A method and device for assimilating meteorological numerical models based on neural network
Through the neural network-based meteorological numerical model assimilation method, an intelligent assimilation system suitable for all numerical models is built, which solves the problems of high development costs and low forecast accuracy in the existing technology, and achieves efficient and accurate weather forecast results.
Patent Information
- Application Number
- CN202210459619.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-28
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2042-04-28
AI Technical Summary
The development cost of mainstream data assimilation methods in existing business assimilation systems is high and cannot be directly applied to different numerical models, making it difficult to ensure forecast accuracy.
Using the meteorological numerical model assimilation method based on neural network, an intelligent assimilation system is constructed by determining assimilation variables and time windows, and using neural network training optimization algorithms to iteratively optimize parameters, providing the optimal initial field coordinated dynamically and thermally, suitable for all Numerical mode.
It reduces development costs, improves the accuracy of numerical modes in forecasting, avoids the consumption of modules and resources for development, and has high computing efficiency and real-time performance.
Smart Images
Figure CN115081307B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the cross-technical field of computer and meteorology, and in particular to a method and device for assimilating meteorological numerical models based on neural networks. Background Art
[0002] Numerical weather forecasting is a typical initial value problem. Under given initial conditions and boundary conditions, the basic equations of atmospheric motion are numerically solved, and the atmospheric state at future moments is predicted from the atmospheric state at the known initial moment. It has now become an important basis and fundamental scientific approach for meteorological departments to make operational weather forecasts, and has a status and role that cannot be replaced by other forecasting methods. The Earth's atmosphere is a geophysical system that evolves over time and has nonlinear characteristics. The accuracy of using existing numerical models to simulate the motion of the atmosphere and predict future climate states depends largely on the initial values of the model. In this case, the quality of the initial conditions has an increasingly prominent impact on the reanalysis and forecast results of the Earth's atmosphere.
[0003] At present, the most commonly used method for providing initial fields internationally is the data assimilation method, that is, making full use of various observational data and numerical forecast background fields to provide a dynamically and thermally coordinated optimal initial field for the numerical model. The data assimilation method is to integrate the observational data and the numerical model in the constructed atmospheric dynamics framework, so that the numerical model can continuously rely on the observation value to automatically adjust the model trajectory and reduce the forecast error. However, the mainstream data assimilation methods (four-dimensional variational assimilation and ensemble Kalman filtering) in the existing business assimilation system have the disadvantage of too high development cost, and cannot be directly and reliably applied to different numerical models, making it difficult to guarantee the accuracy of the existing numerical models in forecasting.
[0004] Among them, the four-dimensional variational assimilation method uses the numerical model as a constraint and can assimilate observational data at multiple times within the window. At the same time, the variational-based framework makes it easy to adopt bias correction and quality control processes for observations. However, the four-dimensional variational assimilation method requires the development of the adjoint model of the numerical model, which requires a lot of computing resources and thus limits its development to a certain extent. The assimilation method based on ensemble Kalman filtering can construct flow-dependent background error covariance based on ensemble samples, and can also avoid solving the tangent adjoint model. However, the limited number of ensemble members in the ensemble assimilation method will inevitably cause large sample errors. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a meteorological numerical model assimilation method and device based on neural network, which can be applicable to all numerical models, reduce development costs, and improve the accuracy of existing numerical models in forecasting.
[0006] The object of the present invention can be achieved by the following technical scheme: A meteorological numerical model assimilation method based on neural network, comprising the following steps:
[0007] A. Determine the numerical model that needs to be established for the intelligent assimilation system and encapsulate it, which includes the following sub-steps:
[0008] A.1. Select assimilation variables and determine assimilation period;
[0009] Encapsulating the numerical model so that the numerical model outputs the numerical field of the selected assimilation variable according to the assimilation period;
[0010] A.2. Establish a data interaction interface between the numerical model and the intelligent assimilation system: the numerical model receives the numerical field output by the intelligent assimilation system to perform model integration for the next assimilation cycle, and the intelligent assimilation system reads the output of the numerical model as the model input for the next moment;
[0011] A.3. Determine the assimilation time window, that is, the time span of the observational data that the numerical model needs to assimilate;
[0012] B. Constructing an intelligent assimilation system, including the following sub-steps:
[0013] B.1. Read the meteorological data of the assimilated variables determined in step A.1 and construct the corresponding feature extraction module;
[0014] B.2. Aiming at the uneven grid distribution of the numerical model, construct an observation operator that maps the observation data to the numerical field of the numerical model;
[0015] C. Training the neural network of the intelligent assimilation system includes the following sub-steps:
[0016] C.1. Using the observation operator constructed in step B.2, construct observation image data of the required resolution for neural network training;
[0017] C.2. Perform data preprocessing on the observation data and initial meteorological data of the numerical model input into the neural network;
[0018] C.3. Select an intelligent optimization algorithm, construct an objective function, and iteratively train the parameters in the feature extraction module;
[0019] C.4. The intelligent optimization stops when the number of iterations reaches the set number, and the neural network parameters with the best objective function value are saved;
[0020] D. Using intelligent assimilation system to assist numerical model forecasting includes the following sub-steps:
[0021] D.1. Determine the starting time, the observational data and the initial meteorological data of the numerical model for the starting time of each assimilation cycle, and perform the same data preprocessing as step C.2;
[0022] D.2. Input the optimal parameters of the neural network saved in step C.4 into the intelligent assimilation system, input the observation data after data preprocessing in step D.1 and the initial meteorological data of the numerical model into the feature extraction module of the intelligent assimilation system to perform feature extraction, iterate to the set forecast time, and obtain the forecast result;
[0023] D.3. Based on the forecast results, calculate the set relevant indexes and evaluation indicators.
[0024] Furthermore, the assimilation variables selected in step A.1 are specifically one or more atmospheric variables, one or more land surface variables and one or more ocean variables.
[0025] Furthermore, the forecast results in step D.2 include but are not limited to short-term weather forecast results, medium-term weather forecast results, long-term weather forecast results or climate forecast results.
[0026] Furthermore, the step B.1 specifically adopts one or a combination of convolutional neural networks, recurrent convolutional neural networks, long short-term memory convolutional neural networks, and graph convolutional neural networks to construct a feature extraction module.
[0027] Furthermore, the step B.2 specifically includes the following steps:
[0028] (1) Mapping the acquired observational data to the selected numerical model forecast area, which involves the following operations: slicing, cropping, and splicing;
[0029] (2) Adjust the resolution of the acquired observational data to be consistent with the selected numerical model forecast field, which involves operations including but not limited to interpolation and downsampling.
[0030] Furthermore, the intelligent optimization algorithm in step C.3 is specifically a population biological intelligent heuristic algorithm including a particle swarm algorithm, an ant colony algorithm, and a bee colony algorithm, or a gradient-free optimization algorithm including a Bayesian deep neural network.
[0031] Furthermore, the step C.3 specifically adopts one or a combination of image comparison methods to construct the objective function, and the image comparison methods include but are not limited to root mean square error (MSE), mean absolute error (MSE);
[0032] The output value of the objective function is specifically: within a determined assimilation time window, the sum of all optimization results output by the numerical model after the intelligent assimilation system and the values calculated by the image comparison function of the observation data.
[0033] A meteorological numerical model assimilation device based on a neural network comprises a first determination module, a model training module, a prediction module and an assimilation model evaluation module connected in sequence, wherein the first determination module is used to obtain observation data of assimilated variables and initial meteorological data of numerical models, determine an assimilation period and an assimilation time window, and encapsulate numerical models;
[0034] The model training module is used to convert the observation data into a specified input format, perform data preprocessing on the observation data and initial meteorological data of the numerical model input to the model, and select an intelligent optimization algorithm to train the model;
[0035] The prediction module is used to input the trained model optimal parameters into the intelligent assimilation system and perform iterative prediction to obtain the prediction results;
[0036] The assimilation model evaluation module is used to convert the output data into set meteorological grid data and a variety of quantitative evaluation indicators.
[0037] A computer device includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of the above-mentioned meteorological numerical model assimilation method based on neural network are implemented.
[0038] A readable storage medium stores computer executable instructions. When the computer executable instructions are called and executed by a processor, the computer executable instructions prompt the processor to execute the above-mentioned meteorological numerical model assimilation method based on neural network.
[0039] Compared with the prior art, the present invention has the following advantages:
[0040] (1) Save time and development costs. The present invention can avoid the resource consumption of developing accompanying modules of numerical models and designing localized solutions, and has the characteristics of high computational efficiency and strong real-time performance.
[0041] (2) The model is highly practical. The meteorological numerical model assimilation system based on neural network designed by the present invention is applicable to any numerical model. It creatively uses intelligent optimization algorithm to iteratively optimize the neural network framework, solving the problem of gradient disappearance in the training process of deep learning network output of numerical model.
[0042] (3) High forecast accuracy. The meteorological numerical model assimilation system based on neural network designed by the present invention can provide a dynamic and thermally coordinated optimal initial field for the numerical model, thereby improving the accuracy of the existing numerical model in forecasting. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 It is a schematic diagram of the method flow of the present invention;
[0044] Figure 2 It is a schematic diagram of the method architecture of the present invention;
[0045] Figure 3 It is a schematic diagram of the forecast process in the present invention;
[0046] Figure 4 It is a schematic diagram of the structure of the device of the present invention;
[0047] Figure 5 Schematic diagram of the structure of a computer device in an embodiment. DETAILED DESCRIPTION
[0048] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0049] Example
[0050] like Figure 1 As shown, a meteorological numerical model assimilation method based on neural network includes the following steps:
[0051] A. Encapsulate the numerical model and determine the assimilation variables and parameters, including the following sub-steps:
[0052] A.1. Select assimilation variables, determine assimilation period, and obtain observational meteorological data of assimilation variables;
[0053] This example uses ENSO (El Taking the El Nino-Southern Oscillation event as an example, data assimilation is performed on the sea surface temperature in the numerical model. The assimilation period is one month. The observational meteorological data of the sea surface temperature can be downloaded from the website of the National Center for Atmospheric Research.
[0054] A.2. Establish a data interaction interface between the numerical model and the intelligent assimilation system. The numerical model receives the numerical field output by the intelligent assimilation system to perform model integration for the next assimilation cycle. The intelligent assimilation system reads the output of the numerical model as the model input for the next moment.
[0055] A.3. Determine the assimilation time window, that is, the time span of the observational data that the numerical model needs to assimilate.
[0056] B. Building an intelligent assimilation system and observation operator, including the following sub-steps:
[0057] B.1. Read the observed meteorological data of the assimilated variables determined in step A.1 and construct a feature extraction module for it; in this embodiment, a two-dimensional convolutional neural network is selected to extract features from the observed meteorological data, and its input is the observed meteorological data at the current moment and the initial field of the numerical model at the current moment;
[0058] Specifically, the convolutional neural network selected in this embodiment can manually set the convolution kernel weight parameters, such as Figure 2 As shown, the meteorological data is input into the feature extraction module constructed in step B.1, and the output of the assimilation system is integrated as the initial field of the numerical model at the current moment. After integration, the forecast result of the numerical model at the next moment is read, and it is compared with the observation data at the next moment to calculate the objective function. The intelligent optimization algorithm continuously iterates and optimizes by adjusting the weight parameters until the iteration stops;
[0059] B.2. In view of the uneven grid distribution of the numerical model, an observation operator is constructed to map the observation data to the physical field of the numerical model. First, the acquired observation data is mapped to the selected numerical model forecast area, which may involve operations such as slicing, cropping and splicing. Then, the resolution of the acquired observation data is adjusted to be consistent with the selected numerical model forecast field, which may involve operations such as interpolation and downsampling.
[0060] B.2.1. The images of meteorological observation data are cropped according to the numerical model area. The world ocean map separated from the Pacific Ocean needs to be cropped and spliced in sections;
[0061] B.2.2. Adjust the resolution of the image of the observed meteorological data. The tool used in this embodiment is the resize() function of the opencv component in Python. First, adjust the image resolution in the longitude and latitude directions respectively, then cut it according to the local resolution and perform the corresponding downscaling operation.
[0062] C. Training the neural network of the intelligent assimilation system and saving the optimal network parameters includes the following sub-steps:
[0063] C.1. Perform data preprocessing on the observation data and initial meteorological data of the numerical model input into the neural network;
[0064] This embodiment selects to perform normalization operation on the observation data and the initial field data of the numerical model;
[0065] C.2. Select an intelligent optimization algorithm, construct an objective function, and iteratively train the parameters in the feature extraction module;
[0066] In view of the fact that the gradient disappears when the neural network optimizes the numerical model output using Fortran as the programming language, this embodiment selects the particle swarm algorithm to optimize the weight parameters iteratively;
[0067] In addition, in this embodiment, the MSE function is selected as the image contrast function, and the objective function is the sum of the MSE function values of the model numerical field and the observation data at all output moments within the assimilation period;
[0068] C.3. The model stops when it iterates to the specified number of iterations, and the weight parameters with the optimal objective function value are saved;
[0069] D. Using intelligent assimilation system to assist numerical model forecasting includes the following sub-steps:
[0070] D.1. Determine the starting time, the observational data and the initial meteorological data of the numerical model for the starting time of each assimilation cycle, and perform the same data preprocessing as step C.1;
[0071] D.2. Input the optimal parameters of the neural network saved in step C.3 into the intelligent assimilation system, input the image after data preprocessing (i.e. the observation data and numerical model meteorological data after data preprocessing in step D.1) into the feature extraction module of the intelligent assimilation system for feature extraction, iterate to the forecast time, and obtain the forecast result;
[0072] like Figure 3 As shown in the figure, the optimal parameters determined by the optimization algorithm are input into the intelligent assimilation system, and the image after data preprocessing is input into the S1 module. After continuous iterative integration, it enters the S2 prediction module. The observed meteorological data after data preprocessing and the integration results of the numerical model of the previous month are successively input into the intelligent assimilation system and the numerical model to obtain the first prediction result, and the subsequent forecast results can be obtained by continuing to iterate.
[0073] D.3. Based on the forecast results, calculate various related indexes and evaluation indicators.
[0074] Figure 4 A schematic diagram of a meteorological numerical model assimilation device based on a neural network provided in an embodiment of the present application includes:
[0075] M1. The first determination module is used to obtain the observation data of the assimilated variables and the initial meteorological data of the numerical model, determine the assimilation period and assimilation time window, and encapsulate the numerical model;
[0076] M2. Model training module, used to convert observation data into the specified input format, perform data preprocessing on the observation data and initial meteorological data of the numerical model for input model, and select intelligent optimization algorithm to train the model;
[0077] M3. Prediction module, used to input the optimal parameters of the trained model into the intelligent assimilation system, and perform iterative forecasting to obtain the prediction results;
[0078] M4. Assimilation model evaluation module, used to convert output data into understandable meteorological grid data and a variety of quantitative evaluation indicators.
[0079] The meteorological numerical model assimilation device based on neural network provided in the embodiment of the present application has the same technical features as the meteorological numerical model assimilation method based on neural network provided in the above embodiment, so it can also solve the same technical problems and achieve the same technical effects.
[0080] like Figure 5 As shown, this embodiment provides a computer device E, including: a processor E.1, a memory E.2 and a bus, the memory E.2 stores machine-readable instructions executable by the processor E.1, when the computer device E is running, the processor E.1 communicates with the memory E.2 through the bus, and the processor E.1 executes the machine-readable instructions to perform the steps of the above-mentioned meteorological numerical model assimilation method based on a neural network.
[0081] Specifically, the above-mentioned memory E.2 and processor E.1 can be general-purpose memories and processors, which are not specifically limited here. When the processor E.1 runs the computer program stored in the memory E.2, it can execute the above-mentioned meteorological numerical model assimilation method based on neural network.
[0082] Corresponding to the above-mentioned meteorological numerical model assimilation method based on neural network, an embodiment of the present application also provides a computer-readable storage medium, which stores machine-executable instructions. When the computer-executable instructions are called and executed by the processor, the computer-executable instructions prompt the processor to execute the steps of the above-mentioned meteorological numerical model assimilation method based on neural network.
[0083] In summary, this technical solution encapsulates numerical models by selecting assimilation variables; constructs an intelligent assimilation system and observation operator based on neural networks; performs data preprocessing on the observation data of the input model and the initial meteorological data of the numerical model; optimizes the neural network parameters based on the intelligent optimization algorithm, determines the intelligent assimilation system, obtains the forecast results through iterative operation; and calculates relevant quantitative evaluation indicators. In this way, a meteorological numerical model assimilation scheme based on neural networks is realized, which can be applied to all numerical models, reduce development costs, avoid the resource consumption of developing accompanying modules of numerical models and designing localized schemes, and can provide a numerical model with an optimal initial field that is dynamically and thermally coordinated, thereby improving the accuracy of existing numerical models in forecasting.
[0084] The present invention aims to develop an assimilation system applicable to any numerical model, eliminating the resource consumption of developing the accompanying modules of the numerical model and designing the localized solution, creatively using the intelligent optimization algorithm to iteratively optimize the neural network parameters, and solving the problem of gradient disappearance in the training process of the deep learning network for the output of the numerical model. In addition, the assimilation system based on the neural network framework has the characteristics of easy construction and emphasis on timeliness, and can be integrated into the existing business assimilation system.
Claims
1. A meteorological numerical model assimilation method based on neural network, characterized in that: The following steps are involved: A. Determine the numerical model that needs to be established for the intelligent assimilation system and encapsulate it, which includes the following sub-steps: A.
1. Select assimilation variables and determine assimilation period; Encapsulating the numerical model so that the numerical model outputs the numerical field of the selected assimilation variable according to the assimilation period; A.
2. Establish a data interaction interface between the numerical model and the intelligent assimilation system: the numerical model receives the numerical field output by the intelligent assimilation system to perform model integration for the next assimilation cycle, and the intelligent assimilation system reads the output of the numerical model as the model input for the next moment; A.
3. Determine the assimilation time window, that is, the time span of the observational data that the numerical model needs to assimilate; B. Constructing an intelligent assimilation system, including the following sub-steps: B.
1. Read the meteorological data of the assimilated variables determined in step A.1 and construct the corresponding feature extraction module; B.
2. Aiming at the uneven grid distribution of the numerical model, construct an observation operator that maps the observation data to the numerical field of the numerical model; C. Training the neural network of the intelligent assimilation system includes the following sub-steps: C.
1. Using the observation operator constructed in step B.2, construct observation image data of the required resolution for neural network training; C.
2. Perform data preprocessing on the observation data and initial meteorological data of the numerical model input into the neural network; C.
3. Select an intelligent optimization algorithm, construct an objective function, and iteratively train the parameters in the feature extraction module; C.
4. The intelligent optimization stops when the number of iterations reaches the set number, and the neural network parameters with the best objective function value are saved; D. Using intelligent assimilation system to assist numerical model forecasting includes the following sub-steps: D.
1. Determine the starting time, the observational data and the initial meteorological data of the numerical model for the starting time of each assimilation cycle, and perform the same data preprocessing as step C.2; D.
2. Input the optimal parameters of the neural network saved in step C.4 into the intelligent assimilation system, input the observation data after data preprocessing in step D.1 and the initial meteorological data of the numerical model into the feature extraction module of the intelligent assimilation system to perform feature extraction, iterate to the set forecast time, and obtain the forecast result; D.
3. Based on the forecast results, calculate the set relevant indexes and evaluation indicators.
2. A meteorological numerical model assimilation method based on neural network according to claim 1, characterized in that: The assimilation variables selected in step A.1 are specifically one or more atmospheric variables, one or more land surface variables and one or more ocean variables.
3. The method for assimilating meteorological numerical models based on neural networks according to claim 1, characterized in that: The forecast results in step D.2 include but are not limited to short-term weather forecast results, medium-term weather forecast results, long-term weather forecast results or climate forecast results.
4. The method for assimilating meteorological numerical models based on neural networks according to claim 1, characterized in that: The step B.1 specifically adopts one or a combination of convolutional neural networks, recurrent convolutional neural networks, long short-term memory convolutional neural networks, and graph convolutional neural networks to construct a feature extraction module.
5. The method for assimilating meteorological numerical models based on neural networks according to claim 1, characterized in that: The step B.2 specifically comprises the following steps: (1) Mapping the acquired observational data to the selected numerical model forecast area, which involves the following operations: slicing, cropping, and splicing; (2) Adjust the resolution of the acquired observational data to be consistent with the selected numerical model forecast field, which involves operations including but not limited to interpolation and downsampling.
6. The method for assimilating meteorological numerical models based on neural networks according to claim 1, characterized in that: The intelligent optimization algorithm in step C.3 is specifically a population biological intelligent heuristic algorithm including a particle swarm algorithm, an ant colony algorithm, and a bee colony algorithm, or a gradient-free optimization algorithm including a Bayesian deep neural network.
7. The method for assimilating meteorological numerical models based on neural networks according to claim 1, characterized in that: The step C.3 specifically adopts one or a combination of image comparison methods to construct the objective function, wherein the image comparison methods include but are not limited to the root mean square error method and the mean absolute error method; The output value of the objective function is specifically: within a determined assimilation time window, the sum of all optimization results output by the numerical model after the intelligent assimilation system and the values calculated by the image comparison function of the observation data.
8. A meteorological numerical model assimilation device based on neural network, characterized in that: It includes a first determination module, a model training module, a prediction module and an assimilation model evaluation module connected in sequence, wherein the first determination module is used to obtain observation data of assimilated variables and initial meteorological data of numerical models, determine the assimilation cycle and assimilation time window, and encapsulate the numerical model; The model training module is used to convert the observation data into a specified input format, perform data preprocessing on the observation data and initial meteorological data of the numerical model input to the model, and select an intelligent optimization algorithm to train the model; The prediction module is used to input the trained model optimal parameters into the intelligent assimilation system and perform iterative prediction to obtain the prediction results; The assimilation model evaluation module is used to convert the output data into set meteorological grid data and a variety of quantitative evaluation indicators.
9. A computer device, comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of the meteorological numerical model assimilation method based on a neural network according to any one of claims 1 to 7 are implemented.
10. A readable storage medium storing instructions executable by a computer, characterized in that: When the computer executable instructions are called and executed by the processor, the computer executable instructions prompt the processor to execute the meteorological numerical model assimilation method based on a neural network according to any one of claims 1 to 7.