Weather forecast data assimilation method, device, equipment, medium and program product
By combining particle filtering and four-dimensional variational data assimilation methods, and utilizing graph neural networks and convolutional long short-term memory networks for data assimilation, the problem of inaccurate data assimilation in existing technologies is solved, achieving high-precision and high-real-time meteorological data assimilation.
Patent Information
- Application Number
- CN202511356228.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-22
- Publication Date
- 2026-01-02
AI Technical Summary
Existing data assimilation methods, such as variational and ensemble Kalman filtering, are only applicable to linear or weakly nonlinear systems, resulting in inaccurate assimilation results that do not conform to reality.
A data assimilation method based on particle filter constraints is adopted, which combines four-dimensional variational and particle filter constraints. By minimizing the difference between the observation field sequence data and the background field sequence data, graph neural network and convolutional long short-term memory network are used for data assimilation. The importance analysis is combined with self-attention mechanism to obtain target assimilation sequence data.
It improves the accuracy and robustness of data assimilation, can handle nonlinear dynamic characteristics, dynamically capture the evolution patterns of meteorological data, realizes the complementary advantages of data, and enhances the accuracy and real-time performance of data assimilation.
Smart Images
Figure CN121256684A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data assimilation technology, specifically to methods, apparatus, equipment, media, and program products for assimilating weather forecast data. Background Technology
[0002] With the rapid development of computer technology, numerical weather prediction models have been continuously improved, and numerical weather prediction models have become an important means of providing future weather forecasts. The accuracy of the initial field obtained from numerical weather prediction models is key to improving the accuracy of numerical weather warnings. Data assimilation techniques can be used to improve the accuracy of the initial field and enhance the forecast results of numerical weather prediction models.
[0003] Data assimilation (DA) is a data fusion technique that combines observational data (actual data), short-term forecast data, and corresponding error information to provide numerical models with the most accurate initial field possible.
[0004] Currently, the most widely used data assimilation method is the variational and ensemble Kalman filter (EnKF). The posterior probability density distribution of the variational and ensemble Kalman filter is Gaussian and depends on the linearization operator. It is only applicable to linear or weakly nonlinear systems, which leads to inaccurate data assimilation results that do not conform to the actual situation. Summary of the Invention
[0005] In view of this, the present invention provides a method, apparatus, device, medium and program product for assimilating weather forecast data, in order to solve the problem that the assimilation results of data assimilation methods in related technologies are not accurate enough and do not conform to the actual situation.
[0006] In a first aspect, the present invention provides a method for assimilating weather forecast data, comprising: constrained by minimizing the difference between first observation field sequence data and background field sequence data, performing preliminary assimilation analysis on the background field sequence data based on the first observation field sequence data to obtain first assimilated sequence data; the first observation field sequence data is meteorological sequence data for a first preset time period obtained from an observation platform, and the background field sequence data is meteorological forecast sequence data obtained from a numerical weather prediction model; fusing the second observation field sequence data and the first assimilated sequence data based on a particle filter constraint to obtain second assimilated sequence data; the particle filter constraint is a constraint for selecting data with a better probability distribution in the time series, and the second observation field sequence data is meteorological sequence data for a second preset time period obtained from an observation platform; performing importance analysis on the first assimilated sequence data and the second assimilated sequence data to obtain a first gate value and a second gate value; the first gate value is used to characterize the importance of the first assimilated sequence data, and the second gate value is used to characterize the importance of the second assimilated sequence data; and fusing the first assimilated sequence data and the second assimilated sequence data using the first gate value and the second gate value to obtain target assimilated sequence data.
[0007] The weather forecast data assimilation method of this invention uses minimizing the difference between the first observation field sequence data and the background field sequence data as a constraint. Based on the first observation field sequence data, a preliminary assimilation analysis is performed on the background field sequence data to obtain the first assimilated sequence data. This first observation field sequence data is then used to correct the background field sequence data obtained from the numerical weather prediction model, reducing the deviation between the background field sequence data and the first observation field sequence data. This ensures that the background field sequence data is as close as possible to the actual first observation field sequence data, avoiding unreasonable data and thus improving the accuracy of the first assimilated sequence data. Based on particle filter constraints, this invention fuses the second observation field sequence data and the first assimilated sequence data to obtain the second assimilated sequence data. Through particle filter constraints, data with better probability distributions are selected in the time series, dynamically capturing the evolution of meteorological data over time and effectively handling the nonlinear dynamic characteristics that may exist in the meteorological system, thereby improving the real-time performance of the second assimilated sequence data. This invention analyzes the importance of first and second assimilation sequence data to obtain a first and a second gating value, enabling differentiated evaluation of different assimilation sequence data. The first assimilation sequence data focuses on the initial stage of observation field and background field fusion, while the second assimilation sequence data focuses on temporal dynamic particle filtering optimization. The importance of the two may differ in different scenarios; the introduction of the gating value quantifies this difference in importance, providing a clear weighting basis for subsequent fusion and avoiding data waste or error amplification caused by simple weighting. This invention utilizes the first and second gating values to fuse the first and second assimilation sequence data to obtain the target assimilation sequence data, achieving complementary advantages between the two data sets, improving the real-time performance of the target assimilation sequence data, and enhancing the accuracy and robustness of data assimilation.
[0008] In one optional implementation, with the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, a preliminary assimilation analysis is performed on the background field sequence data based on the first observation field sequence data to obtain the first assimilation sequence data. This includes: inputting the first observation field sequence data and the background field sequence data into a trained first assimilation model to obtain the first observation field sequence data. The first assimilation model is a data assimilation model constructed based on the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data and a graph neural network. The input of the first assimilation model is the first observation field sequence data and the background field sequence data, and the output of the first assimilation model is the first assimilation sequence data.
[0009] In one optional implementation, based on particle filter constraints, the second observation field sequence data and the first assimilation sequence data are fused to obtain the second assimilation sequence data, including: inputting the second observation field sequence data and the first assimilation sequence data into a trained second assimilation model to obtain the second assimilation sequence data; the second assimilation model is a data assimilation model constructed based on particle filter constraints and a convolutional long short-term memory network, the input of the second assimilation model is the second observation field sequence data and the first assimilation sequence data, and the output of the second assimilation model is the second assimilation sequence data.
[0010] In one optional implementation, importance analysis is performed on the first assimilation sequence data and the second assimilation sequence data to obtain a first gate value and a second gate value. This includes: inputting the first assimilation sequence data and the second assimilation sequence data into a trained importance analysis model to obtain the first gate value and the second gate value; the input to the importance analysis model is the first assimilation sequence data and the second assimilation sequence data, and the output of the importance analysis model is the first gate value and the second gate value. The importance analysis model is a model constructed based on a deep learning network using a self-attention mechanism.
[0011] In one optional implementation, the first assimilation sequence data and the second assimilation sequence data are input into the trained importance analysis model to obtain a first gating value and a second gating value, including: using the importance analysis model to extract features from the first assimilation sequence data and the second assimilation sequence data to obtain a first attention score and a second attention score; converting the first attention score and the second attention score into a first attention weight and a second attention weight based on a preset normalization function of the importance analysis model; and converting the first attention weight and the second attention weight into a first gating value and a second gating value based on a preset nonlinear activation function of the importance analysis model.
[0012] In one optional implementation, the first assimilation sequence data and the second assimilation sequence data are fused using a first gate value and a second gate value to obtain target assimilation sequence data, including: obtaining a first product result based on the product of the first assimilation sequence data and the first gate value; obtaining a second product result based on the product of the second assimilation sequence data and the second gate value; and obtaining the target assimilation sequence data based on the sum of the first product result and the second product result.
[0013] Secondly, the present invention provides a weather forecast data assimilation device, comprising: a first assimilation module, configured to perform preliminary assimilation analysis on the background field sequence data based on the first observation field sequence data, with the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, to obtain first assimilated sequence data; the first observation field sequence data is meteorological sequence data for a first preset time period obtained from an observation platform, and the background field sequence data is meteorological forecast sequence data obtained from a numerical weather prediction model; and a second assimilation module, configured to fuse the second observation field sequence data and the first assimilated sequence data based on particle filter constraints to obtain a second assimilation sequence. The data includes: a particle filter constraint for selecting data with a better probability distribution in a time series; a second observation field sequence data for meteorological sequence data in a second preset time period obtained from an observation platform; a gate value determination module for analyzing the importance of the first and second assimilation sequence data to obtain a first gate value and a second gate value; the first gate value characterizes the importance of the first assimilation sequence data, and the second gate value characterizes the importance of the second assimilation sequence data; and a data fusion module for fusing the first and second assimilation sequence data using the first and second gate values to obtain the target assimilation sequence data.
[0014] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the weather forecast data assimilation method of the first aspect or any corresponding embodiment described above.
[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the weather forecast data assimilation method of the first aspect or any corresponding embodiment thereof.
[0016] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the weather forecast data assimilation method of the first aspect or any corresponding embodiment thereof. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0018] Figure 1This is a flowchart illustrating a weather forecast data assimilation method according to an embodiment of the present invention.
[0019] Figure 2 This is a flowchart illustrating another method for assimilating weather forecast data according to an embodiment of the present invention.
[0020] Figure 3 This is a structural block diagram of a weather forecast data assimilation device according to an embodiment of the present invention.
[0021] Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] With the rapid development of computer technology and the continuous improvement of numerical weather prediction models, numerical weather prediction models have become an important means for major operational forecasting centers to provide future weather forecasts. The accuracy of the initial field is key to improving the accuracy of numerical weather prediction, and data assimilation techniques can be used to improve the accuracy of the initial field and enhance the forecast results of numerical weather prediction models.
[0024] Currently, the most widely used data assimilation methods are variational and ensemble Kalman filtering (EnKF). Although variational and ensemble Kalman filtering methods have achieved good results in business applications, they both have their own limitations. In order to maximize the advantages of these two methods and eliminate their shortcomings, a series of techniques have emerged to couple the two methods, known as hybrid data assimilation methods.
[0025] However, variational, EnKF, and mixed data assimilation methods all explicitly or implicitly involve the problem that the posterior probability density distribution is Gaussian and depends on linearization operators. Therefore, they are only applicable to linear or weakly nonlinear systems. As the resolution of weather forecast models increases and the scale becomes smaller, the requirements for nonlinearity also become more stringent. As a result, variational, EnKF, and mixed data assimilation methods achieve poor assimilation results.
[0026] Particle filtering (PF) is a nonlinear filtering method based on the Monte Carlo approach. It simulates the probability density distribution using a weighted set of particles. Compared to variational and EnKF methods, its greatest advantage is its ability to handle strongly nonlinear problems and simulate non-Gaussian posterior probability density distributions.
[0027] However, particle filtering is prone to filtering degradation, and due to the complex calculations involving a large number of particles, particle filtering algorithms are relatively slow. This is especially true in applications with high real-time requirements, such as real-time meteorological data assimilation, where it may not be able to complete calculations in a timely manner to provide the latest forecast results.
[0028] This invention provides a method for assimilating weather forecast data, which improves the accuracy of weather forecast data assimilation by fusing four-dimensional variational constraints and particle filter constraints.
[0029] According to an embodiment of the present invention, a method for assimilating weather forecast data is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0030] This embodiment provides a method for assimilating weather forecast data, which can be used in computer equipment. Figure 1 This is a flowchart of a weather forecast data assimilation method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:
[0031] Step S101: With the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, a preliminary assimilation analysis is performed on the background field sequence data based on the first observation field sequence data to obtain the first assimilation sequence data; the first observation field sequence data is the meteorological sequence data for the first preset time period obtained from the observation platform, and the background field sequence data is the meteorological forecast sequence data obtained from the numerical weather prediction model.
[0032] The first observation field sequence data can be observation time series data of various elements such as temperature, humidity, wind speed, and wind direction obtained from various observation platforms such as ground stations, radiosondes, radar, and satellites within a first preset time period. Time series data is data arranged in chronological order. The first preset time period can be a historical time period, and the specific time period length can be set according to the actual situation. The background field sequence data can be forecast field time series data of numerical weather prediction models, including forecast time series data of various elements such as temperature, humidity, wind speed, and wind direction.
[0033] In some optional implementations, the first observation field sequence data and the background field sequence data are input into the trained first assimilation model to obtain the first observation field sequence data.
[0034] Step S102: Based on particle filter constraints, the second observation field sequence data and the first assimilation sequence data are fused to obtain the second assimilation sequence data; the particle filter constraint is a constraint to select data with better probability distribution in the time series, and the second observation field sequence data is meteorological sequence data of the second preset time period obtained from the observation platform.
[0035] The second preset time period can be the current real-time time period, and the length of the second preset time period can be set according to the actual situation.
[0036] In some alternative implementations, the second observation field sequence data and the first assimilation sequence data are input into the trained second assimilation model to obtain the second assimilation sequence data.
[0037] Step S103: Perform importance analysis on the first assimilation sequence data and the second assimilation sequence data to obtain a first gate value and a second gate value; the first gate value is used to characterize the importance of the first assimilation sequence data, and the second gate value is used to characterize the importance of the second assimilation sequence data.
[0038] In some alternative implementations, the first assimilation sequence data and the second assimilation sequence data are input into the trained importance analysis model to obtain the first gating value and the second gating value.
[0039] Step S104: Using the first gate value and the second gate value, the first assimilation sequence data and the second assimilation sequence data are fused to obtain the target assimilation sequence data.
[0040] In some optional implementations, the first assimilation sequence data and the second assimilation sequence data are fused using a first gate value and a second gate value to obtain target assimilation sequence data, including: obtaining a first product result based on the product of the first assimilation sequence data and the first gate value; obtaining a second product result based on the product of the second assimilation sequence data and the second gate value; and obtaining the target assimilation sequence data based on the sum of the first product result and the second product result.
[0041] The weather forecast data assimilation method provided in this embodiment takes minimizing the difference between the first observation field sequence data and the background field sequence data as a constraint. Based on the first observation field sequence data, a preliminary assimilation analysis is performed on the background field sequence data to obtain the first assimilated sequence data. The first observation field sequence data is then used to correct the background field sequence data obtained from the numerical weather prediction model, reducing the deviation between the background field sequence data and the first observation field sequence data. This ensures that the background field sequence data is as close as possible to the actual first observation field sequence data, avoiding unreasonable data and thus improving the accuracy of the first assimilated sequence data. This embodiment of the invention, based on particle filter constraints, fuses the second observation field sequence data and the first assimilated sequence data to obtain the second assimilated sequence data. Through particle filter constraints, data with better probability distributions are selected in the time series, dynamically capturing the evolution of meteorological data over time and effectively handling the nonlinear dynamic characteristics that may exist in the meteorological system, thereby improving the real-time performance of the second assimilated sequence data. This invention analyzes the importance of a first assimilation sequence and a second assimilation sequence to obtain a first gate value and a second gate value. This enables differentiated evaluation of different assimilation sequence data. The first assimilation sequence focuses on the initial stage of fusion between the observation field and the background field, while the second assimilation sequence focuses on dynamic particle filtering optimization over time. The importance of these two sequences may differ in different scenarios. The introduction of the gate value quantifies this difference in importance, providing a clear weighting basis for subsequent fusion and avoiding data waste or error amplification caused by simple weighting. This invention utilizes the first and second gate values to fuse the first and second assimilation sequence data to obtain the target assimilation sequence data. This achieves complementary advantages between the first and second assimilation sequence data, improves the real-time performance of the target assimilation sequence data, and enhances the accuracy and robustness of data assimilation.
[0042] This embodiment provides a method for assimilating weather forecast data, which can be used in computer equipment. Figure 2 This is a flowchart of another method for assimilating weather forecast data according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:
[0043] Step S201: With the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, a preliminary assimilation analysis is performed on the background field sequence data based on the first observation field sequence data to obtain the first assimilation sequence data; the first observation field sequence data is the meteorological sequence data for the first preset time period obtained from the observation platform, and the background field sequence data is the meteorological forecast sequence data obtained from the numerical weather prediction model.
[0044] Specifically, step S201 includes:
[0045] Step S2011: Input the first observation field sequence data and the background field sequence data into the trained first assimilation model to obtain the first observation field sequence data. The first assimilation model is a data assimilation model constructed based on the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data and a graph neural network. The input of the first assimilation model is the first observation field sequence data and the background field sequence data, and the output of the first assimilation model is the first assimilation sequence data.
[0046] The first assimilation model is used to assimilate the first observation field sequence data and the background field sequence data by minimizing the difference between them. This ensures that the data assimilation process satisfies the physical laws of the four-dimensional domain (such as atmospheric dynamics equations and energy conservation) and avoids physical inconsistencies such as temperature abrupt changes and wind speed exceeding limits in the corrected data.
[0047] In some alternative implementations, a graph neural network (GNN) is used to address the uneven distribution of the first observation sequence data while preserving spatial correlations.
[0048] In some optional implementations, the weather forecast data assimilation method further includes a process of training a first assimilation model. The process of training the first assimilation model includes: acquiring a first historical dataset, including: first historical observation field sequence data, first historical background field sequence data, and first historical analysis field sequence data; wherein, the first historical analysis field sequence data is obtained through a four-dimensional variational method; using the first historical observation field sequence data and the first historical background field sequence data in the first historical dataset as inputs, and the first historical analysis field sequence data as labels, the GNN is trained using the root mean square error loss function to obtain a first assimilation model based on four-dimensional variational constraints.
[0049] Step S202: Based on particle filter constraints, the second observation field sequence data and the first assimilation sequence data are fused to obtain the second assimilation sequence data; the particle filter constraint is a constraint to select data with better probability distribution in the time series, and the second observation field sequence data is meteorological sequence data of the second preset time period obtained from the observation platform.
[0050] Specifically, step S202 includes:
[0051] Step S2021: Input the second observation field sequence data and the first assimilation sequence data into the trained second assimilation model to obtain the second assimilation sequence data. The second assimilation model is a data assimilation model constructed based on particle filter constraints and convolutional long short-term memory network. The input of the second assimilation model is the second observation field sequence data and the first assimilation sequence data, and the output of the second assimilation model is the second assimilation sequence data.
[0052] Among them, particle filtering constraint uses a large number of particles to simulate probability distribution and select data that is closer to the second observation field sequence data in the time series, dynamically correcting the error of the first assimilation sequence data (especially suitable for capturing nonlinear and short-term changing meteorological features, such as sudden gusts and local humidity fluctuations).
[0053] In some alternative implementations, Convolutional Long Short-Term Memory (Conv-LSTM) networks, combining the spatial feature extraction capabilities of CNN (Convolutional Neural Network) and the temporal dependency modeling capabilities of LSTM (Long Short-Term Memory), can capture both the spatial distribution details of meteorological elements (such as temperature differences between urban and suburban areas) and learn temporal variation patterns (such as the periodic fluctuations of temperature between day and night), so that the resulting second assimilation sequence data takes into account both spatial granularity and temporal continuity.
[0054] In some optional implementations, the weather forecast data assimilation method further includes a process of training a second assimilation model. The training process includes: acquiring a second historical dataset, including: second historical analysis field sequence data, second historical observation field sequence data, and second historical analysis field sequence data obtained through a four-dimensional variational method; wherein the second historical analysis field sequence data is obtained by inputting the second historical analysis field sequence data and the second historical observation field sequence data into a particle filter algorithm; using the second historical analysis field sequence data and the second historical observation field sequence data obtained through the four-dimensional variational method in the second historical dataset as input, and using the second historical analysis field sequence data as labels, optimizing and training the Conv-LSTM using the RMSE (Root Mean Square Error) loss function to obtain a second assimilation model based on particle filter constraints.
[0055] Step S203: Perform importance analysis on the first assimilation sequence data and the second assimilation sequence data to obtain a first gate value and a second gate value; the first gate value is used to characterize the importance of the first assimilation sequence data, and the second gate value is used to characterize the importance of the second assimilation sequence data.
[0056] Specifically, step S203 includes:
[0057] Step S2031: Input the first assimilation sequence data and the second assimilation sequence data into the trained importance analysis model to obtain the first gate value and the second gate value; the input of the importance analysis model is the first assimilation sequence data and the second assimilation sequence data, and the output of the importance analysis model is the first gate value and the second gate value. The importance analysis model is a model constructed based on a deep learning network with a self-attention mechanism.
[0058] The process involves inputting the first and second assimilated sequence data into a trained importance analysis model to obtain a first and second gating value. This includes: extracting features from the first and second assimilated sequence data using the importance analysis model to obtain a first attention score and a second attention score; converting the first and second attention scores into first and second attention weights based on a preset normalization function of the importance analysis model; and converting the first and second attention weights into a first and a second gating value based on a preset nonlinear activation function of the importance analysis model.
[0059] Specifically, the Transformer (a deep learning network with self-attention mechanism) model is used to extract features from the first and second assimilation sequence data to learn the complex patterns and global dependencies of the input data. The first and second attention scores are obtained from the output of the Transformer model. The first and second attention scores are converted into first and second attention weights through the Softmax (normalization) function. The first and second attention weights are converted into first and second gating values through the Sigmoid (non-linear activation) function.
[0060] In some optional implementations, the first attention score and the second attention score are used to represent the credibility of the first assimilation sequence data and the second assimilation sequence data; the sum of the first attention weight and the second attention weight is 1, quantifying the relative importance of the two; the first gate value and the second gate value, for example, the first gate value corresponding to the first assimilation sequence data is 0.3, and the first gate value corresponding to the second assimilation sequence data is 0.7, which means that the second assimilation sequence data is more important.
[0061] Step S204: Using the first gate value and the second gate value, the first assimilation sequence data and the second assimilation sequence data are fused to obtain the target assimilation sequence data. For details, please refer to [link to details]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.
[0062] The weather forecast data assimilation method provided in this embodiment fully leverages the advantages of two different data assimilation techniques: four-dimensional variational and particle filtering. This improves the accuracy and robustness of data assimilation and mitigates the result bias caused by particle degeneration in the particle filtering algorithm. A data-driven model is used to simulate the four-dimensional variational and particle filtering data assimilation models, improving computational efficiency and meeting real-time and high-efficiency requirements. Using GNN to simulate the four-dimensional variational algorithm and Conv-LSTM to simulate the particle filtering algorithm, it can fuse multi-source heterogeneous data and extract temporal and spatial features, fully utilizing the correlation between spatiotemporal features and improving the accuracy of the assimilation results.
[0063] This embodiment also provides a weather forecast data assimilation device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.
[0064] This embodiment provides a weather forecast data assimilation device, such as... Figure 3 As shown, it includes:
[0065] The first assimilation module 301 is used to perform preliminary assimilation analysis on the background field sequence data based on the first observation field sequence data, with the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, to obtain the first assimilation sequence data; the first observation field sequence data is the meteorological sequence data of the first preset time period obtained from the observation platform, and the background field sequence data is the meteorological forecast sequence data obtained from the numerical weather prediction model.
[0066] The second assimilation module 302 is used to fuse the second observation field sequence data and the first assimilation sequence data based on particle filter constraints to obtain the second assimilation sequence data; the particle filter constraint is a constraint to select data with better probability distribution in the time series, and the second observation field sequence data is meteorological sequence data of the second preset time period obtained from the observation platform.
[0067] The gating value determination module 303 is used to perform importance analysis on the first assimilation sequence data and the second assimilation sequence data to obtain a first gating value and a second gating value; the first gating value is used to characterize the importance of the first assimilation sequence data, and the second gating value is used to characterize the importance of the second assimilation sequence data.
[0068] The data fusion module 304 is used to fuse the first assimilation sequence data and the second assimilation sequence data using the first gate value and the second gate value to obtain the target assimilation sequence data.
[0069] In some alternative implementations, the first assimilation module 301 includes:
[0070] The first assimilation unit is used to input the first observation field sequence data and the background field sequence data into the trained first assimilation model to obtain the first observation field sequence data. The first assimilation model is a data assimilation model constructed based on the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data and a graph neural network. The input of the first assimilation model is the first observation field sequence data and the background field sequence data, and the output of the first assimilation model is the first assimilation sequence data.
[0071] In some alternative implementations, the second assimilation module 302 includes:
[0072] The second assimilation unit is used to input the second observation field sequence data and the first assimilation sequence data into the trained second assimilation model to obtain the second assimilation sequence data. The second assimilation model is a data assimilation model constructed based on particle filter constraints and convolutional long short-term memory network. The input of the second assimilation model is the second observation field sequence data and the first assimilation sequence data, and the output of the second assimilation model is the second assimilation sequence data.
[0073] In some alternative implementations, the gate value determination module 303 includes:
[0074] The gating value determination unit is used to input the first assimilation sequence data and the second assimilation sequence data into the trained importance analysis model to obtain the first gating value and the second gating value. The input of the importance analysis model is the first assimilation sequence data and the second assimilation sequence data, and the output of the importance analysis model is the first gating value and the second gating value. The importance analysis model is a model constructed based on a deep learning network with a self-attention mechanism.
[0075] In some optional implementations, the gate value determination unit includes:
[0076] The feature extraction subunit is used to extract features from the first and second assimilation sequence data using an importance analysis model to obtain the first attention score and the second attention score.
[0077] The first transformation subunit is used to transform the first attention score and the second attention score into the first attention weight and the second attention weight based on the preset normalization function of the importance analysis model.
[0078] The second transformation subunit is used to transform the first attention weight and the second attention weight into the first gating value and the second gating value based on the preset nonlinear activation function of the importance analysis model.
[0079] In some alternative implementations, the data fusion module 304 includes:
[0080] The first multiplication unit is used to obtain the first product result based on the product of the first assimilation sequence data and the first gate value.
[0081] The second multiplication unit is used to obtain the second product result based on the product of the second assimilation sequence data and the second gate value.
[0082] The data fusion unit is used to obtain the target assimilation sequence data based on the sum of the first product result and the second product result.
[0083] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.
[0084] In this embodiment, the weather forecast data assimilation device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.
[0085] This invention also provides a computer device having the above-described features. Figure 3 The device shown is for assimilating weather forecast data.
[0086] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 4 Take a processor 10 as an example.
[0087] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.
[0088] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.
[0089] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0090] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.
[0091] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.
[0092] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.
[0093] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.
[0094] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A method for assimilating weather forecast data, characterized in that, The method includes: With the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, a preliminary assimilation analysis is performed on the background field sequence data based on the first observation field sequence data to obtain the first assimilated sequence data; the first observation field sequence data is the meteorological sequence data for a first preset time period obtained from the observation platform, and the background field sequence data is the meteorological forecast sequence data obtained from the numerical weather prediction model; Based on particle filter constraints, the second observation field sequence data and the first assimilation sequence data are fused to obtain the second assimilation sequence data; the particle filter constraints are constraints for selecting data with better probability distribution in the time series, and the second observation field sequence data are meteorological sequence data for a second preset time period obtained from the observation platform; Importance analysis is performed on the first assimilation sequence data and the second assimilation sequence data to obtain a first gate value and a second gate value; the first gate value is used to characterize the importance of the first assimilation sequence data, and the second gate value is used to characterize the importance of the second assimilation sequence data; Using the first gate value and the second gate value, the first assimilation sequence data and the second assimilation sequence data are fused to obtain the target assimilation sequence data.
2. The method according to claim 1, characterized in that, The process involves minimizing the difference between the first observation field sequence data and the background field sequence data. Based on the first observation field sequence data, a preliminary assimilation analysis is performed on the background field sequence data to obtain the first assimilated sequence data, including: The first observation field sequence data and the background field sequence data are input into the trained first assimilation model to obtain the first observation field sequence data. The first assimilation model is a data assimilation model constructed based on the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data and a graph neural network. The input of the first assimilation model is the first observation field sequence data and the background field sequence data, and the output of the first assimilation model is the first assimilation sequence data.
3. The method according to claim 1 or 2, characterized in that, The process of fusing the second observation field sequence data and the first assimilation sequence data based on particle filter constraints to obtain the second assimilation sequence data includes: The second observation field sequence data and the first assimilation sequence data are input into the trained second assimilation model to obtain the second assimilation sequence data. The second assimilation model is a data assimilation model constructed based on particle filter constraints and convolutional long short-term memory network. The input of the second assimilation model is the second observation field sequence data and the first assimilation sequence data, and the output of the second assimilation model is the second assimilation sequence data.
4. The method according to claim 1 or 2, characterized in that, The step of performing importance analysis on the first assimilation sequence data and the second assimilation sequence data to obtain a first gate value and a second gate value includes: The first assimilation sequence data and the second assimilation sequence data are input into the trained importance analysis model to obtain the first gate value and the second gate value. The input of the importance analysis model is the first assimilation sequence data and the second assimilation sequence data, and the output of the importance analysis model is the first gate value and the second gate value. The importance analysis model is a model constructed based on a deep learning network with a self-attention mechanism.
5. The method according to claim 4, characterized in that, The step of inputting the first assimilated sequence data and the second assimilated sequence data into the trained importance analysis model to obtain the first gate value and the second gate value includes: The importance analysis model is used to extract features from the first assimilation sequence data and the second assimilation sequence data to obtain the first attention score and the second attention score; Based on the preset normalization function of the importance analysis model, the first attention score and the second attention score are converted into a first attention weight and a second attention weight. Based on the preset nonlinear activation function of the importance analysis model, the first attention weight and the second attention weight are transformed into the first gating value and the second gating value.
6. The method according to claim 1 or 2, characterized in that, The step of fusing the first assimilation sequence data and the second assimilation sequence data using the first gate value and the second gate value to obtain the target assimilation sequence data includes: The first product result is obtained by multiplying the first assimilation sequence data and the first gate value; The second product result is obtained by multiplying the second assimilation sequence data and the second gate value; The target assimilation sequence data is obtained by summing the first product result and the second product result.
7. A weather forecast data assimilation device, characterized in that, The device includes: The first assimilation module is used to perform preliminary assimilation analysis on the background field sequence data based on the first observation field sequence data, with the constraint of minimizing the difference between the first observation field sequence data and the background field sequence data, to obtain the first assimilation sequence data; the first observation field sequence data is meteorological sequence data for a first preset time period obtained from the observation platform, and the background field sequence data is meteorological forecast sequence data obtained from the numerical weather prediction model; The second assimilation module is used to fuse the second observation field sequence data and the first assimilation sequence data based on particle filter constraints to obtain the second assimilation sequence data; the particle filter constraint is a constraint to select data with better probability distribution in the time series, and the second observation field sequence data is meteorological sequence data of a second preset time period obtained from the observation platform; The gate value determination module is used to perform importance analysis on the first assimilation sequence data and the second assimilation sequence data to obtain a first gate value and a second gate value; the first gate value is used to characterize the importance of the first assimilation sequence data, and the second gate value is used to characterize the importance of the second assimilation sequence data; The data fusion module is used to fuse the first assimilation sequence data and the second assimilation sequence data using the first gate value and the second gate value to obtain the target assimilation sequence data.
8. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the weather forecast data assimilation method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the assimilation method of weather forecast data according to any one of claims 1 to 6.
10. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the assimilation method of weather forecast data according to any one of claims 1 to 6.