Large atmosphere model and large ocean model coupling method based on deep learning
By designing a deep learning-based coupling method for atmospheric and ocean big models, and using deep learning couplers to achieve efficient interaction between AI meteorological and ocean big models, the problem of lack of effective coupling mechanism in the existing technology is solved, and the accuracy of meteorological and ocean forecasts is significantly improved.
Patent Information
- Application Number
- CN202411592664.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2025-02-28
AI Technical Summary
The lack of effective coupling mechanisms in the prior art has led to the inability of deep learning-based meteorological and marine models to interact effectively and work together, and the inability to fully utilize the advantages of deep learning for meteorological-ocean AI coupling.
A method of coupling atmospheric large-model and ocean large-model based on deep learning is designed, including data preparation and preprocessing module, deep learning coupler design module, model training module, AI meteorological and ocean large-model interface module, coupling prediction module, post-processing module and evaluation and optimization module, and efficient data exchange and state update between AI meteorological and ocean large-models is achieved through deep learning couplers.
It achieves seamless integration with modern AI meteorological and marine models, improves the consistency and accuracy of the overall prediction system, can adapt to the model coupling needs of different resolutions and time scales, and significantly improves the accuracy of meteorological and marine forecasts of AI meteorological models.
Smart Images
Figure CN119539030A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of large model coupling methods, and in particular to a coupling method of a large atmospheric model and a large ocean model based on deep learning. Background Art
[0002] In recent years, with the rapid development of deep learning technology, AI big models in the fields of meteorology and oceanography have made significant progress. For example, meteorological big models such as GraphCast, and ocean big models such as Xihe and AI-GOMS have demonstrated strong prediction capabilities in their respective fields. However, these models usually run independently and lack effective coupling mechanisms.
[0003] In meteorology, the interaction between the atmosphere and the ocean is an extremely important process involving complex exchanges of heat, momentum and matter. These exchanges play a key role in the evolution of the global climate system and the formation of weather phenomena. For example, the interaction between the atmosphere and the ocean has an important impact on the formation and development of the El Nino-Southern Oscillation (ENSO), monsoon systems and tropical cyclones.
[0004] At present, the coupling of traditional fluid dynamics-based atmospheric and ocean models mainly relies on "couplers", such as CPL7 and MCT, etc. For AI-based meteorological and ocean large models, there is no coupler that can organically couple them.
[0005] Therefore, how to build an efficient and flexible meteorological-oceanographic AI coupling system that can fully utilize the advantages of deep learning has become a key issue that needs to be urgently addressed in the current meteorological and oceanographic field. Summary of the invention
[0006] The purpose of the present invention is to solve at least one of the technical problems existing in the prior art and to provide a method for coupling a large atmospheric model and a large ocean model based on deep learning, which can solve the above-mentioned problem.
[0007] To achieve the above objectives, the present invention provides the following technical solutions: an atmospheric large model based on deep learning, comprising a data preparation and preprocessing module, a deep learning coupler design module, a model training module, an AI meteorological large model interface module, an AI ocean large model interface module, a coupling prediction module, a post-processing module and an evaluation and optimization module.
[0008] Preferably, the data preparation and preprocessing module collects historical operation data of traditional couplers (such as CPL7, MCT), including input and output data of atmosphere and ocean models.
[0009] Preferably, the deep learning coupler design module designs a deep neural network structure suitable for meteorological-ocean coupling, such as resnet-1d.
[0010] Preferably, the model training module uses the collected historical data to train a deep learning coupler to learn the complex interaction relationship between atmospheric and ocean elements.
[0011] Preferably, the AI meteorological big model interface module is designed to have a data exchange interface with an existing AI meteorological big model (such as GraphCast, Pangu-Weather).
[0012] Preferably, the AI ocean big model interface module is designed to have a data exchange interface with an existing AI ocean big model (such as AI-GOMS).
[0013] Preferably, the coupled prediction module: during the prediction process, uses the trained deep learning coupler to achieve efficient data exchange and status update between the AI meteorological and ocean large models.
[0014] Preferably, the post-processing module performs post-processing on the coupling prediction results, including data correction and deviation correction.
[0015] Preferably, the evaluation and optimization module: evaluates the accuracy of the coupling prediction results, and continuously optimizes the deep learning coupler based on the evaluation results.
[0016] Preferably, the deep learning-based method for coupling a large atmospheric model with a large ocean model comprises the following steps: S1. Data preparation and preprocessing: collecting historical operating data of traditional couplers (such as CPL7, MCT), including input and output data of atmospheric and ocean models;
[0017] Among them, the data is aligned in time and space to ensure that the data from different sources match in time and space dimensions;
[0018] Among them, data standardization and normalization are carried out to make different physical quantities comparable;
[0019] Among them, construct training, validation and test data sets;
[0020] S2 Deep Learning Coupler Design: Using a general architecture similar to ResNet, we design a deep learning model that can handle heterogeneous data;
[0021] Among them, the attention mechanism (such as se-resnet) is introduced to highlight the important sea-air interaction process;
[0022] S3 model training: Use the collected historical data to train the deep learning coupler, adopt appropriate loss functions such as mean square error (MSE) and correlation coefficient, and use optimizers (such as Adam) and learning rate scheduling strategies to optimize the model;
[0023] Among them, regularization techniques (such as L2 regularization and Dropout) are implemented to prevent overfitting;
[0024] S4AI meteorological big model interface: Design the data exchange interface with GraphCast and Pangu-WeatherAI meteorological big models to ensure that the interface can efficiently process large-scale three-dimensional meteorological field data;
[0025] Among them, data format conversion and coordinate system alignment are achieved;
[0026] S5AI Ocean Large Model Interface: Design a data exchange interface with the AI-GOMSAI ocean large model, taking into account the particularities of the ocean model, such as processing irregular grids and vertical layer structures, to achieve an efficient ocean data reading and writing mechanism;
[0027] S6 coupled prediction: At each time step, read the current status of the AI meteorological and ocean models, input the status data into the deep learning coupler, and obtain the coupled prediction results;
[0028] Among them, the status of AI meteorological and ocean models is updated to make the next prediction, and parallel computing technology is used to improve the computational efficiency of the coupling process;
[0029] S7 post-processing: bias correction of coupled prediction results and construction of correction model using historical observation data;
[0030] Among them, the statistical downscaling method is applied to improve the prediction accuracy of local areas and generate prediction products of various meteorological and ocean elements, such as temperature field, pressure field, and ocean current field;
[0031] S8 Evaluation and Optimization: Use a variety of evaluation indicators (such as root mean square error, correlation coefficient) to evaluate the coupled prediction results;
[0032] Among them, the advantages and disadvantages of deep learning couplers are analyzed by comparing with traditional coupling methods;
[0033] Among them, based on the evaluation results, the structure and parameters of the deep learning coupler are continuously optimized;
[0034] Among them, an online learning mechanism is established to enable the model to adapt to new observation data and climate change.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] 1. This deep learning-based atmospheric large model and ocean large model coupling method achieves seamless integration with modern AI meteorological and ocean large models, improves the consistency and accuracy of the overall prediction system, can adapt to the model coupling requirements of different resolutions and time scales, and is easy to integrate with various AI meteorological and ocean large models.
[0037] 2. The deep learning-based coupling method of the atmospheric large model and the ocean large model has low coupling: it is easy to integrate new ocean and meteorological large models without affecting the operation and training of the large model itself. At the same time, computing resource utilization: fully utilize hardware acceleration such as GPU to improve computing efficiency and resource utilization.
[0038] 3. The deep learning-based atmospheric large model and ocean large model coupling method realizes the effective coupling of the AI meteorological large model and the AI ocean large model by constructing a deep learning coupler, and uses the existing coupler data to train the deep learning model to achieve efficient atmosphere-ocean element exchange in the prediction process. The present invention provides a possibility for the organic integration of the AI meteorological large model and the AI ocean large model, significantly improves the accuracy of the meteorological and ocean forecasts of the AI meteorological large model, provides new technical support for weather forecasting and climate research, and has important scientific value and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The present invention is further described below in conjunction with the accompanying drawings and embodiments:
[0040] Figure 1 It is a schematic diagram of the model structure of the present invention. DETAILED DESCRIPTION
[0041] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.
[0042] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.
[0043] In the description of the present invention, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.
[0044] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.
[0045] See also Figure 1 , the present invention provides a technical solution: a coupling method of an atmospheric large model and an ocean large model based on deep learning, an atmospheric large model based on deep learning, including a data preparation and preprocessing module, a deep learning coupler design module, a model training module, an AI meteorological large model interface module, an AI ocean large model interface module, a coupling prediction module, a post-processing module and an evaluation and optimization module;
[0046] Among them, the data preparation and preprocessing module: collects historical operation data of traditional couplers (such as CPL7, MCT), including input and output data of atmospheric and ocean models;
[0047] Among them, the deep learning coupler design module: designs a deep neural network structure suitable for meteorological-ocean coupling, such as resnet-1d;
[0048] Among them, the model training module: uses the collected historical data to train the deep learning coupler to learn the complex interaction relationship between atmospheric and ocean elements;
[0049] Among them, the AI meteorological big model interface module: designs the data exchange interface with the existing AI meteorological big models (such as GraphCast, Pangu-Weather);
[0050] Among them, AI ocean big model interface module: design data exchange interface with existing AI ocean big model (such as AI-GOMS);
[0051] Among them, the coupled prediction module: in the prediction process, the trained deep learning coupler is used to realize efficient data exchange and status update between AI meteorological and ocean large models;
[0052] Among them, the post-processing module: post-processes the coupling prediction results, including data correction and deviation correction;
[0053] Among them, the evaluation and optimization module: evaluates the accuracy of the coupling prediction results, and continuously optimizes the deep learning coupler based on the evaluation results.
[0054] The deep learning-based coupling method of the atmospheric large model and the ocean large model includes the following steps: S1. Data preparation and preprocessing: Collect historical operation data of traditional couplers (such as CPL7 and MCT), including input and output data of atmospheric and ocean models;
[0055] Among them, the data is aligned in time and space to ensure that the data from different sources match in time and space dimensions;
[0056] Among them, data standardization and normalization are carried out to make different physical quantities comparable;
[0057] Among them, construct training, validation and test data sets;
[0058] S2 Deep Learning Coupler Design: Using a general architecture similar to ResNet, we design a deep learning model that can handle heterogeneous data;
[0059] Among them, the attention mechanism (such as se-resnet) is introduced to highlight the important sea-air interaction process;
[0060] S3 model training: Use the collected historical data to train the deep learning coupler, adopt appropriate loss functions such as mean square error (MSE) and correlation coefficient, and use optimizers (such as Adam) and learning rate scheduling strategies to optimize the model;
[0061] Among them, regularization techniques (such as L2 regularization and Dropout) are implemented to prevent overfitting;
[0062] S4AI meteorological big model interface: Design the data exchange interface with GraphCast and Pangu-WeatherAI meteorological big models to ensure that the interface can efficiently process large-scale three-dimensional meteorological field data;
[0063] Among them, data format conversion and coordinate system alignment are achieved;
[0064] S5AI Ocean Large Model Interface: Design a data exchange interface with the AI-GOMSAI ocean large model, taking into account the particularities of the ocean model, such as processing irregular grids and vertical layer structures, to achieve an efficient ocean data reading and writing mechanism;
[0065] S6 coupled prediction: At each time step, read the current status of the AI meteorological and ocean models, input the status data into the deep learning coupler, and obtain the coupled prediction results;
[0066] Among them, the status of AI meteorological and ocean models is updated to make the next prediction, and parallel computing technology is used to improve the computational efficiency of the coupling process;
[0067] S7 post-processing: bias correction of coupled prediction results and construction of correction model using historical observation data;
[0068] Among them, the statistical downscaling method is applied to improve the prediction accuracy of local areas and generate prediction products of various meteorological and ocean elements, such as temperature field, pressure field, and ocean current field;
[0069] S8 Evaluation and Optimization: Use a variety of evaluation indicators (such as root mean square error, correlation coefficient) to evaluate the coupled prediction results;
[0070] Among them, the advantages and disadvantages of deep learning couplers are analyzed by comparing with traditional coupling methods;
[0071] Among them, based on the evaluation results, the structure and parameters of the deep learning coupler are continuously optimized;
[0072] Among them, an online learning mechanism is established to enable the model to adapt to new observation data and climate change.
[0073] Furthermore, this method achieves seamless integration with modern AI meteorological and oceanographic large models, improves the consistency and accuracy of the overall prediction system, can adapt to the coupling requirements of models with different resolutions and time scales, and is easy to integrate with various AI meteorological and oceanographic large models.
[0074] Furthermore, the method has low coupling: it is easy to integrate new ocean and meteorological large models without affecting the operation and training of the large models themselves, and computing resource utilization: it makes full use of hardware acceleration such as GPU to improve computing efficiency and resource utilization.
[0075] Furthermore, the method realizes the effective coupling of the AI meteorological big model and the AI ocean big model by constructing a deep learning coupler, uses the existing coupler data to train the deep learning model, and realizes efficient exchange of atmosphere-ocean elements in the prediction process. The present invention provides a possibility for the organic integration of the AI meteorological big model and the AI ocean big model, significantly improves the accuracy of the meteorological and ocean forecasts of the AI meteorological big model, provides new technical support for weather forecasting and climate research, and has important scientific value and application prospects.
[0076] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the technical field without departing from the purpose of the present invention.
Claims
1. A large atmospheric model based on deep learning, characterized by: It includes data preparation and preprocessing module, deep learning coupler design module, model training module, AI meteorological large model interface module, AI ocean large model interface module, coupled prediction module, post-processing module and evaluation and optimization module.
2. The atmospheric large model based on deep learning according to claim 1, characterized in that: The data preparation and preprocessing module collects historical operation data of traditional couplers (such as CPL7, MCT), including input and output data of atmospheric and ocean models.
3. The atmospheric large model based on deep learning according to claim 1, characterized in that: The deep learning coupler design module designs a deep neural network structure suitable for meteorological-ocean coupling, such as ResNet-1D.
4. The atmospheric large model based on deep learning according to claim 1, characterized in that: The model training module uses the collected historical data to train the deep learning coupler to learn the complex interaction relationship between atmospheric and ocean elements.
5. The atmospheric large model based on deep learning according to claim 1, characterized in that: The AI meteorological big model interface module is designed to exchange data with existing AI meteorological big models (such as GraphCast and Pangu-Weather).
6. The atmospheric large model based on deep learning according to claim 1, characterized in that: The AI ocean large model interface module is designed to have a data exchange interface with the existing AI ocean large model (such as AI-GOMS).
7. The atmospheric large model based on deep learning according to claim 1, characterized in that: The coupled prediction module: During the prediction process, the trained deep learning coupler is used to realize efficient data exchange and status update between the AI meteorological and ocean large models.
8. The atmospheric large model based on deep learning according to claim 1, characterized in that: The post-processing module performs post-processing on the coupling prediction results, including data correction and deviation correction.
9. The atmospheric large model based on deep learning according to claim 1, characterized in that: The evaluation and optimization module evaluates the accuracy of the coupling prediction results and continuously optimizes the deep learning coupler based on the evaluation results.
10. A method for coupling a large atmospheric model with a large ocean model based on deep learning, characterized in that: It includes the following steps: S1 Data preparation and preprocessing: Collect historical operation data of traditional couplers (such as CPL7 and MCT), including input and output data of atmospheric and ocean models; Among them, the data is aligned in time and space to ensure that the data from different sources match in time and space dimensions; Among them, data standardization and normalization are carried out to make different physical quantities comparable; Among them, construct training, validation and test data sets; S2 Deep Learning Coupler Design: Using a general architecture similar to ResNet, we design a deep learning model that can handle heterogeneous data; Among them, the attention mechanism (such as se-resnet) is introduced to highlight the important sea-air interaction process; S3 model training: Use the collected historical data to train the deep learning coupler, adopt appropriate loss functions such as mean square error (MSE) and correlation coefficient, and use optimizers (such as Adam) and learning rate scheduling strategies to optimize the model; Among them, regularization techniques (such as L2 regularization and Dropout) are implemented to prevent overfitting; S4AI meteorological big model interface: Design the data exchange interface with GraphCast and Pangu-WeatherAI meteorological big models to ensure that the interface can efficiently process large-scale three-dimensional meteorological field data; Among them, data format conversion and coordinate system alignment are achieved; S5AI Ocean Large Model Interface: Design a data exchange interface with the AI-GOMSAI ocean large model, taking into account the particularities of the ocean model, such as processing irregular grids and vertical layer structures, to achieve an efficient ocean data reading and writing mechanism; S6 coupled prediction: At each time step, read the current status of the AI meteorological and ocean models, input the status data into the deep learning coupler, and obtain the coupled prediction results; Among them, the status of AI meteorological and ocean models is updated to make the next prediction, and parallel computing technology is used to improve the computational efficiency of the coupling process; S7 post-processing: bias correction of coupled prediction results and construction of correction model using historical observation data; Among them, the statistical downscaling method is applied to improve the prediction accuracy of local areas and generate prediction products of various meteorological and ocean elements, such as temperature field, pressure field, and ocean current field; S8 Evaluation and Optimization: Use a variety of evaluation indicators (such as root mean square error, correlation coefficient) to evaluate the coupled prediction results; Among them, the advantages and disadvantages of deep learning couplers are analyzed by comparing with traditional coupling methods; Among them, based on the evaluation results, the structure and parameters of the deep learning coupler are continuously optimized; Among them, an online learning mechanism is established to enable the model to adapt to new observation data and climate change.