and the AI ​​deep learning model is used to estimate the next moment. t + dt The dynamic exchange coefficient of the air-sea interface C d ' ,t+dt , through iteration, the air-sea interface dynamic exchange coefficients for multiple future time steps are calculated. C d ' prediction.

[0010] The second inventive object of the present invention is to provide a computer program product, including computer instructions, for executing the above-mentioned method for estimating and predicting the dynamic exchange coefficient of the sea-air interface based on AI and multi-source fusion data.

[0011] The third inventive object of the present invention is to provide a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for estimating and predicting the dynamic exchange coefficient of the sea-air interface based on AI and multi-source fusion data is implemented.

[0012] (III) Technical Effect Compared with the prior art, the method, medium and program product of the present invention for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data have the following technical effects: (1) The present invention reconstructs the dynamic exchange coefficient of the air-sea interface C d ' The calculation formula comprehensively considers multiple meteorological and oceanographic variable factors including wind speed at 10 meters, sea surface current velocity, Earth's Coriolis force, sea surface geostrophic current, ocean mixed layer thickness and atmospheric density, providing a comprehensive model that can simultaneously integrate multi-source data. The calculation formula more accurately reflects the momentum and heat exchange process at the sea-air interface, especially under complex meteorological and oceanic conditions. Compared with the existing single parameterization scheme, the present invention can significantly improve the estimation accuracy of the sea-air interface dynamic exchange coefficient, thereby improving the overall performance of the numerical simulation and forecasting model.

[0013] (2) This invention effectively integrates ocean drifting buoy data, satellite altimeter remote sensing data and atmospheric reanalysis data by introducing the collection and preprocessing of multi-source data, ensuring the diversity and accuracy of the data used. The consistency and integrity of the data are guaranteed by performing quality control, outlier detection, missing value processing and spatiotemporal interpolation on these multi-source data. This not only improves the model's ability to process data from different sources, but also enhances its tolerance for data uncertainty in complex environments.

[0014] (3) The present invention constructs an AI deep learning model, which can use deep learning algorithms to extract spatiotemporal features from multi-source fusion data, and establish meteorological and ocean variable factors and sea-air interface dynamic exchange coefficients based on nonlinear mapping. C d ' This innovative AI model can adaptively update parameters under dynamically changing meteorological and ocean conditions, further improving C d 'At the same time, the Adam optimizer and RMSE loss function training method, as well as the Dropout layer and L2 regularization method, ensure the stability and generalization ability of the model when processing large-scale data.

[0015] (4) By introducing a regular update mechanism, the present invention can retrain the model and update its parameters according to changes in new observation data, so that the model can always maintain efficient adaptability to the latest climate and ocean conditions. In particular, when the model performance drops to a preset threshold, the system can automatically trigger the model update process and select incremental learning or full retraining strategies based on actual conditions. This flexible update mechanism can ensure the continued effectiveness of the model in long-term applications.

[0016] (5) The present invention obtains C d ' Applied to numerical simulation and forecasting models of air-sea interaction, replacing the existing air-sea interface parameterization scheme can not only improve the estimation accuracy of air-sea flux, but also improve the simulation effect of ocean upper structure and atmospheric boundary layer. In addition, this method has shown strong applicability in extreme weather events (such as hurricanes and typhoons) and climate change research, and can provide more accurate support for meteorological forecasting and ocean dynamics research. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the implementation of the method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data of the present invention; Figure 2 The figure shows a schematic diagram of the AI ​​deep learning model using multi-scale spatiotemporal feature extraction process; Figure 3 Shown is a schematic diagram of the update and retraining process of the AI ​​deep learning model. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the implementation of the present invention clearer, the technical scheme in the embodiment of the present invention will be described in more detail below in conjunction with the drawings in the embodiment of the present invention. The described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The embodiments described below with reference to the drawings are exemplary and are intended to be used to explain the present invention, and should not be understood as limiting the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] The present invention aims to provide a method, medium and program product for estimating and predicting the dynamic exchange coefficient of the sea-air interface based on AI and multi-source fusion data. By combining AI technology and multi-source heterogeneous data, considering the 10m wind field ( U 10 ) and matching the data and models used to achieve the dynamic exchange coefficient ( C d ), thereby improving the accuracy of numerical simulations of air-sea interactions and providing more reliable technical support for marine meteorological forecasting and scientific research.

[0020] As a specific example, Figure 1 As shown, the method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data in an embodiment of the present invention is used to improve the accuracy of numerical simulation, prediction and scientific research of air-sea interaction, and the implementation thereof includes the following steps: SS1. Reconstruct the calculation formula of the dynamic exchange coefficient of the air-sea interface:

[0021] Based on the Lagrangian fluid motion equation and the geostrophic flow equation, a variety of meteorological and oceanic variable factors including at least 10m wind speed, sea surface current velocity, Earth's Coriolis force, sea surface geostrophic current, ocean mixed layer thickness, and atmosphere and ocean density are comprehensively considered to reconstruct a sea-air interface dynamic exchange coefficient that can integrate multiple meteorological and oceanic data and simultaneously consider multiple different variable factors C d ' : in, U 10 is the wind field at 10 meters above the atmosphere, u 10 and v 10 They are U 10 Components in the X and Y directions; u and v are the components of the sea surface current velocity in the X and Y directions respectively; f is the Earth's Coriolis force, which reflects the effect of the Earth's rotation on fluid motion; u g and v g are the components of the geostrophic current in the X and Y directions, respectively, describing the influence of the pressure gradient at the air-sea interface; K is the horizontal kinetic energy of a unit mass particle at the air-sea interface and .

[0022] Specifically, the embodiment of the present invention aims to integrate a variety of meteorological and oceanographic data and comprehensively consider a variety of different variable factors when deriving a new calculation formula for the air-sea interface dynamic exchange coefficient. In the following formula, the time step is dt If there is no time specification for a variable, the default is the current time. t The variables at other times are t and dt The combination of is represented as follows, the next moment is represented as t + dt . u For the surface current X The direction of the component, v For the surface current Y The direction of the component, f For the Earth's scientific force, h is the ocean mixed layer thickness, ρ o is the ocean density, ρ a is the atmospheric density. u g and v g The surface geostrophic current X and Y The weight in direction, u 10 and v 10 The wind speed at 10 m is X and Y The weight in direction, U 10 It is the wind field at 10m in the atmosphere. C d is the traditional drag coefficient between the atmosphere and the ocean, C d ' The new drag coefficient between the sea and air interfaces defined by the present invention.

[0023] The Lagrangian fluid motion equations are shown in equations (1) and (2), and the geostrophic flow equations are shown in equations (3) and (4). Substituting equations (3) and (4) into equations (1) and (2) yields equations (5) and (6): Formula (5) × u We get formula (7), formula (6) × v We get formula (8), which is: Formula (7) + formula (8) to get formula (9), which is: Because there are calculation formulas (10) and (11) and mass micelle K The definition of horizontal kinetic energy is (12): X-direction force F x The calculation formula is (13), the Y direction force F y Calculation formula (14): Substituting equations (10)-(14) into equation (9), we get equation (15), which is: make , Substituting formula (16) into formula (15) we get formula (17): After transforming formula (17), we get formula (18): In formula (18), K , u , v and f It can be calculated from ocean drifting buoy data. u g and v g It can be obtained from satellite altimeter remote sensing data. U 10 , u 10 and v 10 It can be obtained from atmospheric reanalysis data. In this way, the buoy data, satellite data and atmospheric reanalysis data can be integrated to calculate C d ' .

[0024] But the calculated C d ' , is to use different moments that have already occurred ( t and t + dt ) data is back-calculated, and it is impossible to calculate ( t ) C d ' To predict the next moment ( t + dt) In the simulation of real air-sea interaction, it is necessary to calculate the relevant variables according to the current time ( t ) of the existing variables, calculate Cd ' , and then use C d ' To predict the next moment ( t + dt) variables. C d ' It is determined by many factors, but previous studies have generally only focused on the effect of wind in the lower atmosphere on C d ’ , but ignores the combined effects of other factors, such as ( x 1 ,x 2 ...x n ), of which x are the current moment ( t ) data and factors. Therefore, artificial intelligence methods can be used to convert the current moment ( t ) related variables ( x 1 , x 2 ...x n ) and the calculated C d ' Establishing connections, the specific details of artificial intelligence are described in detail later.

[0025] SS2. Collection and preprocessing of multi-source data: Collect and organize data for estimating the dynamic exchange coefficient of the air-sea interface C d ' The multi-source meteorological and oceanographic data collected at least include ocean drifting buoy data, satellite altimeter remote sensing data and atmospheric reanalysis data. The collected multi-source data are then quality controlled and preprocessed. Through outlier detection, missing value processing, data consistency check, and spatiotemporal interpolation and format unification of data from different sources, a multi-source fusion dataset containing multiple meteorological and oceanographic factors is constructed. The multi-source fusion dataset is divided into independent training and validation sets based on time series stratified sampling.

[0026] Preferably, in step SS2, the preprocessing of the satellite altimeter remote sensing data may further include the following additional preprocessing steps to improve the accuracy and reliability of the satellite altimeter data and provide higher quality input data for the subsequent estimation of the sea-air interface dynamic exchange coefficient: SS21. Use cross-calibration data from multiple satellites (such as Jason-3, Sentinel-3, etc.) to establish a systematic bias correction model between different satellite data; SS22. Use spectrum analysis method to identify and filter out the influence of tidal signal on sea surface height; SS23. Use the inverse pressure correction model to eliminate the effect of atmospheric pressure changes on sea surface height; SS24. Application of the moist tropospheric delay correction algorithm to correct measurement errors caused by changes in water vapor content.

[0027] Preferably, in step SS2, during the multi-source data collection and preprocessing process, a weighted average method is preferably used to fuse data from different sources to ensure that the contribution of each data source in the final result is proportional to its credibility, and weights are assigned to each data source according to its historical performance, observation accuracy and timeliness, thereby improving C d ' This method can effectively reduce the errors caused by inconsistent data sources and improve the consistency and reliability of the final results.

[0028] SS3. Construction and training of AI deep learning models: Based on the reconstruction in step SS1 C d ' The AI ​​deep learning model suitable for multi-source fusion data is designed and constructed by using the calculation formula and deep learning algorithm. The input layer is used to receive the relevant meteorological and ocean factors in the multi-source fusion data, and the hidden layer is responsible for extracting the spatiotemporal characteristics and establishing the meteorological and ocean variable factors and the sea-air interface dynamic exchange coefficient through nonlinear mapping. C d ' The output layer is used to generate the dynamic exchange coefficient of the air-sea interface C d ' The model is trained based on the model training data set divided in step SS2, and the model parameters are updated using the Adam optimizer, the root mean square error (RMSE) loss function and / or the mini-batch gradient descent method, and the Dropout layer and L2 regularization are added to the model network to prevent overfitting. Specifically, after obtaining ( x 1 ,x 2 ...x n )and C d 'After the relationship between them is established by artificial intelligence (i.e. AI deep learning model), it can be t The data and factors ( x 1 ,x 2 ...x n ), calculate the current time t of C d ' .

[0029] Preferably, in step SS3, during the AI ​​deep learning model construction and training process, multiple deep learning models (such as LSTM, CNN, GRU, etc.) are compared and the most suitable model for air-sea interaction simulation is selected through cross-validation, and each model should be trained and evaluated on the same data set to determine its performance. C d ' The performance in the estimation is used to select the best model.

[0030] Preferably, in step SS3, the constructed AI deep learning model can adopt a multi-scale spatiotemporal feature extraction structure to better capture the complex spatiotemporal correlation in the process of sea-air interface dynamic exchange and improve the prediction accuracy and generalization ability of the model, such as Figure 2 As shown, when implemented, it includes: SS31. Extract spatial features using multi-layer convolutional neural networks (CNNs) with kernel sizes ranging from 3×3 to 7×7 to capture spatial correlations at different scales. SS32. Use long short-term memory (LSTM) or gated recurrent unit (GRU) to extract time series features, considering time scales ranging from 1 hour to 30 days; SS33. Introduce the attention mechanism to adaptively adjust the weights of features at different spatiotemporal scales; SS34. Use residual connections to ensure that deep networks can learn effectively and avoid the gradient vanishing problem.

[0031] SS4. Evaluation and optimization of AI deep learning models: Based on the model validation data set divided in step SS2, the performance of the AI ​​deep learning model trained in step SS3 is evaluated. The evaluation indicators include root mean square error (RMSE), mean absolute error (MAE) and / or correlation coefficient (R 2), and the K-fold cross-validation method is used to evaluate the generalization ability of the model. Then, based on the evaluation results, the model performance is optimized by adjusting the network structure (such as the number of layers, the number of neurons, etc.), optimizing hyperparameters (such as learning rate, batch size, etc.), and trying different feature combinations. The optimization process is iterated until the model performance reaches the preset threshold or the performance is no longer significantly improved.

[0032] Preferably, in step SS4 above, the model evaluation and optimization may further include the following additional evaluation and optimization sub-steps to improve the interpretability and reliability of the model: SS41. Use the additive explanation (SHAP) method to analyze the contribution of different input variables to the model prediction results, and identify the key factors based on the SHAP value. C d ' Estimate the key variables with the greatest impact; SS42. Conduct sensitivity analysis on key variables to assess their impact on model performance and adjust the weights of input features or consider introducing new relevant variables based on the results of sensitivity analysis; SS43. Use ensemble learning methods (such as random forests or gradient boosted trees) to compare and integrate the results of deep learning models to further improve prediction accuracy.

[0033] SS5. Dynamic exchange coefficient of air-sea interface C d ' Estimates and forecasts: Based on the AI ​​deep learning model optimized in step SS4, input the current time t Meteorological and oceanographic variable data are used to generate the dynamic exchange coefficient of the air-sea interface C d ' The current estimated value of and the next moment are calculated by the following formula t + dt The kinetic energy of the ocean K t+dt and the components of the surface velocity in the X and Y directions u t+dt , v t+dt ,in dt is the time step, representing the time interval between each iteration: Calculating the next moment t + dt of u t+dt and v t+dtAfter that, it is used as the input of the air-sea interaction numerical simulation system and a round of air-sea coupling numerical simulation is completed to obtain t + dt The new meteorological and oceanographic variables at the moment are then passed as input data to the AI ​​deep learning model optimized in step SS4, and the AI ​​deep learning model is used to estimate the next moment. t + dt The dynamic exchange coefficient of the air-sea interface C d ' ,t+dt , through iteration, the air-sea interface dynamic exchange coefficients for multiple future time steps are calculated. C d ' prediction.

[0034] Specifically, Substituting into equation (17) yields the following equation (20): Substituting equations (13) and (14) into equations (5) and (6) respectively, we can obtain equations (21) and (22): Substituting equation (16) into equations (21) and (22), we can obtain equations (23) and (24): Will Substituting into equations (23) and (24), we can obtain equations (27) and (28): After transforming equations (27) and (28), we can obtain equations (29) and (30):

[0035] Using formula (20), we can get the next moment K t+dt , using formulas (29) and (30), we can get the next moment u t+dt and v t+dt .

[0036] As a preferred embodiment, in step SS5, the air-sea interface dynamic exchange coefficient is iteratively predicted for multiple future time steps. C d ' When the time step is adjusted, a dynamic adaptive time step adjustment method can be introduced to adjust the time step in real time according to the rate of dynamic change of the air-sea interface. dtWhen weather and ocean conditions change dramatically, reduce the time step size dt To improve the calculation accuracy; when the conditions are relatively stable, increase the time step dt To improve computational efficiency, thus achieving more accurate long-term forecasts.

[0037] Preferably, in step SS5, when performing the multi-step calculation for the future time steps, C d ' When making predictions, the prediction results can be smoothed by using the multi-time scale superposition method. C d ' The forecast results are weighted and superimposed to eliminate possible spike errors in short-term forecasts and maintain stability in long-term forecasts. This smoothing method can enhance the model's tolerance to future uncertainties and ensure the stability and accuracy of the forecast.

[0038] SS6. Application and verification of the method: The air-sea interface dynamic exchange coefficient obtained by the AI ​​deep learning model C d ' The new method is applied to numerical simulation and forecasting models of air-sea interaction, replacing the existing traditional air-sea interface parameterization scheme, and conducting comparative experiments in different sea areas and meteorological conditions to compare the effects of the new method on the prediction of air-sea interaction. C d ' The consistency between the simulation results of the traditional methods and the observational data is evaluated, the performance of the new method is evaluated under different spatiotemporal scales and meteorological and oceanic conditions (including extreme weather events and seasonal changes), the impact of the new method on the estimation of sea-air fluxes (such as momentum, heat, and water vapor flux) and the improvement of the simulation of the ocean upper structure and atmospheric boundary layer are analyzed, and based on long-term observational data and reanalysis data, the performance and reliability of the method on the climate scale are evaluated to ensure its applicability in multiple application scenarios such as meteorological forecasting and ocean dynamics simulation.

[0039] SS7. Update and retrain AI deep learning models: Establish an update and maintenance mechanism for AI deep learning models, regularly retrain the models with new observation data, and ensure that the models remain efficient and accurate under changing climate and ocean conditions, such as Figure 3 As shown, the implementation thereof includes at least the following sub-steps: SS71. Continue to collect and integrate new ocean drifting buoy data, satellite altimeter remote sensing data, and atmospheric reanalysis data, and maintain the same quality control and pre-processing process as in step SS2; SS72. Regularly (depending on the application scenario, the frequency of model updates can be set to weekly, monthly, or quarterly) evaluate the performance of the existing model on new data. When performance indicators (such as RMSE, MAE, R 2 ) When the value drops by more than a preset threshold (such as 5%) compared to the benchmark value, the model update process is triggered; SS73. Choose incremental learning or full retraining strategy based on the amount of data and the degree of change in model performance. Incremental learning is suitable for quickly updating model parameters to retain learned knowledge, while full retraining is suitable for large-scale data updates or when model performance drops significantly. SS74. After each retraining, the performance indicators (such as RMSE, MAE, R 2 ) Evaluate model performance to ensure model robustness and generalization ability, and dynamically adjust model hyperparameters to optimize model performance under complex meteorological and ocean conditions; SS75. Establish a feedback mechanism to monitor the performance of the model in real time and make necessary adjustments and optimizations in time to cope with the impact of extreme weather or other emergencies on the air-sea interface dynamic exchange coefficient. C d ' Through this dynamic updating mechanism, this method can continuously improve the accuracy and reliability of numerical simulation of air-sea interaction.

[0040] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention is described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data, characterized in that: The method comprises at least the following steps when implemented: SS1. Reconstruct the calculation formula of the dynamic exchange coefficient of the air-sea interface: Based on the Lagrangian fluid motion equation and the geostrophic flow equation, and taking into account a variety of meteorological and ocean variable factors, the calculation formula of the dynamic exchange coefficient of the air-sea interface is reconstructed: in, C d ' is the dynamic exchange coefficient of the air-sea interface, U 10 is the wind field at 10 meters above the atmosphere, u 10 and v 10 They are U 10 The components in the X and Y directions, u and v are the components of the sea surface current velocity in the X and Y directions, u g and v g are the components of the geostrophic flow in the X and Y directions, f is the Earth's Coriolis force, K is the horizontal kinetic energy of a unit mass particle at the air-sea interface and ; SS2. Collection and preprocessing of multi-source data: Collect and organize multi-source meteorological and oceanographic data, including at least ocean drifting buoy data, satellite altimeter remote sensing data and atmospheric reanalysis data, and then perform quality control and preprocessing on the collected multi-source data. Through outlier detection, missing value processing, data consistency check, spatiotemporal interpolation and format unification, a multi-source fusion data set is constructed, and the multi-source fusion data set is divided into independent training data sets and validation data sets based on time series stratified sampling. SS3. Construction and training of AI deep learning models: Based on the reconstruction in step SS1 C d ' The calculation formula is used to design and construct an AI deep learning model suitable for multi-source fusion data using a deep learning algorithm. The input layer is used to receive relevant meteorological and oceanographic factors in the multi-source fusion data, and the hidden layer is responsible for extracting spatiotemporal features and establishing meteorological and oceanographic variable factors and their correlation through nonlinear mapping. C d ' The relationship between C d ' The model is trained based on the training data set divided in step SS2, and the model parameters are updated using the Adam optimizer, the root mean square error loss function and / or the mini-batch gradient descent method, and a Dropout layer and L2 regularization are added to the model network to prevent overfitting; SS4. Evaluation and optimization of AI deep learning models: Based on the validation data set divided in step SS2, the performance of the trained AI deep learning model is evaluated. The evaluation indicators include root mean square error, mean absolute error and / or correlation coefficient, and the K-fold cross-validation method is used to evaluate the generalization ability of the model. Then, according to the evaluation results, the model performance is optimized by adjusting the network structure, optimizing hyperparameters, and trying different feature combinations. Iterative optimization is performed until the model performance reaches the preset threshold or the performance is no longer significantly improved; SS5. Dynamic exchange coefficient of air-sea interface C d ' Estimates and forecasts: Based on the AI ​​deep learning model optimized in step SS4, input the current time t Meteorological and oceanographic variable data, generating C d ' The current estimated value of and the next moment are calculated by the following formula t + dt The kinetic energy of the ocean K t+dt and the components of the surface velocity in the X and Y directions u t+dt , v t+dt ,in dt is the time step, representing the time interval between each iteration: Calculating the next moment t + dt of u t+dt and v t+dt After that, it is used as the input of the air-sea interaction numerical simulation system and a round of air-sea coupling numerical simulation is completed to obtain t + dt The new meteorological and oceanographic variables at the moment are then passed as input data to the AI ​​deep learning model optimized in step SS4, and the AI ​​deep learning model is used to estimate the next moment. t + dt The dynamic exchange coefficient of the air-sea interface C d ' ,t+dt , through iteration, the air-sea interface dynamic exchange coefficients for multiple future time steps are calculated. C d ' prediction.

2. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS2, the preprocessing of the satellite altimeter remote sensing data also includes: SS21. Use satellite cross-calibration data to establish a systematic bias correction model between different satellite data; SS22. Use spectrum analysis method to identify and filter out the influence of tidal signal on sea surface height; SS23. Use the inverse pressure correction model to eliminate the effect of atmospheric pressure changes on sea surface height; SS24. Application of the moist tropospheric delay correction algorithm to correct measurement errors caused by changes in water vapor content.

3. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS2, during the multi-source data collection and preprocessing process, the weighted average method is used to fuse data from different sources to ensure that the contribution of each data source in the final result is proportional to its credibility, and weights are assigned to each data source according to its historical performance, observation accuracy and timeliness, thereby improving C d ' The estimation accuracy.

4. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS3, during the AI ​​deep learning model construction and training process, multiple deep learning models are compared and the most suitable model for air-sea interaction simulation is selected through cross-validation. Each model should be trained and evaluated on the same data set to determine its performance in the simulation. C d ' Estimated performance.

5. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS3, the constructed AI deep learning model adopts a multi-scale spatiotemporal feature extraction structure, which specifically includes: SS31. Extract spatial features using multi-layer convolutional neural networks with kernel sizes ranging from 3×3 to 7×7 to capture spatial correlations at different scales. SS32. Extract time series features using long short-term memory networks or gated recurrent units, considering time scales ranging from 1 hour to 30 days; SS33. Introduce the attention mechanism to adaptively adjust the weights of features at different spatiotemporal scales; SS34. Use residual connections to ensure that deep networks can learn effectively and avoid the gradient vanishing problem.

6. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS4, model evaluation and optimization also includes the following sub-steps: SS41. Use the additive explanation (SHAP) method to analyze the contribution of different input variables to the model prediction results, and identify the key factors based on the SHAP value. C d ' Estimate the key variables with the greatest impact; SS42. Conduct sensitivity analysis on key variables to assess their impact on model performance and adjust the weights of input features or consider introducing new relevant variables based on the results of sensitivity analysis; SS43. Use ensemble learning methods to compare and integrate the results of deep learning models to further improve prediction accuracy.

7. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS5, the air-sea interface dynamic exchange coefficients are iteratively predicted for multiple future time steps. C d ' When the time step is adjusted, a dynamic adaptive time step adjustment method is introduced to adjust the time step in real time according to the rate of dynamic change of the air-sea interface. dt When weather and ocean conditions change dramatically, reduce the time step size dt To improve the calculation accuracy; when the conditions are relatively stable, increase the time step dt To improve computational efficiency, thus achieving more accurate long-term forecasts.

8. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: In the above step SS5, when performing multiple time steps in the future C d ' When making predictions, the prediction results are smoothed by using the multi-time scale superposition method. C d ' The forecast results are weighted and superimposed to eliminate possible spike errors in short-term forecasts and maintain stability in long-term forecasts.

9. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: The method also includes step SS6 of applying and verifying the method when implemented, specifically including: applying the air-sea interface dynamic exchange coefficient obtained by the AI ​​deep learning model C d ' The new method is applied to numerical simulation and forecasting models of air-sea interaction, replacing the existing traditional air-sea interface parameterization scheme, and conducting comparative experiments in different sea areas and meteorological conditions to compare the effects of the new method on the prediction of air-sea interaction. C d ' The consistency between the simulation results of the traditional methods and the observational data is evaluated, the performance of the new method is evaluated under different spatiotemporal scales and meteorological-ocean conditions, the impact of the new method on the estimation of air-sea flux and the improvement of the simulation of the upper ocean structure and the atmospheric boundary layer are analyzed, and the performance and reliability of the method on the climate scale are evaluated based on long-term observational data and reanalysis data.

10. The method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data according to claim 1 is characterized in that: When implemented, the method further includes step SS7 of updating and retraining the AI ​​deep learning model, which specifically includes the following sub-steps: First, continue to collect and integrate new ocean drifting buoy data, satellite altimeter remote sensing data, and atmospheric reanalysis data, and maintain the same quality control and preprocessing process as in step SS2; Secondly, the performance of the existing model on new data is regularly evaluated. When the performance indicator drops by more than a preset threshold compared to the baseline value, the model update process is triggered; After that, select incremental learning or full retraining strategy based on the amount of data and the degree of change in model performance. Incremental learning is suitable for quickly updating model parameters to retain learned knowledge, while full retraining is suitable for large-scale data updates or when model performance drops significantly. Furthermore, after each retraining, the model performance is evaluated through performance indicators to ensure the robustness and generalization ability of the model, and the model hyperparameters are dynamically adjusted to optimize the model's performance under complex meteorological and ocean conditions; Finally, a feedback mechanism is established to monitor the performance of the model in real time and make necessary adjustments and optimizations in time to cope with the impact of extreme weather or other emergencies on the air-sea interface dynamic exchange coefficient. C d ' Estimated impact.

11. A computer program product comprising computer instructions, characterized in that: The computer instructions are used to execute the above-mentioned method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the method for estimating and predicting the dynamic exchange coefficient of the air-sea interface based on AI and multi-source fusion data is implemented.

Images